{"co-sus-backend-matrix.b2ad098fb3":"Cost saved","co-sus-backend-matrix.915dfba90a":"EUR 69,320","co-sus-backend-matrix.99335c026a":"per server, 5 yr","co-sus-backend-matrix.c36f191943":"EUR 72,210 GPU energy against EUR 2,890 Dweve, at the Dutch average rate.","co-sus-backend-matrix.e407d370bf":"Grid freed","co-sus-backend-matrix.ef3f36ce8a":"231,065 kWh","co-sus-backend-matrix.97c43c9af5":"Capacity returned to the grid that a homes-and-hospitals queue can use instead.","co-sus-backend-matrix.ff9fa89178":"CO2 avoided","co-sus-backend-matrix.b8f58b723a":"90,809 kg","co-sus-backend-matrix.f9d3d807fa":"Emissions never produced, at 393g CO2e per kWh on the Dutch grid mix.","co-sus-backend-matrix.4d46e2ec70":"Water drawn","co-sus-backend-matrix.afc6c45769":"Zero","co-sus-backend-matrix.7022f47fc1":"air cooled","co-sus-backend-matrix.122650a943":"No liquid cooling loop, so no operational water footprint for thermal management.","co-sus-backend-matrix.a767bb0a51":"On-premise","co-sus-backend-matrix.236c31b762":"Standard x86 or ARM servers, inside your own data centre, no GPU fleet.","co-sus-backend-matrix.14be826220":"Sovereign cloud","co-sus-backend-matrix.e5937f8f78":"EU-pinned region, same capability, no hot-path crossing a border you did not approve.","co-sus-backend-matrix.5fdc5540b0":"Standard outlet","co-sus-backend-matrix.7f4ee0b2e8":"Low enough draw to run on an ordinary power connection, even a solar installation.","co-sus-backend-matrix.5acab7406e":"impact.matrix / 5yr-horizon","co-sus-backend-matrix.865dfb9622":"per server, switch one","co-sus-backend-matrix.74d24f6960":"Dutch grid average","co-sus-backend-matrix.8a24a36fc6":"Where it earns these figures","co-sus-backend-matrix.ea89037a5b":"The model scales linearly. If the reported 96% energy comparison transfers, fewer equivalent data centres would be needed, so\n          the land, concrete, steel, and skilled workers that a hyperscale campus consumes could return to\n          the housing and infrastructure Europe actually needs.","co-sus-benefits-deck.29c1ffaff3":"No special infrastructure","co-sus-benefits-deck.9bef75b932":"In the stated low-draw comparison, a standard office outlet could be enough. Confirm the site connection and workload before treating that as a deployment promise.","co-sus-benefits-deck.752c194a85":"Wall outlet","co-sus-benefits-deck.f524532da1":"No data centre","co-sus-benefits-deck.3a73246f15":"Solar-powered AI","co-sus-benefits-deck.7eb0218621":"A small solar installation could power serious AI workloads in the stated low-draw scenario. Battery backup becomes more attainable. This is solar-powered AI as a testable deployment goal, not a blanket carbon-neutral claim.","co-sus-benefits-deck.8689c27e40":"Small solar","co-sus-benefits-deck.4b5e0c17e5":"Battery backup","co-sus-benefits-deck.123a99127a":"Deploy anywhere","co-sus-benefits-deck.eca29fe299":"In your office building. On a hospital floor. In a rural clinic. On a ship. If humans can work there, this AI can run there. No data centre required.","co-sus-benefits-deck.f587c404b1":"Office or clinic","co-sus-benefits-deck.ccecc84377":"On a ship","co-sus-benefits-deck.ffbed03960":"Democratised access","co-sus-benefits-deck.bedc304016":"SMEs, startups, research labs, non-profits. If the reported comparison transfers, lower operating energy could make advanced AI attainable beyond tech-giant budgets.","co-sus-benefits-deck.548366ab43":"Small teams","co-sus-benefits-deck.9585e2d8fc":"Research labs","co-sus-benefits-deck.5ef14ad617":"what.changes / for-you","co-sus-benefits-deck.50eb9c9ffa":"green AI, in practice","co-sus-benefits-deck.ecd812d4a8":"4 benefits","co-sus-benefits-deck.b4151c17f9":"If you have power for computers, you have power for this.","co-sus-benefits-deck.62004c8b3d":"Reported lower operating energy, advanced AI for more organisations","co-sus-benefits-deck.dd95079a81":"Plug it in","co-sus-benefits-deck.0a82742c4e":"Run on sunlight","co-sus-benefits-deck.c2b0d66607":"Anywhere people work","co-sus-benefits-deck.7ccd8cfc31":"For everyone","co-sus-business-glyph.cd1d97db3d":"One year, one server","co-sus-business-glyph.8b81ad6fbf":"The same workload, both ways","co-sus-business-glyph.5902e88c6f":"96% less","co-sus-business-glyph.afcb7d5403":"Electricity for the same work","co-sus-business-glyph.d5a7cbd732":"FP16 GPU baseline","co-sus-business-glyph.f6fd158830":"Dweve server","co-sus-business-glyph.09afd522a1":"headroom left on the grid","co-sus-business-glyph.c11fd2bbcd":"Where the cost lands","co-sus-business-glyph.5569ac434d":"Cooling","co-sus-business-glyph.50c2dd7b6e":"Air cooled, no water footprint.","co-sus-business-glyph.b76ab65957":"Hardware","co-sus-business-glyph.af525627fe":"Standard servers, longer useful life.","co-sus-business-glyph.701c483f81":"Grid","co-sus-business-glyph.bc78471db9":"15,014 companies waiting for capacity.","co-sus-business-glyph.dd67e322a8":"Netbeheer Nederland, end of 2025","co-sus-business-glyph.1527cf4ae8":"Emissions never produced","co-sus-business-glyph.2ed0487c34":"Built and run in Europe","co-sus-business-glyph.f92bbb8348":"Dweve-reported comparison","co-sus-business-glyph.2f9d578c58":"393 g CO2e/kWh","co-sus-business-glyph.f83e928509":"1,926 kWh","co-sus-business-glyph.e612abafb2":"48,139 kWh","co-sus-calm-answers.2b4bd51a0f":"Do I have to do anything?","co-sus-calm-answers.c1f5e1a5fa":"No. You will never touch this. It works quietly behind the shops, clinics, and offices you already use. The only difference is they use less power to do it.","co-sus-calm-answers.0b049b5346":"Nothing to do","co-sus-calm-answers.be184774e5":"Does it really use less electricity?","co-sus-calm-answers.eee3af6324":"Yes. It is built to run on a normal computer plugged into a normal wall socket, instead of a shed full of power-hungry machines that need water to stay cool.","co-sus-calm-answers.de918cf835":"Far less","co-sus-calm-answers.6e4a5297ea":"Where does it run?","co-sus-calm-answers.0b94d7271e":"In Europe, often in the same building that uses it. Because it sips power, it does not need a giant data centre far away. Your information can stay close to home.","co-sus-calm-answers.107c92ba67":"In Europe","co-sus-calm-answers.fd1666942b":"Is it better for the planet?","co-sus-calm-answers.b02dbedc39":"Yes. Less electricity burned means less pollution. It also frees up land, water, and the grid for homes and hospitals instead of more server sheds.","co-sus-calm-answers.d963f32724":"Kinder","co-sus-calm-answers.07265d0af1":"The honest answers","co-sus-calm-answers.e76ee762ad":"no catch","co-sus-calm-answers.378bb9af36":"That is the whole story. Useful work, far less power, kept close to home, and kinder to the\n          world your grandchildren will live in.","co-sus-city-grid.fb61c8a8eb":"Netherlands","co-sus-city-grid.66ad4a8160":"15,014 consumption requests (end 2025)","co-sus-city-grid.24eb7a78ec":"Netbeheer Nederland reports 15,014 companies in the queue; connection timing varies by area and connection type.","co-sus-city-grid.2dce283e0e":"Amsterdam","co-sus-city-grid.78e1f98248":"Complete ban on new data centres","co-sus-city-grid.8ba619178a":"The grid simply cannot handle more demand, so the city stopped approving new data centres altogether.","co-sus-city-grid.d5d8846569":"Frankfurt","co-sus-city-grid.d537045c92":"Connection queue past 2030","co-sus-city-grid.1f366d0f73":"Power connection queues stretch beyond 2030. Critical infrastructure cannot expand on a grid this constrained.","co-sus-city-grid.c4a265ca6d":"Dublin","co-sus-city-grid.037963e5d4":"80% renewable-electricity assumption","co-sus-city-grid.3498f62202":"EirGrid uses 80% as a renewable-generation assumption for future connections; it does not show Dublin's AI share. The greater Dublin area is described as fully constrained.","co-sus-city-grid.a59ba24d63":"NL","co-sus-city-grid.95e84083aa":"grid.monitor / europe","co-sus-city-grid.55f83a436c":"4 cities, reported constraints","co-sus-city-grid.c21ba7f72d":"not a forecast","co-sus-city-grid.5a4c86d5de":"The cost is real. Grid congestion costs the Dutch economy","co-sus-city-grid.28825ab053":"EUR 40 billion","co-sus-city-grid.05c9d3c90b":"annually,\n            and upgrading Europe's grid to handle current AI growth will require","co-sus-city-grid.bd3be4756e":"more than EUR 200 billion","co-sus-city-grid.a90faafb3d":"by 2040.\n            Connection timing varies, and this page does not establish that AI data centres\n            get priority access to shrinking grid capacity.","co-sus-city-grid.status-queue-timing":"Timing varies by area","co-sus-city-grid.status-moratorium":"Moratorium reported","co-sus-city-grid.status-queue-beyond":"Queue reported beyond 2030","co-sus-city-grid.status-grid-load":"Data-centre load","co-sus-compliance-desk.6a99300a46":"EU AI Act scope to review","co-sus-compliance-desk.8c4cc42b98":"ready for review","co-sus-compliance-desk.89c4ed7221":"procurement / compliance sign-off","co-sus-compliance-desk.039037cc6b":"Report-ready","co-sus-compliance-desk.722e9c37c3":"Audit posture","co-sus-compliance-desk.24dea8ab84":"Reported 96% comparison","co-sus-compliance-desk.1d243cbe98":"Grid load","co-sus-compliance-desk.9531949785":"EUR 69,320 per server","co-sus-compliance-desk.805202f5eb":"Five-year saving","co-sus-compliance-desk.2a4888f6b9":"EU jurisdiction end to end, with workload placement held in region. Choose managed Fabric on the public Mesh, licensed customer-controlled infrastructure, or physically isolated operation.","co-sus-compliance-desk.ffeea86281":"Data residency and jurisdiction","co-sus-compliance-desk.b6c464f702":"The stated low-draw scenario is intended to avoid special infrastructure; site connection and workload checks still apply. A standard office outlet may be enough for the described setup.","co-sus-compliance-desk.7493abd0b2":"Power and grid connection","co-sus-compliance-desk.303c28c0ff":"A June 2026 German government proposal moves the 100% renewable target from 2027 to 2030; verify the enacted scope. A lower-draw comparison may reduce the energy burden, but does not replace compliance review.","co-sus-compliance-desk.772ac58fb5":"German renewable-energy thresholds","co-sus-compliance-desk.b1c566c782":"In an on-premise deployment on a standard power connection, data can stay in the building. Residency and regulatory outcomes still depend on configuration and scope.","co-sus-compliance-desk.d911b430b9":"Where data is processed","co-sus-compliance-desk.0762d8f49a":"The reported 96% comparison is described for a stated workload; independent verification and raw output are not published on this page.","co-sus-compliance-desk.f77484afd7":"EU AI Act GPAI energy documentation","co-sus-compliance-desk.status-reported":"Dweve-reported","co-sus-compliance-desk.status-on-premise":"On-premise","co-sus-compliance-desk.status-assumptions":"Assumptions apply","co-sus-compliance-desk.status-connection":"Connection varies","co-sus-compliance-desk.status-eu-held":"EU-held","co-sus-compute-mechanism.3f633c3c8c":"Dweve substrate","co-sus-compute-mechanism.dd5ca315b8":"GPU stack","co-sus-compute-mechanism.160bbfbc4d":"Axis","co-sus-compute-mechanism.22e055e63d":"Reported 96% cut at source","co-sus-compute-mechanism.8c006f2296":"mechanism resolved","co-sus-compute-mechanism.5fed2e95be":"compute.model / dense vs binary","co-sus-compute-mechanism.ec2f085fd7":"Lower board-level draw means standard air cooling is enough. No liquid loop, no operational water footprint, and one fewer class of mechanical failure point.","co-sus-compute-mechanism.f975c17000":"Standard air, no water","co-sus-compute-mechanism.3a8192cbd7":"Industrial liquid cooling, water draw","co-sus-compute-mechanism.fc36c7df6a":"Thermals","co-sus-compute-mechanism.ab9259524c":"The page's dense-accelerator comparison uses a continuous figure near 700W per accelerator. The same stated work maps to SIMD lanes on standard CPUs or FPGAs, so there is no proprietary accelerator to power, cool, or replace on a 1 to 3 year treadmill.","co-sus-compute-mechanism.ba3f28beb5":"Commodity x86, ARM, RISC-V, or FPGA with SIMD","co-sus-compute-mechanism.0d00cdabb8":"Specialised GPUs at ~700W each, continuous","co-sus-compute-mechanism.ba8ac97e12":"Silicon","co-sus-compute-mechanism.90725da6d0":"Fixed-point integer arithmetic is cheaper per operation than FP16 and carries zero numerical drift. It needs no floating-point unit array, which is one of the largest power sinks on a GPU die.","co-sus-compute-mechanism.358989911d":"Correctly-rounded fixed-point integers","co-sus-compute-mechanism.79495159a5":"FP16 floating point, hardware FPUs","co-sus-compute-mechanism.45f053dc73":"Arithmetic","co-sus-compute-mechanism.da448fbd46":"A dense network activates every parameter for every token. The 528-domain-specialist mixture routes to only 4 to 8 domain specialists per query, so the vast majority of weights stay cold and draw nothing.","co-sus-compute-mechanism.73d94fe39a":"Sparse: 4 to 8 domain specialists active per query","co-sus-compute-mechanism.66f413d82f":"Every parameter fires per decision","co-sus-compute-mechanism.b9d3621863":"Activation","co-sus-compute-mechanism.8c55e7ecb3":"A GPU inference is billions of floating-point multiply-accumulates. The described substrate decides in binary constraints, so the dense matrix multiply attributed to the power budget does not run on the stated critical path.","co-sus-compute-mechanism.9fcf00586d":"Binary yes / no constraint decisions","co-sus-compute-mechanism.ac7d5506ba":"Billions of FP16 multiply-accumulates","co-sus-compute-mechanism.69fa1aa564":"Unit of work","co-sus-consumer-glyph.99c2e90a08":"The same answer, far less power","co-sus-consumer-glyph.23aa1bf0c5":"What that means at home","co-sus-consumer-glyph.e91519807e":"Power for one job","co-sus-consumer-glyph.192a56759d":"used","co-sus-consumer-glyph.56b5d6e690":"left for everyone else","co-sus-consumer-glyph.d65d552b71":"Where one way needs twenty five, the reported figure is one.","co-sus-consumer-glyph.553e2efc51":"Cooled by air","co-sus-consumer-glyph.8fa6a6386a":"A fan, not a water loop.","co-sus-consumer-glyph.a4471bd7bb":"Ordinary computers","co-sus-consumer-glyph.66d989418a":"Standard machines, kept for longer.","co-sus-consumer-glyph.ca007cf01d":"More left over","co-sus-consumer-glyph.6ca8788565":"Power stays free for homes and shops.","co-sus-consumer-glyph.e92bbdeefd":"Less power for the same answer","co-sus-consumer-glyph.6af74613cf":"Built in Europe","co-sus-energy-ledger.cb281ee6a6":"Power draw","co-sus-energy-ledger.3c60eb4968":"9,150 W","co-sus-energy-ledger.ea760bfbec":"366 W","co-sus-energy-ledger.a2cd878c0b":"Reported board-level comparison. The GPU figure is one published dense-accelerator server with PUE 1.5x overhead; Dweve is standard CPU or FPGA silicon.","co-sus-energy-ledger.15c3142b9e":"Annual energy","co-sus-energy-ledger.e612abafb2":"48,139 kWh","co-sus-energy-ledger.f83e928509":"1,926 kWh","co-sus-energy-ledger.3d4db522ae":"One server, 60% utilisation, full year. The reported 25x ratio between the two is the comparison in one number.","co-sus-energy-ledger.002e89ed7d":"Annual