The Data Center
Power
Playbook.
A living document, provided as an open free resource for the broader AI infrastructure industry to de-risk and streamline power.
The AI revolution is now an energy and infrastructure race. Power is the primary bottleneck to scaling compute, and the industry's current defaults of grid-dependent procurement and on-site gas baseload face structural constraints that compound over time.
But there is a faster, more reliable, lower-risk path: solar+storage centric energy co-development paired with flexible grid interconnection, enabling firm compute capacity in months rather than years. This is the Playbook for how to execute.
The Problem
Power is now the binding constraint of the AI infrastructure race. Energy velocity will determine who wins it.
Target Scale
500MW+
Per site, per deployment
Time to Power
12–18mo
Not years
The Challenge
Why Now
“Power is now the binding constraint of the AI infrastructure race.”
The exponential growth of AI has created a structural compute shortage. The expansion of compute capacity is constrained by power. Industry forecasts, consistently revised upward, expect US electricity load growth as high as 50% through 2035, with AI data center load expected to surpass 11–12% of total US electricity consumption by 2030 — requiring >50GW of new capacity through the end of the decade. As of April 2026, US hyperscalers have committed over $700B in cumulative capital expenditure announcements, with annual AI infrastructure spending now approaching 2.1% of US GDP — the second-largest capital mobilization in American history (Nat Bullard, Halcyon).
As server racks continue to climb in power density, from ~130kW toward ~600kW by 2027 and 1MW by the end of the decade, compute power demand is further amplified. The grid was not built for the pace and scale demanded by AI. Median US grid interconnection timelines now stretch past 5 years and continue to increase.
Power is now the binding constraint of the AI infrastructure race, and energy velocity — defined as the speed that firm, reliable, scalable power can be secured — will determine who wins it.
50%
US electricity load growth through 2035
Driven primarily by AI data centers
$700B+
Cumulative hyperscaler CapEx committed
2nd-largest capital mobilization in US history
>50GW
New data center capacity needed by 2030
Up from ~25GW in 2024
5+ yrs
Median US grid interconnection timeline
Up from under 2 years pre-2008
The AI industry is facing a structural compute shortage, with expansion constrained by power.
The exponential growth in AI advancements and usage has created a structural shortage of accelerated compute. Total global accelerated compute base (compute attributable to AI) stands at ~8GW of installed capacity in 2025, with industry forecasts expecting a 10x increase to ~80GW globally by 2030. Annual performance improvements support an even higher increase in compute output (50x increase in compute expected to be supported by the 10x increase in power by 2030).
The scale of this demand growth is underscored by recent data from hyperscalers. In July 2025, Google cited a 100x year-over-year increase in monthly tokens processed (to ~1.3 quadrillion tokens per month). Even with algorithmic and hardware advances improving efficiency through this same period, Google needed to increase accelerated compute dedicated to token generation by an estimated 3–10x. This meant that Google needed to shift a portion of its accelerated compute away from training to token generation to serve its users. As general purpose computing continues to shift to accelerated computing, hyperscalers shift from CPUs to GPUs, and new AI applications arise, demand will only continue to outpace supply, worsening the structural compute shortage.
The capital commitments match the demand signal. US hyperscalers have already committed over $700B in cumulative capital expenditure announcements, with 2026 spending projected near $400B. Google alone has committed $175–185B in annual CapEx, with the majority directed at AI infrastructure. At 2.1% of US GDP, annual AI infrastructure spending now represents the second-largest capital mobilization in American history (% share of US GDP for year specified) — second only to the Louisiana Purchase exceeding the US railroad buildout, the Interstate Highway System, the telecom buildout, and the Apollo program (Nat Bullard, Halcyon, Transition-AI 2026).
Nvidia CEO Jensen Huang estimates $3–4T CapEx on AI infrastructure by end of decade, with ~60% directed to chips and computing hardware. Presuming $1.8–2.4T spent on compute server rack modules, this implies 600,000–800,000 server racks (~$3M total rack ASP). Power density per rack is accelerating — server racks are currently designed to operate at ~130kW, with industry roadmaps guiding to ~600kW racks by 2027 and 1MW by end of decade. At current power density, this implies +80–105GW of additional power demand. As racks become increasingly power dense, this amplifies power demand further.
Underscoring the structural compute and power supply constraint point, there are no dark GPUs sitting idle, every rack deployed is running at max utilization.
What Does This Mean for Power Demand?
Industry forecasts project US electricity load growth as high as 50% through 2035, with AI data center load projected to represent 75% of this total load growth. AI data center consumption alone is expected to surpass 11–12% of total US electricity by 2030 (up from 4–5% today) — roughly 2x the current electricity load of the entire UK. The US needs >50GW of new data center capacity by 2030 (from ~25GW in 2024), consuming ~606 TWh of electricity annually. Just meeting industry forecasts for global data center capacity growth through 2030 would require delivering as much new power supply over the next five years as the past two decades of cloud infrastructure development combined.
