Measure delivered accelerator-hours, not nameplate megawatts
Useful capacity is the intersection of energized IT, memory and network balance, thermal headroom, software utilization, and job completion.
Accelerators are the visible layer
An AI data center is a power plant in reverse. Each layer can strand the ones above it, and the scarce unit is reliably powered, networked, cooled accelerator-hours, not installed chips.
Grid and onsite power
- What it does
- Generation, transmission, interconnection, substations, storage, and backup establish electrical capacity.
- Measure
- MW · time to energize
Binding constraint: Interconnection queues and firm supply, not equipment delivery.
Every watt entering compute exits as heat
Power conversion can improve, but the first law remains: nearly all electrical energy consumed onsite ultimately becomes heat that must leave the equipment and facility.
Two ways the same watt gets stranded
This is why the binding lead time increasingly sits in interconnection studies, transformers, transmission, construction, and cooling equipment, not in server assembly, which can now move faster than any of them.
Compute procurement moves upstream into infrastructure
The binding lead time increasingly sits in interconnection studies, transformers, transmission, construction, cooling equipment, and community permission rather than in server assembly alone.
Flexible workloads
Shift delay-tolerant jobs across hours or regions so onsite and grid constraints bind less often.
Balanced systems
Buy memory and network capacity that keeps expensive compute occupied instead of stranded.
Thermal co-design
Match rack density, liquid loops, heat rejection, climate, and water constraints to the actual workload.
Firm interconnection
Coordinate grid investment, onsite assets, tariffs, and realistic ramp profiles years ahead of server delivery.
Who is building what
Direct-to-chip liquid manifolds, immersion cooling tanks, medium-voltage switchgear, and dedicated nuclear/geothermal generation bridge the AI power wall. Search the record, or filter by infrastructure domain.
VertivLiebert XDU & Liquid SolutionsCoolant Distribution Units (CDUs) and modular manifold systems delivering treated liquid coolant directly to multi-kW compute racks and cold plates
- Reported evidence
- Core cooling partner for NVIDIA Blackwell NVL72 architectures; high-volume manufacturing of multi-megawatt CDU systems.
- Announced next step
- Standardized liquid infrastructure supporting 100+ kW per rack compute densities with dry-cooler heat rejection.
- Unresolved risk
- Fluid leakage risks inside high-voltage compute racks, quick-disconnect coupling reliability, and pump cavitation under variable pump speeds.
Schneider ElectricEcoStruxure Data CenterIntegrated medium-voltage switchgear, modular prefabricated power skids, and liquid-ready busway architectures designed for rapid multi-megawatt deployment
- Reported evidence
- Deploying turnkey electrical packages across global hyperscale data centers; contracted billions in power infrastructure equipment.
- Announced next step
- Reducing data center electrical installation and commissioning timelines from years down to months.
- Unresolved risk
- Supply chain shortages in heavy medium-voltage transformers and circuit breakers with delivery lead times extending over 3 years.
CoolIT SystemsDirect Contact Liquid CoolingCustom micro-channel cold plates mounted directly on GPUs, CPUs, and memory modules, coupled to stainless steel rack manifolds
- Reported evidence
- Cooled many of the world's fastest supercomputers (Frontier, Aurora) and deployed broadly across hyperscale AI clusters.
- Announced next step
- Liquid cold plates supporting 1,500W+ thermal design power (TDP) per individual silicon package.
- Unresolved risk
- Corrosion in mixed-metal fluid loops, micro-channel clogs from particulates, and tight mechanical tolerances on wafer packaging.
SubmerSmartPod ImmersionSingle-phase immersion cooling submerging entire server chassis in biodegradable synthetic dielectric fluid, eliminating heatsinks and fans
- Reported evidence
- Pilots and commercial deployments with Intel, hyper-scalers, and European research computing centers achieving PUE below 1.05.
- Announced next step
- Ultra-dense AI superclusters running >100 kW per tank with zero water consumption.
- Unresolved risk
- Technician maintenance handling dripping fluid-soaked boards, component cable degradation in hydrocarbon oils, and vendor warranty voiding.
EatonEnergyAware UPSHigh-efficiency double-conversion uninterruptible power supply (UPS) systems capable of fast bidirectional grid frequency regulation
- Reported evidence
- Operating in large hyperscale facilities; allows data centers to monetize multi-megawatt battery reserves to support local grid stability.
- Announced next step
- Grid-interactive data center electrical architectures acting as dispatchable virtual power plants.
- Unresolved risk
- Battery cell degradation from frequent micro-cycling and coordinating fast inverter response with utility protection relays.
Crusoe EnergyDigital Flare Mitigation & Off-Grid ComputeModular data center containers placed directly at oilfield stranded flare gas sites, wind farms, and geothermal plants, bypassing the electrical grid
- Reported evidence
- Operates hundreds of modular compute units across North America; expanded into gigawatt-scale AI data center parks backed by long-term clean power.
- Announced next step
- Multi-gigawatt purpose-built AI data center campuses paired with dedicated behind-the-meter generation.
- Unresolved risk
- High-speed optical fiber connectivity to remote off-grid locations and fuel gas quality variability causing generator maintenance issues.