cost, NL","co-sus-energy-ledger.92890c62bc":"EUR 14,442","co-sus-energy-ledger.ba910da5c8":"EUR 578","co-sus-energy-ledger.ecd31b8ee6":"At the EUR 0.30/kWh Dutch average. Over five years this is EUR 72,210 against EUR 2,890, a EUR 69,320 saving per server.","co-sus-energy-ledger.3eb302108f":"CO2 per year","co-sus-energy-ledger.320bd833ac":"18,919 kg","co-sus-energy-ledger.0416bee0b0":"757 kg","co-sus-energy-ledger.f684e4efc1":"At 393g CO2e per kWh. Over five years, 90,809 kg of emissions avoided per server.","co-sus-energy-ledger.6a2a19d915":"Cooling model","co-sus-energy-ledger.23a56dd665":"Liquid plus water","co-sus-energy-ledger.351723df5c":"Standard air","co-sus-energy-ledger.6603d09732":"No liquid loop, no water draw. Air cooling removes a class of thermal and mechanical failure points.","co-sus-energy-ledger.f4ac22f591":"energy.ledger / per-server","co-sus-energy-ledger.83b7213915":"Published GPU baseline vs Dweve, Dweve-reported comparison","co-sus-energy-ledger.0fe470218e":"Up to 96% cut, 25x reported ratio","co-sus-energy-ledger.b2bb7604c8":"Metric","co-sus-energy-ledger.2dae771c68":"GPU baseline","co-sus-energy-ledger.1bbe39f6af":"Dweve","co-sus-energy-ledger.799b496b24":"Reduction","co-sus-fleet-scale.a19ee5a9fd":"Largest reported fleet","co-sus-fleet-scale.f90eaab8ea":"1,000,000 GPUs","co-sus-fleet-scale.c2c5e67ff9":"These 125,000 servers draw about 1.14 gigawatts of continuous power, enough to power 3.8 million Dutch households. The Netherlands has 7.9 million households in total, so one operator alone uses nearly half a country.","co-sus-fleet-scale.8d546a6dea":"Second reported fleet","co-sus-fleet-scale.2834944e8c":"600,000 GPUs","co-sus-fleet-scale.c7341ca97e":"A second hyperscale fleet in the stated scenario uses dense FP16 servers, each modelled at 9,150W board-level with datacenter overhead at PUE 1.5x.","co-sus-fleet-scale.b780a23b53":"Third reported fleet","co-sus-fleet-scale.c0b31c8556":"100,000 GPUs","co-sus-fleet-scale.23085e235c":"Even the smaller fleet shown is 12,500 servers. Multiply one server's 48,139 kWh a year across a fleet and the grid impact is measured in cities, not racks.","co-sus-fleet-scale.37b0929119":"Per-server draw","co-sus-fleet-scale.6023fd0e6b":"Stated 9,150W into reported 366W","co-sus-fleet-scale.174917e560":"Datacenters needed","co-sus-fleet-scale.67b19151ff":"Up to 96% fewer in the model","co-sus-fleet-scale.1d243cbe98":"Grid load","co-sus-fleet-scale.d71c2beab6":"Reported 25x ratio","co-sus-fleet-scale.f16e7cb653":"fleet.scale / oct-2025","co-sus-fleet-scale.6a4485a41a":"public deployment counts","co-sus-fleet-scale.06af08af62":"dense FP16 fleets","co-sus-fleet-scale.f4f800c096":"At this scale","co-sus-fleet-scale.3b3797dfcb":"The lever, fleet wide","co-sus-fleet-scale.589d117415":"The model scales linearly: if the reported per-server comparison transfers, fewer equivalent data centres would be needed at fleet scale.","co-sus-fork-paths.efc9bb5780":"hitting a wall","co-sus-fork-paths.2b3e2e84a8":"smart growth","co-sus-fork-paths.be32a6816b":"Here is the good news. Efficiency and capability are not opposites. If the reported 25x ratio holds for an equivalent workload,\n          the same infrastructure has more room for AI. We are not choosing between\n          progress and sustainability. We are choosing between","co-sus-fork-paths.54c7586c83":"Imagine if we build smarter from today forward.","co-sus-fork-paths.183e5d692a":"The efficient path","co-sus-fork-paths.82a8a258f1":"Picture Europe in 2030 under the stated 9,150W-per-server GPU-baseline scenario.","co-sus-fork-paths.ae6a4617d8":"If we keep the current path","co-sus-fork-paths.45318a0f51":"the choice is ours","co-sus-fork-paths.f914cc5af1":"two futures, one choice","co-sus-fork-paths.a5454d70e6":"crossroads / europe-2030","co-sus-fork-paths.4e7d490905":"We meet EU climate goals while advancing technology, not by choosing between them.","co-sus-fork-paths.67b4dc1a37":"Every European business can afford advanced AI, not just tech giants with data centre budgets.","co-sus-fork-paths.3cfd1e09e5":"Longer hardware lifecycles mean less waste and lower replacement costs.","co-sus-fork-paths.711a61e364":"Businesses, schools, and homes get power when they need it, not after years of waiting.","co-sus-fork-paths.7fa400a869":"No more data centre moratoriums because efficient AI fits within existing infrastructure.","co-sus-fork-paths.af7795bd74":"Scenario: AI grows 100x more capable while total grid capacity stays at today's level.","co-sus-fork-paths.49af65687c":"We build new power plants for AI while families wait years for electricity connections.","co-sus-fork-paths.41006d4a32":"European AI becomes a luxury only massive corporations with data centre access can afford.","co-sus-fork-paths.e786be7205":"Scenario: mountains of electronic waste from GPU servers replaced every 1 to 3 years.","co-sus-fork-paths.be91586645":"Grid congestion costs balloon from EUR 40 billion to more than EUR 100 billion annually.","co-sus-fork-paths.c80f4f83b4":"Scenario: the business power-connection queue grows from the cited 12,000 baseline to hundreds of thousands.","co-sus-fork-paths.0c7c04ad4b":"More cities join Amsterdam, Frankfurt, and Dublin in banning new data centres.","co-sus-home-wins.c58b570eb5":"Your home keeps its power","co-sus-home-wins.d41d9d1bc0":"When AI sheds need less of the grid, there is more left for the kettle, the heating, and the lights in your street.","co-sus-home-wins.b911e5239a":"Less pressure on bills","co-sus-home-wins.6988f03345":"Power that is not gobbled up by giant machines is power that does not have to be fought over. Scarcity is what pushes prices up.","co-sus-home-wins.f397789473":"Hospitals and schools get connected","co-sus-home-wins.9a0898b4e8":"Today, new homes and public buildings wait years for a power connection while data centres jump the queue. Using less changes who waits.","co-sus-home-wins.434faa4744":"Cleaner air for the grandchildren","co-sus-home-wins.ae58e67bc9":"Less electricity burned means less pollution put into the sky. The same helpful work, with a much lighter footprint.","co-sus-home-wins.f207bca5d3":"It can run right where you are","co-sus-home-wins.713aefd696":"Because it is gentle on power, this kind of AI can sit in a local office or a clinic, not a far-away shed. Your information can stay close to home.","co-sus-home-wins.99a9fcc7cd":"What it means for your day","co-sus-home-wins.734bff7dfb":"Five everyday changes, in order of how close to home they land","co-sus-home-wins.caba223ea6":"None of this asks anything of you. It just means the people building AI are being careful with the power we all share.","co-sus-home-wins.31dedab8bb":"In your home","co-sus-home-wins.bebfe7abe6":"On your bill","co-sus-home-wins.63b3a05e5f":"In your town","co-sus-home-wins.11350b5c99":"In the air","co-sus-home-wins.1a5891b730":"Close to you","co-sus-lifecycle-timeline.9e94f962e3":"do not fully recycle their IT assets,\n          and millions of tons of AI data centre e-waste are projected. Standard server components, by\n          contrast, have established recycling and refurbishment channels.","co-sus-lifecycle-timeline.579058daee":"26% of companies","co-sus-lifecycle-timeline.cf7ec71927":"dense accelerator chassis is\n          discarded as models advance,","co-sus-lifecycle-timeline.7c4220351b":"130 kg","co-sus-lifecycle-timeline.9a1eb5d30c":"The scale of the e-waste problem is real. Each","co-sus-lifecycle-timeline.b468700334":"Total hardware waste","co-sus-lifecycle-timeline.879e32326c":"Year","co-sus-lifecycle-timeline.8839269e24":"GPU vs Dweve","co-sus-lifecycle-timeline.2336f30cb2":"replacement events plotted","co-sus-lifecycle-timeline.3cf4c10a3d":"lifecycle / 10-year","co-sus-lifecycle-timeline.c4bbef1fc1":"Standard CPU or FPGA servers without specialised GPUs. Longer useful life because software improvements do not require new hardware, and components can be upgraded individually.","co-sus-lifecycle-timeline.4ac4a2effd":"15 to 50 kg","co-sus-lifecycle-timeline.98f31c4dfd":"1 to 2 units","co-sus-lifecycle-timeline.3d44269321":"Servers over 10 years","co-sus-lifecycle-timeline.b312039ad4":"~15 to 25 kg","co-sus-lifecycle-timeline.b7debaafad":"Weight per server","co-sus-lifecycle-timeline.2d782d12e6":"Standard lifecycle","co-sus-lifecycle-timeline.2d0b34dff2":"Replacement cycle","co-sus-lifecycle-timeline.f0366b27bf":"Dweve systems","co-sus-lifecycle-timeline.ed288d9d4d":"Rare earth minerals, specialised circuits, and cooling systems that become obsolete as AI advances. Most components cannot be recycled effectively.","co-sus-lifecycle-timeline.5ff85abb7b":"390 to 1,300 kg","co-sus-lifecycle-timeline.57b0a92a61":"3 to 10 units","co-sus-lifecycle-timeline.f3538a8b28":"1 to 3 years","co-sus-lifecycle-timeline.723fb1cfa4":"Traditional GPU servers","co-sus-lifecycle-timeline.3b6faa5c88":"builds","co-sus-plain-cost.fe433fb0d3":"per server, at the Dutch average electricity rate. That is money a school, a clinic or\n          a small business keeps instead of spending it on power.","co-sus-plain-cost.915dfba90a":"EUR 69,320","co-sus-plain-cost.db731ee02b":"The difference, over five years","co-sus-plain-cost.8de92da8e0":"one ordinary computer","co-sus-plain-cost.52c3b618d3":"The gentle kind","co-sus-plain-cost.bacd4ad58b":"a shed full of machines","co-sus-plain-cost.4b15f4c938":"The hungry kind","co-sus-plain-cost.f1453cfe3b":"one server, one year","co-sus-plain-cost.0db75496af":"Two bills for the same year","co-sus-plain-cost.d987ef4821":"An ordinary socket","co-sus-plain-cost.0a6ff27a77":"A dedicated supply","co-sus-plain-cost.bf5b142c12":"Where it plugs in","co-sus-plain-cost.351723df5c":"Standard air","co-sus-plain-cost.23a56dd665":"Liquid plus water","co-sus-plain-cost.5569ac434d":"Cooling","co-sus-plain-cost.f83e928509":"1,926 kWh","co-sus-plain-cost.e612abafb2":"48,139 kWh","co-sus-plain-cost.cff235cd44":"Electricity used","co-sus-plain-cost.ba910da5c8":"EUR 578","co-sus-plain-cost.92890c62bc":"EUR 14,442","co-sus-plain-cost.280f0ac00c":"Electricity for one year","co-sus-plain-power.4abf401340":"The short version. We build the gentle kind, so the lights stay on for everyone, including you.","co-sus-plain-power.7312bf1992":"same job, less power","co-sus-plain-power.aa2500d078":"two kinds of AI","co-sus-plain-power.704869f20f":"In plain words","co-sus-plain-power.ecf4b31705":"When the machines need less power, there is more left over for the people and places that need it.","co-sus-plain-power.e50c742bb5":"It frees up the grid","co-sus-plain-power.e7f28dfdf2":"The hungry kind needs water and big fans to stop it overheating. Ours stays cool with plain air.","co-sus-plain-power.16a7e373e8":"No special cooling","co-sus-plain-power.01dece5f9d":"Think of it like a modern fridge against an old one. Same job, a fraction of the electricity bill.","co-sus-plain-power.a57afcc66e":"It uses far less power","co-sus-plain-power.fa9101b082":"The kind we build sips power instead. It runs on an ordinary computer, plugged into an ordinary wall socket, in an ordinary room. It does the same helpful work without draining the grid your house depends on.","co-sus-plain-power.52c3b618d3":"The gentle kind","co-sus-plain-power.082c809e97":"Most AI today runs in giant sheds full of machines that drink as much electricity as a small town. When too many of them plug in, there is not enough power left for homes, schools, and shops.","co-sus-plain-power.4b15f4c938":"The hungry kind","co-sus-plain-safety.2619890a8d":"You do not have to understand any of the technology. What matters is simple: it can stay\n          close to home, in Europe, and your information does not have to go wandering.","co-sus-plain-safety.5788093488":"Nothing has to leave the walls","co-sus-plain-safety.66588f0f05":"Under one roof","co-sus-plain-safety.ffa5ac4bf0":"Your building, with everything kept inside","co-sus-plain-safety.7615854452":"kept close","co-sus-plain-safety.40392d8c62":"Where your information lives","co-sus-plain-safety.659cd43659":"Your records, locked and local","co-sus-plain-safety.adea79eb76":"The gentle AI, in the building","co-sus-plain-safety.3edd6822e6":"Your front desk","co-sus-plain-safety.bfe08e571f":"It is built and run in Europe, often close to where you live, under the rules you already trust to look after your privacy.","co-sus-plain-safety.16c778f919":"It stays in Europe","co-sus-plain-safety.57e26f6944":"What you tell it does not have to travel anywhere. It can be kept inside the building, behind the same locked doors as the rest of your records.","co-sus-plain-safety.dca5612a20":"Your information stays put","co-sus-plain-safety.486cc2fe09":"Because it sips power, it does not need a giant shed far away. It can run in the same office, clinic, or town hall that uses it.","co-sus-plain-safety.aa4296ee6e":"It can stay in the building","co-sus-plain-scale.4b15f4c938":"The hungry kind","co-sus-plain-scale.c3640ceb95":"Just one big AI company keeps about 125,000 machines running in giant sheds. Together they pull as much power, every hour, as millions of homes.","co-sus-plain-scale.caddbef81d":"Enough electricity for 3.8 million homes","co-sus-plain-scale.52c3b618d3":"The gentle kind","co-sus-plain-scale.68e419d203":"The kind we build does the same helpful work on one ordinary computer, in one ordinary room, plugged into one ordinary wall socket.","co-sus-plain-scale.0890cd46ea":"About the same as one household","co-sus-plain-scale.b715094812":"How much power does it really need?","co-sus-plain-scale.1d1e5c382d":"a picture","co-sus-plain-scale.550619a223":"Two kinds of AI","co-sus-plain-scale.e379fa6650":"The work is the same either way. The only difference is how much power gets used up, and how\n          much is left over for the homes, schools, and shops around you.","co-sus-plain-scale.de349729b4":"homes of power, every hour","co-sus-plain-scale.a9718b95b3":"household of power","co-sus-plain-scale.ab8c7e47e5":"125,000 machines in giant sheds, running day and night.","co-sus-plain-scale.4a20c17f66":"One ordinary computer, in one ordinary room, on one wall socket.","co-sus-plain-scale.d3f72ee733":"The same power, drawn as homes you can picture","co-sus-plain-waste.d241f72b09":"Weight leaving","co-sus-plain-waste.bd88489720":"How often","co-sus-plain-waste.24a8bd149f":"what gets thrown away","co-sus-plain-waste.65b9225761":"Ten years on a shelf","co-sus-plain-waste.8095a5ff4d":"Standard parts with established recycling and refurbishment routes.","co-sus-plain-waste.c281e2f49f":"15 to 50 kg over ten years","co-sus-plain-waste.d609e50a8b":"One or two over the same ten years","co-sus-plain-waste.407e28051e":"kept, upgraded, then handed on","co-sus-plain-waste.41e4dfd670":"Ordinary machines running the gentle kind","co-sus-plain-waste.c3e3e55eff":"Rare earth minerals and cooling parts that are hard to recycle.","co-sus-plain-waste.aa6f0c86c4":"390 to 1,300 kg over ten years","co-sus-plain-waste.3c93d68988":"A new one every 1 to 3 years","co-sus-plain-waste.c37af46c7b":"replaced as the models move on","co-sus-plain-waste.68e128a231":"Machines