Third-party modeling from Aurora Energy Research projects 139 to 280GW of US data center load by 2030, with the range driven primarily by pipeline conversion rates rather than underlying demand. Even the low end of this bracket exceeds the entire capacity of any single US utility today; the high end represents nearly two-thirds of current peak US electricity demand. To serve this load, Aurora projects total AI infrastructure investment at $2T for basic data center build-out, $3T for AI-ready deployment with advanced tooling, and $6T fully loaded — approximately 25% of US GDP. The scale of demand is not the question, but rather what fraction of the announced pipeline is actually delivered. Roughly 145GW of the announced pipeline is attributed to sponsors with zero operational data center capacity — entities whose announced projects exceed their execution track record by orders of magnitude. This speculative overhang is the primary mechanical driver of the gap between announced and deliverable capacity.
Regardless, the physical grid is not built to meet this demand — but this is not an energy shortage. Generation supply exists in abundance; what's scarce is the interconnection, procurement, and delivery infrastructure to convert available energy into firm, fast capacity. Median US grid interconnection timelines exceed 5 years and are increasing — down from under 2 years pre-2008. PJM queues exceed 5+ years. Texas offers 18–24 months for large loads but remains constrained. Globally, Japan has delayed data center interconnection to 2029; Ireland implemented a moratorium on new data center grid connections, reopening selectively in January 2026 only for facilities with on-site power generation.
The scale is without precedent at the utility level. Arizona Public Service took 140 years to build 8GW of peak load. It now faces over 20GW of uncommitted data center requests, more than double its entire existing system in a fraction of the time (4.5GW committed) (Nat Bullard, Halcyon, Transition-AI 2026).
Equipment procurement compounds the constraint — high-voltage electrical transformers require 2–3 year lead times, and grid upgrades are increasingly expensive and chronically delayed. Even where generation supply and permitting are cleared, execution is labor-bound. Construction timelines for large energy projects have tripled — from roughly six months to eighteen months to break ground — driven by pre-existing skilled-labor constraints that predated the AI buildout and have been severely compounded by it. Regulatory reform alone does not compress this constraint.
Even as industry forecasts highlight significant data center load growth, they still likely understate actual demand — forecasts have consistently been revised upward, and current projections likely embed supply constraints as ceilings.
Energy velocity is the defining metric of the AI infrastructure race. It measures the speed at which firm, reliable, scalable power can be secured.
The AI compute race will be decided by energy strategy, not server procurement. Four strategic priorities, in order of precedence:
Access to Power (Firm Supply)
Without firm power, data centers cannot operate. Site selection and infrastructure deployment must follow power-first logic. Identify firm power sources before committing to facility locations. This inverts traditional fiber-first development.
Speed to Power (Time to Deployment)
Rapid access to reliable power is strategically decisive. The speed at which power can be secured AND scaled determines time-to-revenue. A six-month advantage in power procurement translates directly to months of accelerated compute deployment and tens of millions in competitive margin.
Reliability of Power (Continuity)
GPU infrastructure carries high capital amortization costs. Downtime equals stranded revenue. Firm, reliable power ensures maximum utilization and eliminates the forced shutdown risk inherent in grid-dependent deployments.
Economics of Compute (Optimization)
Cost flexibility exists only within the constraints of the preceding three priorities. Economics must be optimized for total compute cost, not levelized cost of energy (LCOE) in isolation. Power economics are ultimately inputs to compute economics.
This priority ordering is validated by how leading data center developers actually evaluate power decisions today (under a Three C framework): 1) Capacity first — can it deliver firm power on the required timeline; 2) Community second — will it survive regulatory and local scrutiny; 3) Cost third. Energy velocity and social license are the binding constraints, not LCOE.
Energy is the new currency of compute. There is a direct correlation between energy access, compute deployment, token processing, and revenue generation. Energy strategy is no longer optional. Energy velocity will decide the winners of the AI race.
Industry energy defaults are forming now. The frameworks and standards established today will determine how the next 50GW of global compute capacity gets powered. First-mover advantage accrues to those who solve energy velocity now.
The Risk
Why the Current Default Fails
“The data center industry is at risk of becoming the visible face of rising electricity costs.”
Energy represents the biggest risk for the data center industry. AI infrastructure is scaling faster than the grid can support, and the data center industry is increasingly realizing that the old power playbook — passive grid-first procurement as a standard industrial load — breaks at GW scale.