Constellation Energy & MicrosoftCrane Clean Energy Center20-year power purchase agreement to restart the 835 MW Unit 1 nuclear reactor at Three Mile Island, dedicating 100% of power to Microsoft data centers
- Reported evidence
- Landmark corporate nuclear PPA executed in 2024, setting precedent for hyperscalers directly funding baseload nuclear capacity.
- Announced next step
- Commercial restart by 2028, providing 24/7 carbon-free power matching round-the-clock AI training workloads.
- Unresolved risk
- Nuclear Regulatory Commission (NRC) licensing timelines, turbine and steam generator refurbishment costs, and public opposition.
Open Compute Project (OCP)Open Rack v3 (ORV3)Open industry specification standardizing 48V DC busbar power delivery, blind-mate liquid cooling connectors, and universal chassis form factors
- Reported evidence
- Adopted by Meta, Google, Microsoft, and server OEMs globally; eliminated redundant individual server power supplies.
- Announced next step
- Establishing universal standards for high-pressure blind-mate liquid quick disconnects and 100 kW+ rack cooling loops.
- Unresolved risk
- Proprietary server architecture divergence by hyperscalers undermining open hardware interoperability.
Data center Power Usage Effectiveness (PUE) ratings (e.g. 1.1) describe non-compute auxiliary overhead; they do not reduce the physical heat rejection required per accelerator kilowatt.
The optimistic view, with conditions
The data center becomes a grid participant
Facilities that expose safe flexibility, stage load growth, and coordinate onsite assets can connect sooner and use infrastructure more productively while bulk supply catches up.
Stage load growth
Ramp profiles that match realistic construction and interconnection timelines connect sooner than all-at-once requests.
Shift what can move
Delay-tolerant jobs move across hours or regions, buying time for firm supply to arrive.
Coordinate onsite assets with the grid
Storage and onsite generation become bridge capacity, not a substitute for transmission and generation buildout.
What usable AI capacity actually needs
- Firm interconnectionGrid investment, onsite assets, tariffs, and ramp profiles coordinated years ahead of server delivery.
- Balanced compute fabricMemory and network capacity purchased alongside accelerators, not as an afterthought.
- Thermal co-designRack density, liquid loops, heat rejection, climate, and water constraints matched to the workload.
- Flexible workloadsScheduling that can shift delay-tolerant jobs across hours or regions when power is scarce.
- Honest capacity accountingReporting delivered accelerator-hours, not nameplate megawatts or installed chip counts.
Power availability is already measurable
Lawrence Berkeley National Laboratory’s US series rises from 58 TWh in 2014 to 176 TWh in 2023; its earlier 2028 range was 325–580 TWh. A later DOE update gives a 649 TWh 2030 reference case, or 11.8% of projected US electricity, under revised assumptions. These are successive model vintages, not a single measured trajectory. Ireland’s energy agency counted data centres at 21.2% of national electricity in 2024, demonstrating that local network limits can bind before a global energy limit does.
The component bottleneck has a dated lead time. DOE’s March 2026 transformer review says distribution-transformer orders moved from roughly three to six months in 2019 to one to two years or longer in 2024. Transformer availability does not equal time to energize a 100 MW campus, which also needs land, substations, generation capacity and interconnection studies. A useful falsifier for the infrastructure thesis would be sustained reductions in median time from a signed large-load agreement to delivered power, reported with a fixed load threshold.
Onsite generation is a live response, but it does not make electricity demand disappear. The IEA tracks a 2026 US pipeline of onsite gas projects; that is project intent, not delivered low-carbon power. The useful output measure is verified model work per MWh at a stated quality and availability level, while cooling water, grid emissions and peak power are reported on their own boundaries.
Global demand and local speed have different scales
The IEA's 2026 update estimates 485 TWh of global data-centre electricity in 2025 and about 950 TWh in its 2030 case. That world projection cannot be added to the Berkeley Lab US projection above; geography and model vintage differ. IEA publishes the underlying regional PUE and load-factor dataset for comparison. A 100 kW IT rack with PUE 1.2 draws about 120 kW at the facility boundary and ultimately rejects roughly 120 kW of heat; this is an engineering identity, not a claim about a particular rack product.
Water and energy cannot be collapsed into one universal litres-per-token figure: cooling towers evaporate water locally, dry cooling trades water for power, and power plants can consume water offsite. The reported unit should state direct site water per MWh, upstream generation water, climate and cooling design. For speed, the measurable test is median months from a signed >100 MW service request to energized power, with an interconnection cohort defined; a transformer lead time alone is only one component.
Sources, method, and boundaries
Energy quantities are Berkeley Lab scenarios, not commitments or measured 2030 demand. Flexibility mechanisms come from its 2026 grid-integration review; no scenario is presented as a forecast certainty. The stack decomposition is an architectural description of how AI campuses are built, not a benchmark of any single deployed site.
- PUE
- Power usage effectiveness: total facility energy divided by IT equipment energy; lower is more efficient.
- Interconnection queue
- The utility process for studying and approving a new large electrical load before it can be energized.
- Stranded accelerator
- Installed compute that cannot be productively used because power, cooling, memory, or network capacity is missing.