built for the hungry kind","co-sus-power-anatomy.f0c3c2910a":". Scenario context: the entire\n          population of the Netherlands is 17.9 million people in 7.9 million households. The cited fleet\n          figure is a scale comparison, not an independent measurement on this page.","co-sus-power-anatomy.9df97728f1":"3.8 million Dutch households","co-sus-power-anatomy.43dff28b0c":"of continuous power. That is\n          enough to power","co-sus-power-anatomy.58836dfb3f":"1.14 gigawatts","co-sus-power-anatomy.ecc0ae4103":"Published scenario: 125,000 servers draw","co-sus-power-anatomy.09262e48d2":"October 2025","co-sus-power-anatomy.9ac314a594":"The scale of AI infrastructure","co-sus-power-anatomy.9347f19739":"Modelled equivalent of keeping 92 households supplied for a year, from one server at 60% utilisation.","co-sus-power-anatomy.1a7b2eb7e9":"Annual impact, 60% utilisation","co-sus-power-anatomy.1b72ae1e92":"Total power draw","co-sus-power-anatomy.1b374da0b3":"Subtotal, IT equipment","co-sus-power-anatomy.72c95eae36":"Where the stated 9,150W goes","co-sus-power-anatomy.6b7fde8840":"Stated 9,150W total","co-sus-power-anatomy.33a7336d4c":"one server, 60% utilisation","co-sus-power-anatomy.8cac609a4a":"power.anatomy / gpu-baseline","co-sus-power-anatomy.479a5997de":"AI servers, around 100K GPUs.","co-sus-power-anatomy.b780a23b53":"Third reported fleet","co-sus-power-anatomy.03430cefe7":"AI servers, around 600K GPUs.","co-sus-power-anatomy.8d546a6dea":"Second reported fleet","co-sus-power-anatomy.eb5bfbb92b":"AI servers deployed, around 1M GPUs.","co-sus-power-anatomy.a19ee5a9fd":"Largest reported fleet","co-sus-power-anatomy.73de291a9b":"92 households","co-sus-power-anatomy.fdd6abe4d3":"Equivalent to","co-sus-power-anatomy.d8673481aa":"18,919 kg CO2e","co-sus-power-anatomy.83f05c43f2":"CO2 emissions, 393g/kWh","co-sus-power-anatomy.44b82b55f6":"EUR 14,442/yr","co-sus-power-anatomy.171bac4aac":"Cost, EUR 0.30/kWh NL avg","co-sus-power-anatomy.d8047facf8":"48,139 kWh/yr","co-sus-power-anatomy.ee07d0e1a1":"Energy consumption","co-sus-power-anatomy.2608841282":"Datacenter overhead, PUE 1.5x","co-sus-power-anatomy.3e23b969fb":"CPU, memory, storage","co-sus-power-anatomy.c5450dde92":"8x dense GPU accelerators","co-sus-resource-balance.27c3d34b19":"EU-wide shortage","co-sus-resource-balance.44e94dde0e":"9.6 million","co-sus-resource-balance.b5b811f052":"Homes missing across Europe, 3.5% of total housing stock in 2024.","co-sus-resource-balance.fb1674217e":"Netherlands, worst hit","co-sus-resource-balance.afd99c6cab":"Growing to 453,000 by 2027. Only 82,000 homes built in 2024 against a 100,000 target.","co-sus-resource-balance.17d53e0e6a":"Germany","co-sus-resource-balance.8f3b975e5d":"Homes in 2025, down 19%. Building permits down more than 40% since 2021.","co-sus-resource-balance.aebc71dceb":"Construction costs, 2010 to 2023","co-sus-resource-balance.994f15930d":"Materials and skilled labour shortages making it worse every year.","co-sus-resource-balance.9487f58282":"Land required, typical hyperscale","co-sus-resource-balance.dd097b63e4":"500 to 1,000","co-sus-resource-balance.8f61a3be8d":"Acres per campus, of which only 30% to 40% can be developed.","co-sus-resource-balance.dc46f616c6":"Building size","co-sus-resource-balance.745e41b8e1":"4M+ sq ft","co-sus-resource-balance.f9792bd500":"Typical large-scale data centre facility footprint.","co-sus-resource-balance.a843599aca":"Concrete, carbon impact","co-sus-resource-balance.397f25034e":"Of data centre construction carbon footprint.","co-sus-resource-balance.aba366f863":"Steel, carbon impact","co-sus-resource-balance.e31466263c":"Additional structural and reinforcement steel carbon cost.","co-sus-resource-balance.6c833b7a7c":"same land, materials, workers","co-sus-resource-balance.fd1fa9aaba":"homes vs data centres","co-sus-resource-balance.df10c3b5a5":"Europe housing crisis","co-sus-resource-balance.d8100fcf08":"What one data centre consumes","co-sus-resource-balance.fc59f1e580":"In a housing crisis, every square metre, every ton of steel, and every skilled worker matters.\n          Efficient AI is not just about saving energy. It is about building the Europe we want to live in.\n          If the reported 96% comparison transfers, fewer equivalent data centres would be needed and those resources could return to homes.","co-sus-resource-balance.691a77cd21":"vs","co-sus-savings-readout.c00d96a702":"Dutch grid average, five-year horizon","co-sus-savings-readout.a3f8ad8095":"Reported comparison: same intelligence, less modelled impact on the grid.","co-sus-savings-readout.a17a57d6b8":"Reported 25x ratio","co-sus-savings-readout.865dfb9622":"per server, switch one","co-sus-savings-readout.1a0df6cb5b":"savings.receipt / 5-year","co-sus-savings-readout.5ad8c6b6b0":"Standard air cooling, no liquid required.","co-sus-savings-readout.5dfe49cb34":"Water consumption","co-sus-savings-readout.afc6c45769":"Zero","co-sus-savings-readout.a3620b7293":"Not produced over five years.","co-sus-savings-readout.d74488d0ba":"CO2 emissions avoided","co-sus-savings-readout.b8f58b723a":"90,809 kg","co-sus-savings-readout.6491d20292":"Returned to Europe over five years.","co-sus-savings-readout.605464ff75":"Grid capacity freed","co-sus-savings-readout.ef3f36ce8a":"231,065 kWh","co-sus-savings-readout.2cd75441d1":"Over five years at the Dutch average rate.","co-sus-savings-readout.a7f7a9e205":"Energy cost savings","co-sus-savings-readout.915dfba90a":"EUR 69,320","co-sus-silicon-spec.0b525ca7aa":"Sustainability, substrate","co-sus-silicon-spec.56fdceba49":"One stack, four instruction sets. The efficiency comes from the compute model, not a\n            proprietary chip you cannot buy.","co-sus-silicon-spec.ee1d573904":"Runs unchanged on","co-sus-silicon-spec.75c83d54b1":"no custom accelerator","co-sus-silicon-spec.45f7f5e080":"binary compute, fixed-point","co-sus-silicon-spec.bcfad38212":"substrate.spec / why-it-holds","co-sus-silicon-spec.489ec77f31":"Fixed-function fabric","co-sus-silicon-spec.cef7f9bd36":"FPGA","co-sus-silicon-spec.706799ff62":"Open silicon","co-sus-silicon-spec.f88bb72fe2":"Edge and data centre","co-sus-silicon-spec.29d8dec43e":"ARM","co-sus-silicon-spec.e798ef2254":"Commodity server CPU","co-sus-silicon-spec.f561a3d86f":"EU, on-premise or sovereign cloud, no foreign hot-path","co-sus-silicon-spec.5d1cc3c5c6":"Jurisdiction","co-sus-silicon-spec.aee5903681":"Standard air cooling, no liquid loop, no water draw","co-sus-silicon-spec.fc36c7df6a":"Thermals","co-sus-silicon-spec.ed52b07e5d":"Up to 96% less than an FP16 GPU stack, reported 25x ratio","co-sus-silicon-spec.437bcb1511":"Energy","co-sus-silicon-spec.d48992ac9c":"Hundreds of thousands of hand-tuned ops across the backends","co-sus-silicon-spec.d5617d27c9":"Kernels","co-sus-silicon-spec.3c8e270e11":"Binary-first, no dense FP16 matrix multiply on the stated critical path","co-sus-silicon-spec.dfe65f6bd5":"Compute model","co-sus-silicon-spec.fe6baa176d":"Correctly-rounded fixed-point, zero numerical drift","co-sus-silicon-spec.45f053dc73":"Arithmetic","co-sus-sovereign-stack.7e625f8d56":"The reported low-draw comparison is the enabler: 366W on a standard outlet is what can make on-premise sovereignty practical.","co-sus-sovereign-stack.dd614fa0fd":"Deployment modes","co-sus-sovereign-stack.c750223864":"Reported 366W comparison per server","co-sus-sovereign-stack.e5e6dce47e":"EU jurisdiction, no foreign hot-path","co-sus-sovereign-stack.69eba08e3b":"deploy.sovereign / placement","co-sus-sovereign-stack.5fdc5540b0":"Standard outlet","co-sus-sovereign-stack.8b65207849":"Power connection","co-sus-sovereign-stack.9681edd3b5":"Documented","co-sus-sovereign-stack.4d9b3e8802":"EU AI Act GPAI energy documentation","co-sus-sovereign-stack.5d18f4b57d":"In region","co-sus-sovereign-stack.c10e83c453":"Data residency","co-sus-sovereign-stack.87a7e2a04c":"Foreign hot-path","co-sus-sovereign-stack.e2c5dab8a8":"Federated EU compute across nodes, with privacy-preserving cryptography at the layers that touch customer data. Distributes the workload without centralising it in one hyperscale shed.","co-sus-sovereign-stack.bde2fbb6f5":"Encrypted","co-sus-sovereign-stack.e1a2d1b150":"Mesh-distributed","co-sus-sovereign-stack.f45f7070d1":"Hosted inside EU jurisdiction with data residency held in region. Same binary substrate, same figures, with the workload placement and audit trail kept under EU rules.","co-sus-sovereign-stack.c88cdae185":"EU only","co-sus-sovereign-stack.4919c6c9e9":"EU sovereign cloud","co-sus-sovereign-stack.609beffada":"The full stack runs with no outbound connection at all. Because there is no FP16 accelerator and no foreign hot-path dependency, the deployment is self-contained and verifiable offline.","co-sus-sovereign-stack.fe6168e59f":"No network","co-sus-sovereign-stack.5df2ebe7c0":"Air-gapped","co-sus-sovereign-stack.d33aecf711":"The reported 366W comparison is intended to fit an ordinary outlet in your own building, subject to workload and site checks. In an on-premise deployment, advanced AI may not require a datacenter contract and data can stay on the premises.","co-sus-sovereign-stack.c1eaeffeb5":"No egress","co-sus-sovereign-stack.a767bb0a51":"On-premise","co-sus-sovereign-stack.7548ab52c3":"Power","co-sus-sovereign-stack.21a1ed6ef2":"Reported 366W outlet","co-sus-sovereign-stack.e5e429bcc9":"Data","co-sus-sovereign-stack.4140eed4ac":"On premises","co-sus-sovereign-stack.8e05a902ed":"Egress","co-sus-sovereign-stack.53ebc572b4":"Network","co-sus-sovereign-stack.dda6ac27b9":"Verify","co-sus-sovereign-stack.e01fa717ba":"Offline","co-sus-sovereign-stack.f1938141fd":"Foreign path","co-sus-sovereign-stack.e92452c8d3":"Absent","co-sus-sovereign-stack.64e63eea6a":"Residency","co-sus-sovereign-stack.bb11a8e3f8":"Rules","co-sus-sovereign-stack.d3dca5bc6d":"EU jurisdiction","co-sus-sovereign-stack.bee831a169":"Substrate","co-sus-sovereign-stack.25dbebb1f1":"Same figures","co-sus-sovereign-stack.68f4127b67":"Compute","co-sus-sovereign-stack.03600fcf60":"Federated EU","co-sus-sovereign-stack.e849494484":"Crypto","co-sus-sovereign-stack.ddd64b4fdd":"Privacy-preserving","co-sus-sovereign-stack.ab89d05c4b":"Placement","co-sus-sovereign-stack.dd7e2b98d4":"No central shed","co-sus-sovereignty-pillars.df2685d893":"On-premise intelligence","co-sus-sovereignty-pillars.e312caed98":"The reported low-draw comparison is intended to make advanced AI feasible in an office on a standard power connection. In an on-premise deployment, data can stay in the building; residency, security, and regulatory outcomes depend on configuration.","co-sus-sovereignty-pillars.6a99300a46":"EU AI Act, scoped","co-sus-sovereignty-pillars.46fe139555":"The AI Act requires known or estimated energy consumption in technical documentation for providers of general-purpose AI models (GPAI) under Annex XI; systemic-risk GPAI has additional energy-efficiency assessment. The Dweve comparison is not a compliance certification.","co-sus-sovereignty-pillars.93170b486f":"Grid-friendly AI","co-sus-sovereignty-pillars.72e9bd2451":"No special power infrastructure needed. No data centre moratoriums to navigate. No years-long wait for grid connections. Our grid-friendly AI deploys where your business needs it, when you need it.","co-sus-sovereignty-pillars.b8cfc79c7a":"Renewable requirements","co-sus-sovereignty-pillars.e7bc8b52db":"A June 2026 German government proposal would move the requirement for data centres to 100% renewable energy from 2027 to 2030; verify the enacted scope before relying on it.","co-sus-sovereignty-pillars.40c7609b24":"sovereignty / in-practice","co-sus-sovereignty-pillars.3051a0bb68":"EU jurisdiction, end to end","co-sus-sovereignty-pillars.090ac7b7dc":"4 pillars","co-sus-sovereignty-pillars.f2b76b78d8":"Lower-carbon AI can support data sovereignty when deployment scope is controlled.","co-sus-sovereignty-pillars.90d3c9d022":"EU, on-premise or sovereign cloud","co-sus-sovereignty-pillars.status-on-premise":"Data can stay on-premise","co-sus-sovereignty-pillars.status-gpai-scope":"GPAI scope","co-sus-sovereignty-pillars.status-existing-connection":"Existing power connection","co-sus-sovereignty-pillars.status-proposal-scope":"Proposal, verify scope","co-sus-technical-glyph.a6f393ebbd":"Energy ledger","co-sus-technical-glyph.51af3d32a6":"per server, one year","co-sus-technical-glyph.6d3551d70d":"reported","co-sus-technical-glyph.5f3c530858":"Continuous draw","co-sus-technical-glyph.d5a7cbd732":"FP16 GPU baseline","co-sus-technical-glyph.f6fd158830":"Dweve server","co-sus-technical-glyph.552b568d09":"accelerators","co-sus-technical-glyph.86dd1cf451":"host","co-sus-technical-glyph.c03158d3b0":"datacentre overhead","co-sus-technical-glyph.3b96cdb14b":"not drawn","co-sus-technical-glyph.7c2c05a297":"Per server, per year","co-sus-technical-glyph.437bcb1511":"Energy","co-sus-technical-glyph.5569ac434d":"Cooling","co-sus-technical-glyph.ba8ac97e12":"Silicon","co-sus-technical-glyph.6a353bfd21":"Hardware life","co-sus-technical-glyph.6a0e7ca9af":"liquid loop","co-sus-technical-glyph.f6392ec9ea":"air, no water","co-sus-technical-glyph.03c21c2db5":"custom accelerator","co-sus-technical-glyph.94206e8443":"standard hardware","co-sus-technical-glyph.85fa5dd16e":"one to three years","co-sus-technical-glyph.2c068d13ef":"standard server life","co-sus-technical-glyph.87e8463980":"overhead","co-sus-technical-glyph.89ba77a7db":"utilisation","co-sus-technical-glyph.d223dbe234":"carbon","co-sus-technical-glyph.551b672481":"tariff","co-sus-technical-glyph.bee831a169":"Substrate","co-sus-technical-glyph.9c00a6d2f0":"binary-first decisions","co-sus-technical-glyph.e9442a9346":"fixed-point integer math","co-sus-technical-glyph.58bea7d3f2":"four to eight specialists per query","co-sus-technical-glyph.fc2f9eb280":"no dense matrix multiply","co-sus-technical-glyph.b76ab65957":"Hardware","co-sus-technical-glyph.0b47a1e184":"nothing custom to fabricate","co-sus-technical-glyph.62e71073b0":"longer useful life","co-sus-technical-glyph.0a5e7a0583":"Boundary","co-sus-technical-glyph.828d338a9b":"source","co-sus-technical-glyph.aacd3103f3":"scope","co-sus-technical-glyph.37ed05d8d4":"cooling","co-sus-technical-glyph.47dc77c82b":"Dweve-reported","co-sus-technical-glyph.6b75d1d0a7":"stated comparison","co-sus-technical-glyph.44f36670c4":"air only","co-sus-technical-glyph.85bcbb62eb":"per server","co-sus-technical-glyph.181c2f4376":"one year","co-sus-technical-glyph.7022f47fc1":"air cooled","co-sus-technical-glyph.a268e26994":"96% less energy","co-sus-technical-glyph.ea760bfbec":"366 W","co-sus-technical-glyph.3c60eb4968":"9,150 W","co-sus-technical-glyph.cef7f9bd36":"FPGA","co-sus-technical-glyph.29d8dec43e":"ARM","co-sus-technical-glyph.a0d0966a3d":"EUR 0.30/kWh","co-sus-technical-glyph.2f9d578c58":"393 g CO2e/kWh","co-sus-technical-glyph.fbf4ef967d":"PUE 1.5x","co-sus-technical-glyph.e612abafb2":"48,139 kWh","co-sus-thermal-path.4840d915c3":"What the shorter path removes","co-sus-thermal-path.bf0b0b20a2":"The same heat path, shortened to the parts a standard server already has.","co-sus-thermal-path.99ba82f10d":"Stated