The industry's immediate default response has been a rush to procure on-site gas turbines to deliver 24/7 baseload power. However, overreliance on on-site gas baseload recreates and compounds the same infrastructure constraints the industry is trying to overcome — concentrating energy risk at the exact moment the industry needs to de-risk.
With current defaults, the data center industry is on a collision course with the communities, regulators, and grid systems it depends on. Energy affordability is already a top political priority across federal, regional, state, and local levels, as policy increasingly mandates additionality and curtailment to address it. The data center industry is at risk of becoming the visible face of rising electricity costs, worsening grid reliability, and local community opposition.
Without a proactive energy strategy, AI infrastructure companies face a narrowing window — absorbing regulatory costs, losing permitting battles, and inheriting the blame for rate increases that all erode the timeline to scale the compute we need.
77GW
New gas-fired generation planned 2026–2033
Largest thermal wave in over a decade
195%
Gas turbine price increase since 2019
Wood Mackenzie; orders full through 2028+
$64B
Data center projects canceled from community opposition
As of April 2026
11
States with legislation to ban or restrict data centers
Bipartisan, red and blue districts alike
As the grid interconnection bottleneck has become increasingly undeniable, the AI data center industry's immediate default response has been to procure on-site gas turbine engines to deliver 24/7 baseload power independent of grid timelines.
On paper, this appears to be effective in addressing the speed-to-power challenge. In practice, this recreates and compounds the same infrastructure constraints the industry is trying to escape. These constraints are only amplified further as the data center industry increasingly shifts to on-site gas generation as a new default. Halcyon Gas Power Plant Tracker indicates 77GW of new gas-fired generation planned for addition between 2026 and 2033, the largest wave of thermal buildout in over a decade.
Six specific risks undermine the on-site gas baseload thesis. These risks are correlated and compound upon one another. Data center operators committing to heavy on-site gas baseload are simultaneously exposed to equipment lead times, capital cost inflation, fuel price swings, pipeline bottlenecks, community opposition, and gas system cost and reliability externalities. This creates concentrated risk on a single energy pathway at a moment when the industry's strategic imperative is to diversify and de-risk.
Surging Equipment Lead Times
- Gas turbine orders have surged as data center operators compete for limited manufacturing capacity. Turbine prices have risen 195% since 2019 according to Wood Mackenzie, with equipment contributing 20–30% of total power plant cost. Global orders sit at 110GW against manufacturing capacity of just 65GW, with no new orders accepted until 2028 and six-year delivery cycles now standard. Without pre-existing orders in place, the speed advantage over grid interconnection has effectively disappeared. Leading OEMs now require 10% up-front deposits against delivery windows extending to 2031 — effectively converting gas turbine procurement into a multi-year capital commitment made before any revenue begins.
- Notably, of the 253GW of gas-fired capacity currently in development in the US, 146GW (more than half) have not announced a turbine manufacturer (Global Energy Monitor, H2 2025). The three major suppliers (GE Vernova, Siemens Energy, Mitsubishi Power) account for just 91GW of identified orders, underscoring the gap between announced gas projects and actual equipment procurement.
- Equipment Damage: If gas turbine equipment installed breaks, equipment supply constraints mean that there are no spare turbines immediately available for replacement. This is especially relevant given volatility in AI compute loads that can swing from full power towards zero in milliseconds, cycling ten to twenty times per second — a demand profile no thermal generator can follow. Fast-ramping gas turbines need one to two minutes; the mismatch spans three orders of magnitude. At these frequencies, periodic load profiles create resonant torsional stress on rotating equipment, accelerating fatigue and mechanical failure. Training workloads specifically generate measured oscillations of up to 70% of rated load at millisecond timescales — making battery buffering a structural requirement, not an optional enhancement.
- In contrast, the US now has solar module manufacturing capacity to produce nearly 65GW as of year-end 2025 (SEIA), up over 50% from the prior year, with domestic factories now able to fully meet US demand for the first time in over a decade. Lead times for solar panels are measured in weeks, not years.
Inflating Capital Costs
- CCGT CapEx costs have climbed to $2,500–3,000/kW, more than double 2022 levels, with GE Vernova's CEO citing turbine ASPs approaching $3,000/kW and slot reservation payments increasing toward 25% (March 2026). This has been driven by supply chain constraints, materials inflation, and surging demand. The capital intensity of gas baseload is approaching levels that fundamentally challenge project economics, especially for speculative data center developments without committed compute offtake.
Gas Pipeline Interconnection Constraints
- On-site gas generation requires firm natural gas supply, which means pipeline interconnection. Similarly to the grid, existing gas pipeline capacity is also becoming increasingly constrained. New pipeline construction faces its own permitting timelines (often 2–5 years), environmental review requirements, and community opposition. Data center operators seeking to avoid electric grid interconnection constraints by shifting to an on-site gas baseload strategy are now encountering and exacerbating gas grid interconnection constraints.