Dweve substrate","co-sus-thermal-path.4959f251ac":"Four stages, each with its own inventory and its own failure modes.","co-sus-thermal-path.51b6ad0d79":"Dense accelerator baseline","co-sus-thermal-path.b6ac35f81d":"one chassis, two paths","co-sus-thermal-path.9930c6594d":"Where the heat goes","co-sus-thermal-path.300fa089a2":"No leak path across live boards","co-sus-thermal-path.c949abd6dc":"No operational water footprint","co-sus-thermal-path.f404093094":"No pump or coolant inventory","co-sus-thermal-path.a4a68c30bc":"No liquid loop","co-sus-thermal-path.77a33cac0f":"Ordinary building ventilation","co-sus-thermal-path.daa8fdbd6b":"Room air","co-sus-thermal-path.deb463da22":"Standard air movement, no coolant","co-sus-thermal-path.d3bfa3d970":"Chassis fans","co-sus-thermal-path.b414b1445c":"Reported 366W across the whole server","co-sus-thermal-path.e879b0be01":"Board","co-sus-thermal-path.cf54470edc":"Operational consumption, continuous","co-sus-thermal-path.533bc49559":"Water draw","co-sus-thermal-path.f415ed4413":"Loop handed to facility water","co-sus-thermal-path.7de5a58cde":"Heat exchanger","co-sus-thermal-path.9d9c3a2a33":"Coolant moved under pressure","co-sus-thermal-path.d2122f906e":"Pumped loop","co-sus-thermal-path.bbe553f246":"Bolted to each accelerator","co-sus-thermal-path.ce623da3a8":"Cold plate","company-sustainability.43864e310a":"EU end to end","company-sustainability.6ce59c3b7c":"Same substrate","company-sustainability.059ec357d2":"Federated EU compute with privacy-preserving crypto.","company-sustainability.e1a2d1b150":"Mesh-distributed","company-sustainability.4c03b125cd":"Data residency held in region under EU jurisdiction.","company-sustainability.4919c6c9e9":"EU sovereign cloud","company-sustainability.f8619a12de":"Self-contained, verifiable offline, no outbound connection.","company-sustainability.5f88a423db":"Air-gapped, no network","company-sustainability.e8273c2824":"A reported 366W comparison on a standard outlet can keep data inside the building.","company-sustainability.4cb182083c":"On-premise, no egress","company-sustainability.df1a9da39d":"Egress by mode","company-sustainability.273e66d878":"DEPLOYMENT POSTURE","company-sustainability.b4e55a3efa":"NO FOREIGN PATH","company-sustainability.b9836279e6":"EU ONLY","company-sustainability.ebf8bb83be":"Sovereignty is an engineering property here. With no FP16 accelerator or foreign dependency in the hot path, the full stack can run on-premise from a standard outlet, air-gapped without a network, in an EU sovereign cloud, or Mesh-distributed across federated EU nodes. The reported 366W comparison is what makes advanced AI feasible inside a building you control, because the draw fits the supply that building already has.","company-sustainability.8625265d68":"on-premise sovereignty real","company-sustainability.f701d81727":"Low draw is what makes","company-sustainability.0b01815d4d":"Efficiency is sovereignty","company-sustainability.1beab6d47b":"Five years","company-sustainability.4867f61838":"Per server","company-sustainability.5fb30b0490":"Air cooled, so no operational water footprint for thermals.","company-sustainability.37d2c84558":"Zero water","company-sustainability.594c4d8a7e":"Emissions never produced at 393g CO2e per kWh.","company-sustainability.99a4932eb9":"90,809 kg avoided","company-sustainability.c46af214d1":"Capacity returned to the grid that a connection queue can use instead.","company-sustainability.9fb570f4c7":"231,065 kWh freed","company-sustainability.d3d63f407c":"EUR 72,210 GPU energy against EUR 2,890 Dweve over five years.","company-sustainability.406987c0ce":"EUR 69,320 saved","company-sustainability.ed5eb8c8eb":"Per server, switch one","company-sustainability.6e4b1262bd":"5-YEAR SAVINGS","company-sustainability.6ef7f44788":"ZERO WATER","company-sustainability.b8f58b723a":"90,809 kg","company-sustainability.ef3f36ce8a":"231,065 kWh","company-sustainability.915dfba90a":"EUR 69,320","company-sustainability.66e9d359c8":"Over five years at the Dutch grid average, the per-server figures compound into modelled totals for cost saved, grid capacity freed, and emissions avoided. If the reported 96% comparison transfers to the stated baseline, the arithmetic implies fewer data centres and less associated land, concrete, steel, and labour. The totals below are one server, so multiply by the racks you actually run.","company-sustainability.cc9c42b1ac":"and read the totals","company-sustainability.5c73278eee":"Switch one server","company-sustainability.9c843bc3ea":"The five-year horizon","company-sustainability.e3d48dbc5e":"Half a country","company-sustainability.98f5ea20f1":"One operator","company-sustainability.94fe7ed937":"Multiply across the fleet","company-sustainability.e612abafb2":"48,139 kWh","company-sustainability.7c2c05a297":"Per server, per year","company-sustainability.c0b31c8556":"100,000 GPUs","company-sustainability.fee09116fd":"Third reported fleet","company-sustainability.2834944e8c":"600,000 GPUs","company-sustainability.7e78ce5e6c":"Second reported fleet","company-sustainability.278b918eba":"1,000,000 GPUs, about 1.14 GW","company-sustainability.651cbdfe15":"Largest reported fleet","company-sustainability.b5852a2755":"Public deployment figures","company-sustainability.37837cc698":"FLEET COUNTS, OCT 2025","company-sustainability.314e9201b1":"96% MODELLED","company-sustainability.acbb85f5fb":"3.8M HOMES","company-sustainability.eaf5b8bf69":"1.14 GW","company-sustainability.41a1693eb1":"125K SERVERS","company-sustainability.6661e3e661":"The per-server figure only matters because it multiplies. Public figures cited in this scenario put one operator at about 125,000 dense FP16 servers, drawing roughly 1.14 gigawatts continuously, enough for 3.8 million Dutch households against the country's 7.9 million. If the reported per-server comparison transfers, the modelled datacenter count, grid load, and e-waste fall with it, and so does the connection queue that gates all three.","company-sustainability.05ad16a09c":"multiplied by the fleet","company-sustainability.efa9599d01":"One server draw","company-sustainability.d423e5e1e8":"At industry scale","company-sustainability.1766aa0717":"Not a chip","company-sustainability.dfe65f6bd5":"Compute model","company-sustainability.f6bd3a576f":"Low draw means standard air cooling, no liquid loop, no water footprint.","company-sustainability.fd4eaa8a10":"Air thermals","company-sustainability.fdc78362c0":"x86, ARM, RISC-V, or FPGA. No proprietary accelerator on the replacement treadmill.","company-sustainability.02db0c8a7c":"Commodity silicon","company-sustainability.00a18548f2":"Correctly-rounded fixed-point arithmetic, zero numerical drift, cheaper per operation.","company-sustainability.dae34fecdf":"Fixed-point math","company-sustainability.828373d3d3":"Binary compute replaces the dense matrix multiply that burns the watts.","company-sustainability.04dd47d8aa":"No FP16 hot path","company-sustainability.8958733c4b":"Substrate properties","company-sustainability.d777cf1cc0":"WHAT REMOVES THE DRAW","company-sustainability.05c1584619":"AIR COOLED","company-sustainability.280208592c":"PORTABLE","company-sustainability.d05fb01e0b":"BINARY","company-sustainability.16433aafaa":"The reported reduction is not presented as a tuning trick, but as a consequence of the compute model. Binary-first compute and correctly-rounded fixed-point arithmetic remove the dense FP16 matrix multiply that dominates a GPU stack. The stated design runs on standard x86, ARM, RISC-V, or FPGA hardware, air cooled, with no custom accelerator to fabricate or replace.","company-sustainability.86aac57c43":"in the substrate","company-sustainability.9503f6a6d5":"The efficiency is","company-sustainability.dc9594d7da":"Why the figures hold","company-sustainability.22e055e63d":"Reported 96% cut at source","company-sustainability.71d4f10913":"No dense hot path","company-sustainability.a3eb573f01":"Maps to x86, ARM, RISC-V, or FPGA. No 700W accelerator to feed.","company-sustainability.62e3b74793":"SIMD silicon","company-sustainability.858d7b0956":"Cheaper per op than FP16, zero drift, no floating-point unit array.","company-sustainability.30695cac10":"Only 4 to 8 of 528 domain specialists fire per query, so most weights stay cold.","company-sustainability.8c245c9f89":"Sparse activation","company-sustainability.1affb64063":"Binary unit of work","company-sustainability.e4b64a1245":"Dense vs sparse binary","company-sustainability.6de691384f":"SIMD","company-sustainability.ca6f1cd269":"SPARSE 4-8","company-sustainability.d82b1d3897":"The reported 96% comparison is not described as a tuning trick, but as a different unit of work. A GPU inference is billions of FP16 multiply-accumulates with every parameter firing. The described substrate decides in binary constraints, routes to 4 to 8 domain specialists per query, and runs fixed-point integer math on SIMD lanes. Read it axis by axis: the dense matrix multiply that burns the power does not run on the stated path.","company-sustainability.d4a88d3f9c":"op by op against binary","company-sustainability.7c9ae9e423":"Dense FP16,","company-sustainability.ad19e15a06":"Where the watts go","company-sustainability.f67dd0037d":"25x reported ratio","company-sustainability.5902e88c6f":"96% less (reported comparison)","company-sustainability.a3434145b1":"From 18,919 kg per year","company-sustainability.0416bee0b0":"757 kg","company-sustainability.3eb302108f":"CO2 per year","company-sustainability.aa62a0faa2":"From EUR 14,442 per year","company-sustainability.ba910da5c8":"EUR 578","company-sustainability.002e89ed7d":"Annual cost, NL","company-sustainability.0901e22876":"From 48,139 kWh per year","company-sustainability.f83e928509":"1,926 kWh","company-sustainability.15c3142b9e":"Annual energy","company-sustainability.474b9ebb94":"From stated 9,150W board-level","company-sustainability.cb281ee6a6":"Power draw","company-sustainability.be4b08a292":"GPU baseline to Dweve","company-sustainability.09163fe72d":"THE REPORTED 96% COMPARISON","company-sustainability.e6e18a38cf":"757 vs 18,919 kg","company-sustainability.374d324788":"EUR 578 vs EUR 14,442","company-sustainability.bef8b9fd58":"1,926 vs 48,139 kWh","company-sustainability.f726016996":"Reported 366W vs stated 9,150W","company-sustainability.28f50aaa4e":"These are Dweve-reported values for a per-server comparison with one published dense-GPU server baseline, at 60% utilisation and the Dutch grid average. Power, energy, cost, CO2, and cooling are shown side by side. The comparison is 96% lower on each line, equivalent to a 25x ratio. Independent verification and raw output are not published on this page, so treat each line as a reported comparison rather than an audited result.","company-sustainability.8eea2d6245":"line by line","company-sustainability.57f99ac4ad":"Per server,","company-sustainability.c66fdc0ada":"The energy ledger","company-sustainability.efffac7f8e":"On your side","company-sustainability.dfad7f6e00":"No catch","company-sustainability.4886023311":"Far less electricity burned to do the work.","company-sustainability.17385cdf47":"Pollution","company-sustainability.526cb7425a":"Less","company-sustainability.a1c21eb73d":"Built in Europe, often close to home.","company-sustainability.15edfb4da4":"Made in","company-sustainability.c82d1b09cf":"EU","company-sustainability.abf4c1dd2c":"Air, like a fan, no water needed.","company-sustainability.5569ac434d":"Cooling","company-sustainability.34118050b2":"Plain","company-sustainability.3cfe646668":"It works quietly behind the places you use.","company-sustainability.c2d8f752ad":"For you to do","company-sustainability.6eef664840":"None","company-sustainability.9a4e0aabc1":"Four plain facts","company-sustainability.11405ca028":"THE SHORT VERSION","company-sustainability.c645a6bf30":"KINDER","company-sustainability.b396600e58":"EUROPEAN","company-sustainability.bb1bed8321":"GENTLE","company-sustainability.493f11f4dd":"EASY","company-sustainability.7093bd8046":"If you are wondering whether you have to do anything, whether it really uses less power, or where your information ends up, here are the plain answers. None of them have a catch.","company-sustainability.86527acadb":"to the obvious questions","company-sustainability.07265d0af1":"The honest answers","company-sustainability.d1bcd80705":"The worries","company-sustainability.31dba1632c":"Kept private","company-sustainability.9777a3ee43":"Close by","company-sustainability.1605d606a0":"Built and run here, under the privacy rules you already trust.","company-sustainability.16c778f919":"It stays in Europe","company-sustainability.05d4cba69e":"What you tell it does not have to travel anywhere far away.","company-sustainability.7c0186c734":"Your details stay put","company-sustainability.d424e36d16":"It sips so little power it can run in the same office or clinic you visit.","company-sustainability.a9bc2e1647":"It stays in the building","company-sustainability.704869f20f":"In plain words","company-sustainability.51ad7a2751":"WHY IT STAYS SAFE","company-sustainability.7d3e660795":"KEPT PRIVATE","company-sustainability.a1f21b5ee4":"IN EUROPE","company-sustainability.22b3558123":"CLOSE TO HOME","company-sustainability.4bba2814ad":"People worry, quite rightly, about where their private details end up. Here is the reassuring part. Because the gentle kind needs so little power, it does not have to live in a giant shed far away. It can sit in the same building that uses it, in Europe, with your information kept behind the same locked doors as the rest of your records.","company-sustainability.f5d7f849a0":"information lives","company-sustainability.62f70e5e95":"Where your","company-sustainability.39279d6d59":"Keeping things safe","company-sustainability.774b65a63f":"You just benefit","company-sustainability.0b049b5346":"Nothing to do","company-sustainability.c175a469bd":"Steadier power, less pressure on bills, cleaner air.","company-sustainability.6e5433950c":"You feel it","company-sustainability.49ee12444a":"There is more electricity left for homes and hospitals.","company-sustainability.374cff77bd":"The grid breathes","company-sustainability.4cb3f4984e":"It does its work on a normal computer, not a power-hungry shed.","company-sustainability.c26fa20390":"It sips power","company-sustainability.0825c2c77e":"A shop, clinic, or office runs the gentle kind of AI.","company-sustainability.08af5c4e3e":"A place near you uses it","company-sustainability.4bd75f518f":"Quietly, in the background","company-sustainability.dc2dfee782":"HOW IT REACHES YOU","company-sustainability.7c58b49271":"YOUR TOWN","company-sustainability.a67b8b0a19":"YOUR BILLS","company-sustainability.d7d4de4dd3":"YOUR HOME","company-sustainability.c08c571516":"You will never see this software or touch it. But when a shop, a clinic, or a council uses the gentle kind instead of the hungry kind, ordinary life gets a little easier. Here are five everyday changes, starting with the ones closest to your own front door. None of them ask anything of you, and none of them depend on you understanding how any of it works.","company-sustainability.79cfac450b":"but your day gets easier","company-sustainability.57e2a7ce05":"You will never see it","company-sustainability.8ac7aa173f":"What you get","company-sustainability.3d1ac91e4f":"Or gentle","company-sustainability.229099a91e":"Hungry","company-sustainability.42335b0ec4":"Both do the job. Only the power differs.","company-sustainability.cee8e25a76":"Helpful work","company-sustainability.b76aee974c":"Same","company-sustainability.7913772910":"About what the gentle kind needs instead.","company-sustainability.52996fa554":"Household","company-sustainability.b58b5a8ced":"One","company-sustainability.28675209f0":"The power those sheds use, every hour.","company-sustainability.2a68b9a645":"Of homes","company-sustainability.2cc4581ed4":"Millions","company-sustainability.1b020b1f05":"Kept running by just one big AI company.","company-sustainability.2940ff1e5f":"Machines","company-sustainability.da93388b5c":"In homes you can picture","company-sustainability.7d7792cac2":"THE SAME WORK, TWO SIZES","company-sustainability.21763d5437":"OR JUST ONE","company-sustainability.00f39c2a3c":"MILLIONS OF HOMES","company-sustainability.dc3955b394":"ONE COMPANY","company-sustainability.cffb46dd6d":"It is hard to picture how hungry the big AI sheds are, so here is a simple way to see it. One large AI company keeps about 125,000 machines running. Together they pull as much electricity, every hour, as millions of homes. The gentle kind needs about as much as a single household. Tap each one and watch the little houses light up.","company-sustainability.56365a97c4":"as much as millions of homes","company-sustainability.3d6105c206":"One company pulls","company-sustainability.4bf820b3c8":"How big is big","company-sustainability.33620b57a2":"Not the only way","company-sustainability.4b15f4c938":"The hungry kind","company-sustainability.208d8304c5":"Burning all that electricity puts a lot of pollution into the air.","company-sustainability.0dd690e433":"It costs the earth","company-sustainability.51ce189ad5":"Power it grabs is power your home, school, and shops do not get.","company-sustainability.20e1851787":"It crowds you out","company-sustainability.7da95d0aae":"It runs so hot it has to be cooled with water and big fans.","company-sustainability.9bf28068b8":"It needs cooling","company-sustainability.a05557134d":"One shed can use as much electricity as a small town, every hour of the day.","company-sustainability.3b8be3147d":"It drinks power","company-sustainability.f603c0224f":"In everyday terms","company-sustainability.3a8b97a247":"WHY THE HUNGRY KIND IS A PROBLEM","company-sustainability.1d1de639e7":"SAME JOB","company-sustainability.131948262a":"HUNGRY","company-sustainability.f519d3e061":"There are two ways to build AI. One way is hungry: it lives in giant sheds full of machines that drink as much electricity as a small town. The other way is gentle: it does the same helpful work on an ordinary computer in an ordinary room. We build the gentle kind. Here is the difference, side by side.","company-sustainability.eea94deaad":"and the gentle kind","company-sustainability.1d2a4e1dd7":"The hungry kind,","company-sustainability.550619a223":"Two kinds of AI","company-sustainability.3a0581fe6b":"Advanced AI for all","company-sustainability.92f3b17c29":"Reported 96% less to run","company-sustainability.320bd833ac":"18,919 kg","company-sustainability.92890c62bc":"EUR 14,442","company-sustainability.ccff5334de":"GPU vs Dweve, per server per year","company-sustainability.ea5d77895a":"WHAT THE REPORTED COMPARISON MEANS","company-sustainability.0ec24810f6":"WITHIN REACH","company-sustainability.5b198b2191":"ANYWHERE","company-sustainability.b48b76c076":"WALL SOCKET","company-sustainability.5b82cf71f1":"Big picture aside, here is what the reported 96% comparison could mean in practice. Plug into a standard office outlet, power serious workloads from a small solar installation, deploy where people work, and put advanced AI within reach of more organisations, not just the tech giants. Each of those four follows from the same property, which is that the draw fits inside the supply a building already has. None of the four needs a new building, a new supply or a new contract with an operator.","company-sustainability.d08599954e":"it is a socket you can use","company-sustainability.26de929e68":"Green AI is not a plan","company-sustainability.286ae60ab8":"What this means for you","company-sustainability.2f29c61829":"Ready to report","company-sustainability.4a7c05bdeb":"Mapped","company-sustainability.fbcf1053fe":"Makes every threshold far cheaper to meet","company-sustainability.65d6d9e84c":"Reported 96% less total","company-sustainability.1bbe39f6af":"Dweve","company-sustainability.ad6200af85":"Proposal target: 100% renewable","company-sustainability.49577901a2":"German renewables","company-sustainability.f3690af660":"Current German scope needs verification","company-sustainability.e3b8204056":"Proposal context","company-sustainability.9681edd3b5":"Documented","company-sustainability.42a4cdf8aa":"EU AI Act energy documentation for GPAI","company-sustainability.e8166d1d0b":"Energy disclosure","company-sustainability.84f62f0ad1":"GPAI documentation","company-sustainability.a52ecd22e7":"And the documented answer","company-sustainability.a7a1c0e2e0":"WHAT EACH RULE ASKS","company-sustainability.d791593d36":"NO GRID WAIT","company-sustainability.380f063397":"EU AI ACT","company-sustainability.0b0f1c6e9e":"Before a deployment clears procurement, someone has to answer the energy, residency, and grid questions in writing. Here is the requirement-by-requirement sheet an operations manager can hand straight to risk: every obligation mapped to a documented Dweve answer, nothing left as a promise.","company-sustainability.29f3d4191a":"ready to sign off","company-sustainability.0b0a8968dc":"The compliance answers,","company-sustainability.3ae8274331":"For procurement and risk","company-sustainability.5028fa51ac":"A lower-draw comparison may make these requirements easier to meet","company-sustainability.ac521d4914":"Less total energy","company-sustainability.5e8b20b424":"A June 2026 German government proposal moves the 100% target from 2027 to 2030; verify the enacted law and scope.","company-sustainability.508f0248a4":"German proposal","company-sustainability.a9965e6129":"Check the enacted renewable-energy scope before relying on it.","company-sustainability.bc4b174f07":"German data centres","company-sustainability.4c2ea2d10d":"Applicable scope","company-sustainability.8dca0c84eb":"Known or estimated energy consumption belongs in GPAI technical documentation; systemic-risk GPAI has additional energy-efficiency assessment.","company-sustainability.d75a4de2ef":"EU AI Act","company-sustainability.c78286b4d6":"What the rules already require","company-sustainability.dc301753ea":"COMPLIANCE LANDSCAPE","company-sustainability.f4b698f1df":"Lower-carbon AI can support data sovereignty, but the deployment determines the outcome. The reported low-draw comparison is intended to make on-premise operation feasible on a standard connection. In an on-premise deployment, data can stay in your building; residency, security, and regulatory obligations still depend on configuration and scope. Sovereignty that needs a new grid connection first is a plan, not a posture.","company-sustainability.4b10d8e3e2":"the foundation of sovereignty","company-sustainability.63b5e856a9":"Low-carbon AI is","company-sustainability.6815758f63":"Built for European sovereignty","company-sustainability.6db08161ec":"Netherlands grid data; Dublin electricity statistics; EU grid upgrade projections","company-sustainability.a195c1c803":"Dweve vs traditional AI per server","company-sustainability.6c04a99537":"Less grid load","company-sustainability.012b5e7be3":"Renewable-electricity assumption","company-sustainability.190702e12c":"EirGrid 80% figure","company-sustainability.dfc954f55d":"To upgrade Europe's grid for AI growth","company-sustainability.6ec18b041a":"By 2040","company-sustainability.75558f9e43":"EUR 200B+","company-sustainability.105adb6d7c":"Grid congestion cost to Dutch economy","company-sustainability.db93755ddf":"Per year","company-sustainability.f7bf3bda1e":"EUR 40B","company-sustainability.239254470d":"Consumption-capacity queue","company-sustainability.806d675b4d":"Businesses in queue","company-sustainability.c4626d9eba":"Why efficiency is not optional.","company-sustainability.040eb8dc69":"The grid crisis in numbers","company-sustainability.d659b2c0a8":"25x reported reach","company-sustainability.330cc95965":"Same infrastructure","company-sustainability.d275a3d5d2":"We are choosing between smart growth and hitting a wall.","company-sustainability.0ecc63411e":"A real choice","company-sustainability.320b1f19b5":"We are not choosing between progress and sustainability.","company-sustainability.d3590d6d5b":"Not a trade","company-sustainability.0e0e65cfdd":"If the reported 25x ratio transfers, deploy more AI on the same grid capacity.","company-sustainability.c737b97c68":"Build efficiently from today and AI fits within the grid we already have.","company-sustainability.7aeb4462d7":"Smart growth","company-sustainability.aaa6fa32e5":"Capability and efficiency together","company-sustainability.0021e06b44":"THE GOOD NEWS","company-sustainability.a09ab32c8e":"OUR CHOICE","company-sustainability.59c91aa625":"25x","company-sustainability.1f9a9973c6":"EFFICIENT PATH","company-sustainability.96d8d604d3":"CURRENT PATH","company-sustainability.3c13b3210d":"One path leads to limits and rationing. The other leads to abundance and accessibility. The technology exists for both, and the choice is ours. Efficiency and capability are not opposites. If the reported 25x ratio holds for an equivalent workload, the same infrastructure has more room for AI, and the queue stops being the thing that decides who gets to build.","company-sustainability.682439eb6d":"a crossroads","company-sustainability.c861fe6b7d":"We are at","company-sustainability.4addcf1251":"Two paths forward","company-sustainability.a17a57d6b8":"25x reported ratio","company-sustainability.175882ed6d":"Same intelligence","company-sustainability.eaf29aa8da":"EUR 2,890","company-sustainability.3cf72f9bf0":"EUR 72,210","company-sustainability.b2f0c3d788":"5-YEAR TOTAL COST","company-sustainability.cdf9bbefc7":"Anywhere","company-sustainability.725cd1baba":"Datacenter only","company-sustainability.d9c1ee16fc":"DEPLOYMENT","company-sustainability.351723df5c":"Standard air","company-sustainability.23a56dd665":"Liquid plus water","company-sustainability.c10728b483":"COOLING","company-sustainability.3bc4b9ed4c":"CO2 PER YEAR","company-sustainability.edd9efdb8e":"ANNUAL COST, NL","company-sustainability.9a6820057e":"ANNUAL ENERGY","company-sustainability.bb1c5bf6b5":"POWER DRAW","company-sustainability.50bd48dddf":"Standard CPU or FPGA silicon, air cooled, no water, deployable anywhere there is a power outlet.","company-sustainability.be5f63d518":"Equivalent capability","company-sustainability.ec27cfba99":"DWEVE SYSTEM","company-sustainability.c58e960d05":"A single dense-transformer GPU server, data centre only, liquid cooled, on the grid every hour of the year.","company-sustainability.93c9895ee6":"Published GPU baseline","company-sustainability.f546467dc3":"TRADITIONAL AI SERVER","company-sustainability.0238c7b348":"SIDE BY SIDE","company-sustainability.f5b02f3797":"Standard air cooling, no liquid cooling required.","company-sustainability.43d7dc3ac6":"Emissions not produced over five years.","company-sustainability.2dfc592712":"Grid capacity returned to Europe over five years.","company-sustainability.d5f3b6af74":"Energy cost savings over five years at the Dutch average rate.","company-sustainability.77a1d50205":"Switch one server, per year","company-sustainability.1c3b09a09f":"25x RATIO","company-sustainability.e9de9c0da8":"LIQUID vs AIR","company-sustainability.670513e0cf":"EUR 14,442 vs 578","company-sustainability.df2bad9258":"9,150W vs 366W","company-sustainability.5952910ddc":"One server, two approaches, dramatically different modelled outcomes for Europe’s energy future. The Dweve comparison reports less impact on the grid against an equivalent GPU baseline; workload parity and measurement boundaries still need to be checked. The gap below is a ratio, not a rounding, which is why it is worth checking rather than assuming.","company-sustainability.cc429a1542":"twenty five times as much","company-sustainability.08b183041d":"One server can cost","company-sustainability.dcae29069c":"Real impact comparison","company-sustainability.28890403f6":"Dweve-reported comparison per server, Dutch grid average, five-year horizon","company-sustainability.49e5b42609":"Per server over five years","company-sustainability.ff9fa89178":"CO2 avoided","company-sustainability.3a4eb72882":"vs EUR 14,442 per year","company-sustainability.c164f7a688":"Annual cost","company-sustainability.45c12bfd4b":"vs a stated 9,150W dense-GPU baseline","company-sustainability.024644dc8d":"Dweve draw","company-sustainability.d0eefa4963":"Same intelligence, less grid","company-sustainability.7c5614d5cb":"More efficient","company-sustainability.b51a1d3207":"vs FP16 GPU baseline","company-sustainability.aa50825f45":"Less energy","company-sustainability.5ba5ac15de":"One server. Radically different outcomes.","company-sustainability.91e777a0e7":"The numbers at a glance","company-sustainability.19508dfdac":"Longer life","company-sustainability.7e6dc8e8c4":"Less waste","company-sustainability.0afccfaa37":"Lower power and simple air cooling reduce thermal stress and mechanical failure points.","company-sustainability.a13334bc76":"Air cooled","company-sustainability.b46c059bf0":"Standard components are upgraded individually instead of replacing whole servers.","company-sustainability.76b4bb9bce":"Upgrade in place","company-sustainability.943c53d411":"Software improvements increase capability without hardware changes.","company-sustainability.02495de897":"Software, not silicon","company-sustainability.af7021b29f":"No need to replace servers when new GPU generations arrive.","company-sustainability.642aba6b00":"No specialised GPUs","company-sustainability.14737091c8":"Standard, upgradable, air cooled","company-sustainability.8a09c77cde":"WHY DWEVE CREATES LESS WASTE","company-sustainability.557a67b9cd":"TONS OF WASTE","company-sustainability.0ffc9ea370":"26% UNRECYCLED","company-sustainability.e315be06a6":"130 kg","company-sustainability.6eb15cf431":"1-3 YEARS","company-sustainability.e718995017":"Energy consumption is only part of the story. Datacenter GPU lifespans can be as short as 1 to 3 years as AI models advance, creating a hardware replacement cycle that traditional IT equipment never faced. A dense accelerator chassis of that class weighs 130 kg, and 26% of companies do not fully recycle their IT assets. Dweve runs on standard servers with longer useful lives and far less waste.","company-sustainability.7fe55033a3":"is what gets thrown away","company-sustainability.85164061bb":"The real waste","company-sustainability.64cb563226":"The hidden cost of AI hardware","company-sustainability.b7132fa231":"Fewer data centres","company-sustainability.1d65f43f54":"More land, materials, and workers available for what Europe actually needs.","company-sustainability.0d97f170a3":"Resources freed","company-sustainability.a21cfecfd1":"A modelled 96% reduction would free construction workers, electricians, and engineers from the stated data-centre baseline.","company-sustainability.d9432e9792":"Skilled workers","company-sustainability.c1c86fa8ab":"A modelled 96% reduction would divert less concrete and steel from housing. The construction-carbon comparison needs a cited method and boundary.","company-sustainability.5318a1ed56":"Construction materials","company-sustainability.e0bf5f1489":"A modelled 96% reduction would require less land than the stated server-farm baseline. A hyperscale campus could instead make room for other uses.","company-sustainability.9be8221174":"Scarce land","company-sustainability.3386ef8055":"Fewer data centres, more homes","company-sustainability.02c2b0cd04":"What 96% frees","company-sustainability.0e92d3d803":"4M SQ FT","company-sustainability.0756a5a8f1":"1,000 ACRES","company-sustainability.c3974499f6":"9.6M HOMES","company-sustainability.90e614c496":"AI data centres can draw on the same scarce land, concrete, steel, and skilled workers Europe needs for housing. If the reported 96% energy comparison translates into fewer equivalent facilities, those resources could return to other priorities; the result depends on workload, utilisation, and build assumptions. The land and the labour are the part of the bill that never appears on an energy invoice.","company-sustainability.dbf5094fee":"about electricity","company-sustainability.69004661d2":"It is not just","company-sustainability.c78a94bbfb":"When data centres compete with homes","company-sustainability.809f4ca37c":"Equivalent for one full year","company-sustainability.079af1ae6d":"92 homes","company-sustainability.c9fdc73a90":"Households powered","company-sustainability.c99265569c":"At 393g per kWh","company-sustainability.279d893900":"CO2 emissions","company-sustainability.f2d9d6a98d":"At EUR 0.30/kWh Dutch average","company-sustainability.e12ce5fe47":"Per year at 60% utilisation","company-sustainability.9b9355172f":"Energy consumed","company-sustainability.10cf2ad525":"One server, 60% utilisation","company-sustainability.6c20061045":"ANNUAL IMPACT","company-sustainability.0d175b02a7":"92 HOMES","company-sustainability.d8673481aa":"18,919 kg","company-sustainability.319a4ea873":"48,139 kWh","company-sustainability.9ecf672c49":"Let us look at the numbers, not to scare you, but because understanding the problem is the first step to fixing it. Here is the stated draw for one published dense-GPU server comparison point, and what the model implies when multiplied across an industry. The figures below are per server and per year, at 60% utilisation and the Dutch grid average, so they can be checked against your own baseline rather than taken on trust.","company-sustainability.4363d8ed44":"real impact","company-sustainability.a4f98e91a7":"A single AI server’s","company-sustainability.08b05cf341":"How AI broke the grid","company-sustainability.a3ce03e5f9":"Not a forecast","company-sustainability.abc0ea9859":"Happening now","company-sustainability.d801989d50":"EirGrid describes greater Dublin as fully constrained; its 80% figure is a renewable-electricity assumption, not an AI share.","company-sustainability.4a865a62e2":"Dublin area","company-sustainability.4acb452990":"For power connections in the Netherlands.","company-sustainability.56596260e1":"To upgrade Europe’s grid for current AI growth.","company-sustainability.2ecc424f86":"Grid congestion cost to the Dutch economy.","company-sustainability.ecd1d5da21":"What congestion already costs","company-sustainability.5ead564252":"THE COST IS REAL","company-sustainability.0ad081a1e7":"DUBLIN","company-sustainability.81f855cdec":"FRANKFURT","company-sustainability.83c352ddc9":"AMSTERDAM","company-sustainability.7850d12ea4":"NETHERLANDS","company-sustainability.2a0bf9df49":"It is not a forecast and it is not far away. Across Europe, the grid is already the constraint that decides who gets power and who waits. These are the places where the crunch is visible right now. Each of the four is a published position from the operator or the city itself, not a projection we assembled, and each one names a queue rather than a target date.","company-sustainability.3b557d5437":"one warning","company-sustainability.791b0a296d":"Four cities,","company-sustainability.cea5f51373":"Across the continent","company-sustainability.8e91e28bcf":"The stated dense-GPU comparison point is 9,150W: roughly 5,600W across eight accelerators, around 500W for CPU, memory, and storage, and a further 3,050W of datacenter overhead at PUE 1.5x. At 60% utilisation that model gives 48,139 kWh a year. The page attributes most of the draw to dense FP16 matrix multiply and its cooling; the boundary and raw measurements are not published here. Read every figure below against that boundary rather than as a bare number.","company-sustainability.841add635a":"in the matrix multiply.","company-sustainability.103cfb637f":"A GPU stack burns power","company-sustainability.7ee0cbcf9f":"01. Where the draw comes from","company-sustainability.f5a685bba6":"There is only so much electricity to go around. When giant AI facilities add demand, there is less headroom for everything else. Netbeheer Nederland reported 15,014 companies in the consumption-capacity queue at the end of 2025; connection timing varies by area and connection type. If AI uses materially less power, this pressure could ease, which is the problem we set out to address. The rest of this page is that idea in plain words, one everyday comparison at a time.","company-sustainability.ff76cd5953":"we all share.","company-sustainability.623491df30":"Power is something","company-sustainability.332ff8b94b":"01. Why this matters","company-sustainability.6cdefabb40":"Grid congestion already costs the Dutch economy EUR 40 billion every year. The page’s EUR 200 billion upgrade figure is a forward-looking scenario, not a sourced forecast here. Netbeheer Nederland reported 15,014 consumption-capacity requests at the end of 2025; timing varies, and this page does not establish that AI data centres receive priority everywhere. The constraint is unfolding across Europe today.","company-sustainability.6c4a87dcab":"it is happening now","company-sustainability.66a2eeef8a":"This is not a future problem.","company-sustainability.7f68063ec6":"01. The European grid crisis","company-sustainability.8540a26caf":"EU, no foreign hot-path","company-sustainability.5d1cc3c5c6":"Jurisdiction","company-sustainability.f975c17000":"Standard air, no water","company-sustainability.0c36ef3513":"x86, ARM, RISC-V, FPGA","company-sustainability.d35260a00f":"Targets","company-sustainability.45d8304c2d":"Binary-first, no FP16 hot path","company-sustainability.efb2a92b6a":"Correctly-rounded fixed-point","company-sustainability.45f053dc73":"Arithmetic","company-sustainability.16b3e8fb1c":"96%, 25x reported comparison","company-sustainability.c78c200d2b":"Energy reduction","company-sustainability.b921ae2a68":"Power draw, per server","company-sustainability.cda63691fe":"BINARY COMPUTE","company-sustainability.cd78c7ea11":"SUBSTRATE PROFILE","company-sustainability.72b1b3c52c":"The substrate","company-sustainability.fa1dc07b5b":"The reported 96% comparison is presented as a property of the substrate, not a procurement slogan. Binary-first compute and correctly-rounded fixed-point arithmetic replace the dense FP16 matrix multiply of the GPU baseline, and the same workload lands at 366W on standard air-cooled hardware.","company-sustainability.be0437f7bf":" the compute model.","company-sustainability.fa01f2fdee":"The efficiency is in","company-sustainability.d5625c8cd1":"Company, Sustainability","company-sustainability.32de4a0f1f":"Nothing, it just works","company-sustainability.ab3d85316b":"What you do","company-sustainability.576347ec82":"Europe","company-sustainability.58f1aea70d":"More power left for homes","company-sustainability.90e3bebca9":"Your street","company-sustainability.5033cbb707":"Less power fought over","company-sustainability.4a25a18ce4":"Your bills","company-sustainability.236530b0b4":"A normal computer","company-sustainability.52c3b618d3":"The gentle kind","company-sustainability.55e76dc8ce":"A shed the size of a town","company-sustainability.f7a9a67b82":"Plain air, like a fan","company-sustainability.c4ddca0690":"An ordinary room, ordinary socket","company-sustainability.1d197d39c5":"Where it runs","company-sustainability.21ebf689b2":"GENTLE ON POWER","company-sustainability.d0a4f77931":"IN PLAIN WORDS","company-sustainability.c800e447e9":"What it means for you","company-sustainability.d2fb002652":"You have heard that AI uses a lot of electricity. It is true: most of it runs in giant sheds that drink as much power as a small town. We build a gentler kind, one that does the same helpful work on an ordinary computer plugged into an ordinary wall socket.","company-sustainability.ea9e01cc84":" hog the power.","company-sustainability.3a7c25c6e8":"AI that does not","company-sustainability.90d3c9d022":"EU, on-premise or sovereign cloud","company-sustainability.2945ba4c78":"Hosting jurisdiction","company-sustainability.84ec9a6555":"Anywhere, standard outlet","company-sustainability.327a55f82d":"Deployment","company-sustainability.ed73d1eb07":"Annual cost, per server","company-sustainability.6a2a19d915":"Cooling model","company-sustainability.07eaf7e53a":"Standard CPU or FPGA","company-sustainability.88b98a2661":"Hardware class","company-sustainability.74b5380d9f":"Efficiency gain","company-sustainability.0bb58e1efd":"Reported 96% vs FP16 GPU baseline","company-sustainability.f4646a7586":"96% LESS IN REPORTED COMPARISON","company-sustainability.b7697131d4":"ENERGY PROFILE","company-sustainability.fa219bfb59":"The impact comparison","company-sustainability.6329e9d6d2":"Netbeheer Nederland reported 15,014 companies in the consumption-capacity queue at the end of 2025, and AI data centres add demand to grids that are already constrained. In a Dweve-reported comparison, the stated neurosymbolic workload used 96% less energy than an FP16 GPU baseline.","company-sustainability.8943cc43dd":" for everyone.","company-sustainability.b3d26ea9b3":"Keep the lights on,","company-sustainability.1dd6a17cb4":"How it works","company-sustainability.f14bbe37d4":"Start saving energy","company-sustainability.6c6724f739":"The grid crisis is real. The AI build-out is happening. But we do not have to choose between them. Efficiency is not a limitation, it is an opportunity.","company-sustainability.e765df05d7":"Let’s build the smarter future.","company-sustainability.3fe2fe497b":"EU-sovereign","company-sustainability.a268e26994":"Up to 96% less in reported comparison","company-sustainability.31bc1b17f9":"Sustainability","company-sustainability.7a1994999d":"Company","company-sustainability.9d1829a2ea":"Company, Sustainability, engineering","company-sustainability.eba8b87b6f":"Company, Sustainability, in plain words","company-sustainability.acda47627f":"COST","company-sustainability.e7404b86c9":"GRID","company-sustainability.e2487d5f4b":"CO2","company-sustainability.04efabdc79":"H2O","company-sustainability.aad3abdeab":"PREM","company-sustainability.62de62f4ea":"GAP","company-sustainability.9170a8b2fb":"EU","company-sustainability.3debcd27e7":"MESH","company-sustainability.79ffa1c22b":"EUROPE","company-sustainability.fba8ef2fb8":"GPU BASELINE","company-sustainability.bf35b544c6":"HOUSING","company-sustainability.84239a06e9":"LIFECYCLE","company-sustainability.e70b29902d":"PER SERVER","company-sustainability.35d23e720d":"SIGN-OFF","company-sustainability.056415ec85":"SAVINGS","company-sustainability.00db09ab06":"PLAIN","company-sustainability.35e0d0360a":"LIVE","company-sustainability.10a87133a3":"SAFE","company-sustainability.407d07ad0e":"MECHANISM","company-sustainability.ab2d889213":"SUBSTRATE","company-sustainability.e01c9e4eb5":"DENSE FP16","company-sustainability.7e4e26183a":"HORIZON","company-sustainability.1b8f6afa16":"PLACEMENT","company-sustainability.512b30c9e0":"We are not choosing between progress and sustainability. We are choosing between smart growth and hitting a wall.","company-sustainability.580f85794b":"There is more electricity left for homes and hospitals, so power stays steadier and bills feel less pressure.","company-sustainability.ad0d1560c7":"Rising threshold","company-sustainability.13879517b0":"Four cities, one warning","company-sustainability.be4b5e2b19":"Grid congestion costs the Dutch economy EUR 40 billion a year, and upgrading Europe's grid for current AI growth runs past EUR 200 billion by 2040. Every delayed connection is a business that cannot grow.","company-sustainability.68c485e903":"Stated 9,150W per server: about 5,600W across eight dense accelerators, 500W for CPU, memory, and storage, and 3,050W of datacenter overhead at PUE 1.5x.","company-sustainability.e40cee380a":"Correctly-rounded fixed-point replaces dense FP16 matrix multiply, so the same result lands without the heat that made it expensive.","company-sustainability.1fa05c24ef":"It does its work on an ordinary computer, not a giant power-hungry shed. So it can sit close to home and sip electricity instead of gulping it.","company-sustainability.0082d008a1":"When AI stops competing for scarce electricity, there is more to go around. That means steadier power, less pressure on everyone's bills, and cleaner air where we live.","company-sustainability.3e02eb2299":"Because the gentle kind needs so little power, it does not have to live in a giant shed far away. It can stay in Europe, in the same building that uses it, with your details behind locked doors.","company-sustainability.2f09eaa13a":"What low draw unlocks","company-sustainability.f5135bd2ce":"The enabler thesis","company-sustainability.401f309ba5":"No datacenter contract","company-sustainability.e4045d4928":"The reported 366W comparison is intended to run from an ordinary wall outlet, so advanced AI may not need a hyperscale lease.","company-sustainability.34e526d09c":"Fits existing buildings","company-sustainability.f37043d2fc":"The whole stack sits in the room that uses it, with no new grid connection to wait for.","company-sustainability.c15c13adc9":"No foreign accelerator","company-sustainability.d7fe818a24":"With no FP16 GPU in the hot path, nothing outside the EU sits on the critical path.","company-sustainability.7a54a2349c":"Verifiable offline","company-sustainability.bd81c1972a":"Air-gapped runs prove the deployment is self-contained, with no outbound call to trust.","company-sustainability.66db06f846":"Low draw","company-sustainability.47482ae636":"is the lever","company-sustainability.de17a59c46":"Plain air cooling","company-sustainability.c5224dd98c":"Made in the EU","company-sustainability.665135c40f":"Less pollution","company-sustainability.9c6e1a2f70":"15,014 (end-2025 queue)","company-sustainability.4d7e2c1a90":"80% renewable assumption","company-sustainability.0c6a2e4f81":"2030 proposal","company-sustainability.0b7d3f6a92":"Scope check","company-sustainability.7f2e4a1c90":"Claim scope","company-sustainability.8c4d0f2e71":"Context sources","company-sustainability.1a7c9e5d42":"The energy figures below are Dweve-reported values from a stated per-server comparison, not an independent certification. Confirm workload parity, correctness, hardware and software versions, power boundary, utilisation, date, and raw output before using them for procurement or reporting.","company-sustainability.2d1e6f4a80":"EirGrid on constrained data-centre areas","company-sustainability.4e8b2a6d91":"Netbeheer Nederland queue figures","company-sustainability.5f3a7c1d62":"German government proposal","company-sustainability.6b9d4e2f73":"European Commission AI Act FAQ","company-sustainability.3f8b1d5e24":"These sources provide policy and grid context; they do not prove Dweve's benchmark.","company-sustainability.9e2c6b4a18":"AI Act Annex XI technical documentation","company-sustainability.773d16340d":"an existing connection avoids it","company-sustainability.6f9b5aec1a":"The queue is the constraint","company-sustainability.c687763832":"A server sized to the supply already on site skips the whole path.","company-sustainability.fa968e885b":"Existing connection","company-sustainability.7bfd801f4c":"Frankfurt reports allocation dates stretching past 2030.","company-sustainability.2541f6bb9f":"Queue allocation","company-sustainability.dcb60dad3a":"Local capacity assessed against existing and committed load.","company-sustainability.b369de1e41":"Network study","company-sustainability.322e852dd7":"Consumption capacity requested from the network operator.","company-sustainability.14c5af556b":"Connection request","company-sustainability.ba04a9b858":"where the years actually go","company-sustainability.fe0f0a8051":"One connection request","company-sustainability.90d3e10bd0":"IN PLACE","company-sustainability.3b2751bbe9":"NO NEW SUPPLY","company-sustainability.4bb6f073c7":"EXISTING LINE","company-sustainability.3d456a478b":"An efficiency figure only matters if it changes what can be deployed. Across Europe the binding constraint on new compute is not silicon supply, it is the grid connection. Netbeheer Nederland reports 15,014 companies in the consumption queue at the end of 2025, Frankfurt reports connection queues stretching past 2030, and Amsterdam has stopped approving new data centres altogether. A server sized to the supply already in the building never enters that queue.","company-sustainability.05ed40e697":"is a connection queue","company-sustainability.8952c0bfa3":"The deployment blocker","company-sustainability.1bafbe15af":"The constraint this answers","company-sustainability.1a272fd282":"ZERO","company-sustainability.e5361d84d9":"Dweve sustainability, one lever with four consequences","company-sustainability.6254ec2d71":"The reported cut follows from the unit of work, not from tuning.","company-sustainability.2b6bd62982":"Where the figure comes from","company-sustainability.092ff9c0ed":"Reported 366W at board level, standard air, no water loop.","company-sustainability.9d819fdfba":"Power and heat","company-sustainability.7a7afd315d":"Commodity x86, ARM, RISC-V or FPGA with SIMD lanes.","company-sustainability.ba8ac97e12":"Silicon","company-sustainability.b85f94e8a6":"Correctly-rounded fixed-point integers, with zero numerical drift.","company-sustainability.785c9ae026":"Binary yes or no constraint decisions instead of dense multiply-accumulates.","company-sustainability.69fa1aa564":"Unit of work","company-sustainability.ea4406ba3d":"Downstream","company-sustainability.2303836920":"One chain, four hops","company-sustainability.90a61e21a6":"Each hop below is downstream of the one above it. Change the unit of work and the arithmetic, the silicon, the thermals and the power connection all move with it.","company-sustainability.60e84ac781":"The draw is not a tuning result. It is what happens when the unit of work stops being a dense floating-point matrix multiply and becomes a binary constraint decision over fixed-point integers.","company-sustainability.8e862630f3":"Everything else follows.","company-sustainability.0142d49e03":"The lever is at the bottom.","company-sustainability.544e447f1b":"From unit of work to outlet","company-sustainability.cd5bdaa1ce":"fewer units through the same rack position","company-sustainability.0fe864151c":"Replacement is a capital cycle","company-sustainability.17850dace8":"26% gap","company-sustainability.e32608e7b9":"26% of companies do not fully recycle their IT assets; standard components have established channels.","company-sustainability.24c9309b03":"Recycling route","company-sustainability.186ca17f05":"End of life","company-sustainability.e95bd5e1b3":"15 to 25 kg","company-sustainability.5f2c4c6c41":"Roughly 15 to 25 kg for a standard server against 130 kg for a dense accelerator chassis.","company-sustainability.b7debaafad":"Weight per server","company-sustainability.90f6c6dfa8":"Per unit","company-sustainability.4ac4a2effd":"15 to 50 kg","company-sustainability.ced10bf8d9":"Longer useful life, because software improvements do not require new hardware.","company-sustainability.0a907ff1c8":"Standard CPU or FPGA servers","company-sustainability.5272676149":"Over 10 years","company-sustainability.5ff85abb7b":"390 to 1,300 kg","company-sustainability.b498ada0a1":"Rare earth minerals, specialised circuits and cooling systems that become obsolete as models advance.","company-sustainability.723fb1cfa4":"Traditional GPU servers","company-sustainability.f3538a8b28":"1 to 3 years","company-sustainability.8251ff1a22":"threshold and cost per rung","company-sustainability.785b399d4a":"Replacement, as a spec","company-sustainability.46c54fb7b9":"LESS MASS","company-sustainability.fdd4266743":"UPGRADEABLE","company-sustainability.bb89b27390":"STANDARD PARTS","company-sustainability.0af262d610":"10 YEARS","company-sustainability.e23fa1f0b1":"Energy is a running cost; hardware is a capital cycle. GPU servers bought for one model generation are commonly replaced within one to three years, which puts three to ten units through a single rack position over ten years. Standard CPU or FPGA servers stay useful longer, because a software improvement does not require new silicon and individual components can be upgraded in place. Counted in units rather than in euros, that is the difference between buying a rack position once and buying it again every other year, and it is the part of the estate a five year energy model never shows. Counted over a decade, the replacement cycle is usually the larger number of the two.","company-sustainability.52fa58ffb4":"counted in units","company-sustainability.3c2cf8c06c":"A ten year estate,","company-sustainability.f9824582a1":"Replacement cadence","company-sustainability.04d94bf064":"Dweve sustainability, placement is a residency decision","company-sustainability.5b7feb260e":"Same binary substrate, same arithmetic, same reported figures.","company-sustainability.539575bb23":"What does not change","company-sustainability.3d21a78f79":"The compute model and the reported draw do not change.","company-sustainability.356c88f931":"Same substrate, same figures","company-sustainability.c081cbf113":"Hosted inside EU jurisdiction with residency held in region.","company-sustainability.0c3847fb96":"EU-pinned region","company-sustainability.12c1dca0d7":"The full stack runs with no outbound connection at all.","company-sustainability.6f2b150dfb":"No outbound path required","company-sustainability.bab4c38c53":"Standard x86 or ARM servers behind your own door.","company-sustainability.37635ed161":"In your own building","company-sustainability.a767bb0a51":"On-premise","company-sustainability.e7332f9619":"Your choice","company-sustainability.d76faa574b":"One substrate, two homes","company-sustainability.f407b875f9":"The choice is a residency and operations decision, not a capability one. Neither placement changes the compute model, the arithmetic or the reported figures.","company-sustainability.da277b0bb7":"The stated draw is what makes both placements possible. A server that fits an ordinary power connection can sit inside your own building, and the same image, unchanged, can run in an EU-pinned region instead.","company-sustainability.0d7861d25b":"Two places to run it.","company-sustainability.132737216f":"Same binary substrate.","company-sustainability.5eb73a67ce":"Two placements, one substrate","company-sustainability.96468f47c5":"no operational water footprint","company-sustainability.a4a68c30bc":"No liquid loop","company-sustainability.fb35dab7c1":"No coolant loop and no operational water","company-sustainability.b5ccad5677":"Continuous operational water draw","company-sustainability.4190d6d0f5":"Consumables","company-sustainability.bd2e876757":"Rejected into the room like any other server","company-sustainability.fa70b1e557":"Heat exchanger handed to facility water","company-sustainability.6f178e1fcd":"Rejection","company-sustainability.c2a1493875":"Room air moved by ordinary building ventilation","company-sustainability.d526487544":"Coolant moved under pressure by a pump","company-sustainability.c10d76c9a4":"Transport","company-sustainability.7efabf48cf":"Chassis fans a standard server already carries","company-sustainability.d657342217":"Cold plate bolted to each accelerator","company-sustainability.5d2784b1c4":"Heat removal","company-sustainability.99ba82f10d":"Stated Dweve substrate","company-sustainability.51b6ad0d79":"Dense accelerator baseline","company-sustainability.4956dcf766":"what the shorter path removes","company-sustainability.5b56919aa3":"The heat path, part by part","company-sustainability.9ad17c5791":"FEWER PARTS","company-sustainability.495ca68063":"STOCK CHASSIS","company-sustainability.2181522623":"NO WATER","company-sustainability.482f3a662f":"A dense accelerator board needs a cold plate, a pumped loop, a heat exchanger and a continuous water draw to stay inside its thermal envelope. At the reported board-level draw the same heat leaves through the fans a standard chassis already has. Removing the loop also removes the pump, the coolant inventory and the leak path across live boards. It removes a maintenance regime as well, because there is no loop to drain, refill or test, and it removes the water line as a siting constraint on where a rack can go. A rack that needs only air is also a rack that can sit in a room that was never built for water.","company-sustainability.c7dc8b356e":"it is a consequence","company-sustainability.9d1d507249":"Air cooling is not a feature","company-sustainability.fc36c7df6a":"Thermals","company-sustainability.180db80637":"Dweve sustainability, the measurement frame stated once","company-sustainability.9e5f72c807":"Annual energy, cost, emissions and the five-year totals all follow from these four.","company-sustainability.cf1c2c03f2":"What is derived","company-sustainability.f36b6e9c9c":"393g CO2e per kWh on the Dutch grid mix.","company-sustainability.cb2fc39dbd":"Emissions basis","company-sustainability.5c55f67dee":"EUR 0.30 per kWh, the Dutch average rate.","company-sustainability.12d3cc0fa4":"Price basis","company-sustainability.d5e21f31b1":"Held at 60% for a full year, the same as the baseline.","company-sustainability.e894c53064":"Utilisation","company-sustainability.a3d224aea2":"Draw measured across the whole server, not per accelerator.","company-sustainability.747dbac645":"Board level","company-sustainability.763f230359":"Stated","company-sustainability.d4a77ca73c":"Four stated conditions","company-sustainability.a1d11a7920":"Every derived figure on this page, annual energy, annual cost, emissions and the five-year totals, is that one measurement multiplied out. Change a condition and the derived figures move with it.","company-sustainability.8d64e06e09":"The reported 366W is a board-level draw for one server held at 60% utilisation, priced on the Dutch grid average and converted at 393g CO2e per kWh. Those four conditions are the whole measurement frame.","company-sustainability.be3cec55cb":"Both are stated.","company-sustainability.ef911da173":"One number, one boundary.","company-sustainability.deafb21533":"The boundary of the figure","company-sustainability.adcaa1c5b9":"fewer machines buried","company-sustainability.62cac53880":"Fewer machines bought","company-sustainability.654b10fece":"26% of companies do not fully recycle their old equipment.","company-sustainability.eb07bef469":"Thrown away","company-sustainability.cc4407917c":"Every 1 to 3 yr","company-sustainability.b626a1e9c6":"Machines built for the hungry kind are swapped out every one to three years.","company-sustainability.05033aaac5":"Replaced","company-sustainability.b6486abb44":"Years 1 to 10","company-sustainability.640ff3effb":"It does the job. An ordinary machine can be upgraded a piece at a time.","company-sustainability.55bd39660c":"Worked","company-sustainability.82a1d63782":"Year 0","company-sustainability.fb3b0a6a4a":"A machine is chosen for the work people want done today.","company-sustainability.6ea4a03474":"Bought","company-sustainability.54565f140e":"bought, worked, replaced","company-sustainability.c891219bc9":"The life of one machine","company-sustainability.cf73f908d2":"LESS WASTE","company-sustainability.033e0ef30c":"LONGER LIFE","company-sustainability.37a0fa7c18":"Machines built for the hungry kind get thrown away every one to three years, because the models they were bought for keep moving on. They are heavy, they are full of parts that are hard to recycle, and there are a great many of them. Ordinary machines are kept for longer, upgraded a piece at a time, and handed on when they finally go. That is waste you never see on an electricity bill.","company-sustainability.3db9be1d76":"half the story","company-sustainability.4907e87776":"The bill is only","company-sustainability.1113c18150":"What happens to the machines","company-sustainability.a05f8c6ed1":"Dweve sustainability, reported figures for one server","company-sustainability.e8b712442d":"EUR 69,320 that a school or a clinic keeps instead.","company-sustainability.a51910713a":"Over five years","company-sustainability.c90bc3ef00":"An ordinary room with an ordinary power socket.","company-sustainability.60a65838a5":"Where it sits","company-sustainability.871020e363":"Plain air, so no water is used to keep it cool.","company-sustainability.c8adaf3ab6":"At the Dutch average price for electricity.","company-sustainability.33620369b5":"The bill","company-sustainability.9412061fb4":"What one machine uses over a full year.","company-sustainability.925cf7f285":"Electricity","company-sustainability.2591cb3615":"One machine, one year","company-sustainability.1d4e429ec9":"You will never see this bill. It is paid by the clinic, the school or the shop that uses the software, and what they do not spend on power they can spend on people.","company-sustainability.282cbbae5e":"These are the four lines that change when the work moves from a shed full of machines to one ordinary computer. They are for one machine, over one year, at the Dutch average price for electricity.","company-sustainability.fe601927e9":"A far smaller bill.","company-sustainability.46df690c0d":"Same helpful work.","company-sustainability.3ac2e7e4ca":"A year of the gentle kind","company-sustainability.5097234849":"money that stays with the people using it","company-sustainability.297b1b0346":"Reported figures, one server","company-sustainability.ec20e93ea4":"A difference of EUR 69,320 that a smaller place gets to keep.","company-sustainability.efc1f8dc57":"No plumbing, no water bill and no special room.","company-sustainability.5ce0bd9213":"Plain air","company-sustainability.3b5c6240e9":"Liquid and water","company-sustainability.328ebbfb95":"Keeping it cool","company-sustainability.c265834b7c":"The same helpful work, drawing far less of the grid.","company-sustainability.cff235cd44":"Electricity used","company-sustainability.3d7ec84df5":"At the Dutch average price, for one machine over one year.","company-sustainability.1cdd8cfc2a":"The electricity bill","company-sustainability.2507f2a7d9":"one machine, one year","company-sustainability.6101b47ba3":"What a year costs","company-sustainability.36e4577dfa":"WITHIN REACH","company-sustainability.a3d961954e":"SMALLER BILL","company-sustainability.461080f781":"ORDINARY SOCKET","company-sustainability.dd6eec877d":"Electricity is the biggest running cost of any kind of AI. The hungry kind needs so much of it that only very large companies can carry the bill. The gentle kind runs on what an ordinary room already has, so a school, a clinic or a small shop can afford to use it too.","company-sustainability.7658a18bab":"a much smaller bill","company-sustainability.9464c3ed65":"The same work,","company-sustainability.fdd0a92de0":"What it costs to run","company-sustainability.fbe0406db7":"Dweve sustainability, what the gentle kind does with a question","company-sustainability.d499dec1d1":"The work happens where you are, and it stays where you are.","company-sustainability.f7dd5c66e9":"The short version","company-sustainability.789f17f56b":"What you said does not have to travel to a far-away shed.","company-sustainability.241780221c":"Nothing is copied away","company-sustainability.91220c9854":"Ordinary power, plain air, no journey across a border.","company-sustainability.126b8318a2":"It answers","company-sustainability.0e1c34e993":"The machine doing the work is the one in the same room.","company-sustainability.d925c40150":"It is read in the building","company-sustainability.36ff6916d2":"A question typed at the front desk, or spoken to a helper.","company-sustainability.7ed149da71":"You ask","company-sustainability.ffd4461505":"Stays put","company-sustainability.805eaa77fb":"One question, four steps","company-sustainability.4482265071":"Because the gentle kind is small enough to sit in the building, the same question can be answered in the same room. Nothing has to travel unless somebody decides that it should.","company-sustainability.02e70d5328":"Most helpful software today sends what you type to a shed hundreds of miles away, waits, and sends the answer back. That journey is a large part of why so much power gets used, and it happens somewhere you cannot see.","company-sustainability.544f7560f8":"It never has to leave.","company-sustainability.c302b92020":"You ask a question.","company-sustainability.003a62eee2":"One question, one answer","company-sustainability.52c20f4b5d":"Dweve sustainability, the gentle kind explained plainly","company-sustainability.696ae8ff1d":"The helpful work can happen where you are, not somewhere far away.","company-sustainability.2094f853b7":"What that means","company-sustainability.12a4ee0ad2":"Built and run in Europe, under the rules you already trust.","company-sustainability.749c8540ad":"European rules","company-sustainability.3aadcbb073":"One ordinary computer in one ordinary room does the work.","company-sustainability.ca78602c36":"A room, not a shed","company-sustainability.67e02c9f58":"No water loop and no plumbing to install or maintain.","company-sustainability.1ae6752a41":"Plain air to cool it","company-sustainability.5f54a82fda":"It runs on a normal power connection, not a special supply.","company-sustainability.d987ef4821":"An ordinary socket","company-sustainability.378aa12ec4":"All four","company-sustainability.1b90ab86fa":"Four plain conditions","company-sustainability.d26a5073f6":"None of this asks anything of you. It is simply what has to be true before a clinic, a school or a small office can run this in the building instead of sending everything far away.","company-sustainability.94f6fa3dac":"The gentle kind does not need a giant shed, a water loop or a new power line. It needs an ordinary room and an ordinary socket, which is why it can be put where the work actually happens.","company-sustainability.ee3961d404":"Four things make that possible.","company-sustainability.328767b440":"It can sit in your town.","company-sustainability.19a4069e08":"Close to home","company-sustainability.73c20ded5e":"Reported","company-sustainability.204912d024":"Four reported constraints","company-sustainability.8447a81a76":"None of the four rows below depends on what Dweve does. All four decide what a deployment is allowed to do this year, which is why the draw of one server is a procurement question and not an environmental one.","company-sustainability.7035188d78":"The constraint is not a forecast. It is a queue with a length, a bill with a number, and a set of connection dates that grid operators have already published.","company-sustainability.37c15a8641":"The dates are already set.","company-sustainability.5c6183f6e9":"The same work, side by side","company-sustainability.cd2c2a6087":"90,809 kg per server over five years.","company-sustainability.6938f1807a":"366W at board level against the published dense-GPU baseline.","company-sustainability.282267b150":"One per-server receipt","company-sustainability.9a8b10d174":"The figures are Dweve-reported for a stated workload, on the boundary the claim-scope note above sets out. Nothing here depends on a named outside product.","company-sustainability.e860af3a68":"Every line below is one server at 60% utilisation on the Dutch grid average, held for a full year. Power, energy, cost and emissions are read from the same comparison rather than from four separate ones.","company-sustainability.0aeeb88827":"Radically different outcomes.","company-sustainability.f46174937b":"One server, one year.","company-sustainability.be123c0876":"Dweve sustainability, the grid constraint stated plainly","company-sustainability.f906c14961":"Grid congestion costs the Dutch economy EUR 40 billion a year.","company-sustainability.fa5d08a2a0":"What the queue costs","company-sustainability.89bdc4824c":"A deployment that fits the supply in the building skips the list.","company-sustainability.e5d215cba0":"Frankfurt reports connection queues stretching past 2030.","company-sustainability.f0f5fa62f3":"Queue position","company-sustainability.839e684a06":"The operator checks what the local network can carry today.","company-sustainability.8e8183e81c":"Capacity assessed","company-sustainability.0ed6ec6f51":"A site asks the network operator for a consumption connection.","company-sustainability.2cb084dfde":"Request lodged","company-sustainability.4351a5181d":"Four hops","company-sustainability.8b8ac75d2d":"A workload sized to the supply already in the building never enters that list. Drawing less is not a virtue claim here, it is the difference between deploying this year and applying for capacity.","company-sustainability.fc3362684d":"A new connection request does not get refused. It joins a list, and the list is where the years go. Netbeheer Nederland reports 15,014 companies waiting for a consumption connection at the end of 2025.","company-sustainability.78e5d7146c":"It says wait.","company-sustainability.44112e34f0":"The grid does not say no.","company-sustainability.30998451b9":"Demand arrives as a queue","company-sustainability.609bd48216":"EUR 200B","company-sustainability.dfd6b62cd1":"Air cooling follows the board-level draw; it is not a separate decision.","company-sustainability.dc67e31f8d":"no water draw","company-sustainability.43006f8612":"one chassis, two heat paths","company-sustainability.9930c6594d":"Where the heat goes","company-sustainability.3cc649719a":"One stack, four instruction sets","company-sustainability.f6d0f15ecd":"The mechanism, resolved","company-sustainability.766a4b4c30":"The per-server energy ledger","company-sustainability.68bc8a5031":"Reported figures for one server over ten years.","company-sustainability.531f47e8c7":"ten year view","company-sustainability.b8cd2d642c":"what gets bought and what gets thrown away","company-sustainability.65b9225761":"Ten years on a shelf","company-sustainability.74c5839283":"Reported comparison, one server at 60% use.","company-sustainability.1118c5079f":"five year gap","company-sustainability.1e0da2056a":"one server, one year, side by side","company-sustainability.0db75496af":"Two bills for the same year","company-sustainability.c619fcfa41":"Green AI, in practice","company-sustainability.8b9e39f166":"Two futures, one choice","company-sustainability.00e99b8e48":"Fewer units through the same rack position over ten years.","company-sustainability.f35f8d90ba":"Homes against data centres","company-sustainability.68cdd16313":"The constraint is already binding in all four of these places.","company-sustainability.74b9a2ea91":"same work","company-sustainability.7bc5b138c7":"fewer boxes","company-sustainability.7022f47fc1":"air cooled","company-sustainability.427ac0d564":"Filed","company-sustainability.cd6fdc4f6c":"Assessed","company-sustainability.33d3063284":"Waiting","company-sustainability.3b3a5d7c42":"Deploys","company-sustainability.0bc63392df":"Ordinary","company-sustainability.ad48807e64":"One room","company-sustainability.107c92ba67":"In Europe","company-sustainability.1bb7985fdb":"Asked","company-sustainability.dc99d54d99":"Local","company-sustainability.e0aafffa02":"Answered","company-sustainability.727edd6bf4":"Kept","company-sustainability.c5a5c3e974":"Air only","company-sustainability.1498b3d006":"In place","company-sustainability.90efc21e83":"Four stages","company-sustainability.be38fdd8ee":"NL avg","company-sustainability.bf98fb97c2":"Four parts","company-sustainability.4140eed4ac":"On premises","company-sustainability.5df2ebe7c0":"Air-gapped","company-sustainability.c88cdae185":"EU only","company-sustainability.51ed651132":"Unchanged","company-sustainability.7f2355fba0":"Four rungs","company-sustainability.66639f7d45":"Binary","company-sustainability.ab98f5b857":"Fixed","company-sustainability.38dfc23319":"Commodity","company-sustainability.9519efcd39":"Where the energy argument begins","company-sustainability.b227136061":"The three grounds this page builds on","company-sustainability.1c087e95c9":"Opened below","company-sustainability.86354f46c1":"Three grounds","company-sustainability.f606125231":"Each ground is opened in full further down the page.","company-sustainability.7d78bce46f":"366W","company-sustainability.b5472944c6":"757 kg CO2e","company-sustainability.600bf53111":"REPORTABLE","company-sustainability.211143bdf0":"GRID-READY","company-sustainability.11a9648486":"SOLAR","company-sustainability.afc3485919":"BUILD THE SMARTER FUTURE","company-sustainability.a515cb96c7":"Efficiency is not a limitation","company-sustainability.2414c09e87":"it is an opportunity","company-sustainability.48e07c9f02":"The figures on this page","company-sustainability.e2de41a6cd":"reported comparison","company-sustainability.573c28e9e2":"Per server draw","company-sustainability.1ae9c06cb1":"366W reported against a stated 9,150W","company-sustainability.bd928cea68":"Per server year","company-sustainability.d8623668a3":"1,926 kWh against 48,139 kWh","company-sustainability.0c0892a0d2":"Cost per year","company-sustainability.6edba794f3":"EUR 578 against EUR 14,442","company-sustainability.0c092cf47c":"Grid connection","company-sustainability.b6e252a0bf":"Existing supply, no new line needed","company-sustainability.f72bd18152":"On-premise, EU jurisdiction, air cooled","company-sustainability.0a5e7a0583":"Boundary","company-sustainability.a64162d097":"Modelled comparison, not an independent audit","company-sustainability.c88bbab640":"every figure here is a reported comparison","company-sustainability.c4cd203756":"THE GENTLE KIND","company-sustainability.280b0ce3f2":"does the same job","company-sustainability.6937091a40":"You do not have to choose between useful help and a sensible electricity bill. The gentle kind does the same work in an ordinary room, on an ordinary socket, and it keeps your information close to home.","company-sustainability.f3f0c2eee2":"plain words","company-sustainability.c5840c02e1":"How much power","company-sustainability.6137c74f86":"About as much as a single household","company-sustainability.38306105e3":"In the same building that uses it, in Europe","company-sustainability.dda1e0e0bb":"Your information","company-sustainability.cf9c5a6512":"Kept close to home, behind the same locked doors","company-sustainability.6dbc8c8707":"Small enough for a school, a clinic or a shop","company-sustainability.e73dc00076":"What you have to do","company-sustainability.d7e6810158":"Nothing, you will never see it","company-sustainability.b8179c9f93":"the same job, on an ordinary socket","company-sustainability.362b20e717":"READ THE LEDGER","company-sustainability.962b9c2cfa":"Read the ledger","company-sustainability.114d33cdca":"before you read the claim","company-sustainability.1840f1d4e2":"Every figure on this page is a reported comparison against one published dense-GPU baseline, at 60% utilisation and the Dutch grid average. The mechanism is in the substrate rather than in a tuning pass, and the boundary is stated wherever the number is.","company-sustainability.88d280d965":"The reported ledger","company-sustainability.4df44274e7":"per server, per year","company-sustainability.a4b682382c":"Board draw","company-sustainability.437bcb1511":"Energy","company-sustainability.64ae43e8fe":"Cost","company-sustainability.b12d86c362":"Emissions","company-sustainability.f05a29bfc8":"757 kg against 18,919 kg CO2e","company-sustainability.4acef0d341":"Air cooled, no water loop","company-sustainability.7d21fca653":"Modelled comparison, raw output not published here","company-sustainability.089a8dd0c6":"the mechanism is binary compute, not a tuning pass","core-extras.96d330537a":"Base IR","core-extras.bee831a169":"Substrate","core-extras.61a0822826":"Apps on Core","core-extras.15438cc098":"not supported","core-extras.ad565d9d01":"Feature","core-extras.0155bf03e5":"Segment legend","core-extras.8b7110078e":"Algorithm groups by data type","core-extras.25ab4276a6":"Scrollable capability matrix","page-breadcrumb.c766e66518":"Breadcrumb","section_st_a1_display-split.7c9a7c0610":"Detail","section_st_f1_cta-dark.90e40d5043":"Get started"}