Fuel Price Volatility
- Natural gas prices are inherently volatile and subject to geopolitical disruption. Recent Middle East escalations have amplified commodity price swings, exposing gas turbine operators to fuel input cost risks that directly impact operating margins. BTM gas contracts are increasingly structured as tolling arrangements, where the data center off-taker is required to assume commodity fuel price risk, while the developer is responsible for delivering the turnkey asset with an availability guarantee. In contrast, on-site solar and storage provides pricing locked through fixed long-term PPA structures, without any commodity price exposure (fuel cost is zero). Compounding the fuel price risk, US natural gas production growth in the three largest shale regions, responsible for three-quarters of all US shale gas production, has slowed considerably. With infrastructure assets designed to operate for 20+ years, fuel input cost risk is a structural vulnerability.
Community Opposition & Permitting Risk
- On-site gas generation produces local air emissions — NOx, CO₂, particulate matter — that trigger air quality permitting requirements and community opposition. Multiple data center projects have already faced lawsuits and backlash over air pollution from on-site gas turbines, with communities organizing against developments. xAI's Colossus facility in Memphis deployed as many as 35 natural-gas turbines to power its AI training operations, drawing lawsuits from environmental groups and local community opposition over air quality impacts (The Atlantic, March 2026). As environmental scrutiny of data centers intensifies, gas-heavy power strategies create a permitting and public relations liability that compounds over time.
Gas System Cost & Reliability Externality Risk
- Behind-the-meter gas generation does not eliminate grid dependence, it simply transfers it from the electrical grid to the natural gas grid. Natural gas generates approximately 40% of US electricity, meaning electricity prices are inherently tied closely to natural gas prices. As data center gas consumption scales, behind-the-meter operations will drive up natural gas and electricity prices for all ratepayers, resulting in the same cost-shift backlash dynamic on the gas grid that regulators are already fighting on the electric grid. Many industries (such as petrochemical, manufacturing, industrial heating) that are structurally dependent on natural gas and cannot turn to electricity or on-site renewables as data centers can will face direct cost and supply competition from data centers. On-site gas baseload overreliance simply transfers risk instead of mitigating it.
- Further, with turbine order books full through 2028+, every unit that goes behind the fence is one fewer available for the grid, meaning data centers are not only competing for fuel supply but actively absorbing the dispatchable generation capacity that utilities need to maintain grid reliability.
- Natural gas supply is also weather-dependent: cold winters, as Texas experienced in 2021, can freeze wellheads and crimp supplies dramatically. During Winter Storm Fern in February 2026, gas accounted for 84% of thermal generation outages in ERCOT (850+GWh of total outages), while PJM lost 22GW (16% of committed capacity). This dynamic will force suppliers to choose between keeping data centers running and letting people heat their homes.
- The correlated-failure profile is itself a grid risk. In Northern Virginia alone, a grid event that simultaneously forces 50 to 60 data centers to transfer to backup power would remove the equivalent of two to three nuclear plants' worth of load in seconds — a coordinated demand shock no grid operator has modeled or hedged. On-site gas does not eliminate this risk; it just transfers it. Every behind-the-meter site that loses primary power drops onto the grid simultaneously, and correlated weather, gas supply, or pipeline disruption events cluster that risk further.
The operational and financial risks of gas overreliance outlined above do not exist in isolation — they feed directly into a broader political and community backlash that is reshaping the terms on which data centers can develop, regardless of power source.
Energy affordability has become a forefront political priority across the United States as utility bills and electricity costs continue to rise (+30/40% on average since 2020 according to US EIA and Bureau of Labor Statistics).
Data centers are increasingly at the center of energy affordability discussions, as the scale of AI-driven load growth has drawn direct political attention. In November 2025, New Jersey and Virginia elected new governors who campaigned explicitly on reining in energy costs, with the Virginia governor-elect directly promising to make data center operators pay for their share of grid upgrades and take immediate action on energy affordability upon taking office. This is not a fringe issue. The political backlash around energy affordability is now widespread across state, RTO/ISO, and federal levels.
Electricity affordability is emerging as a defining factor in the 2026 US election cycle, where utility bills are expected to impact 36 upcoming governor's races. Governors are already acting beyond campaign rhetoric: New Jersey's newly elected Governor Sherrill made utility business model reform her first executive order; Indiana has passed legislation tying utility compensation to performance rather than spend. Developers who cannot demonstrate utility rate neutrality at minimum will face political headwinds in battleground states. As of April 2026, an estimated $64B in data center projects have already been canceled due to local community opposition — a remarkably bipartisan pattern spanning red and blue districts alike. Community opposition is now the single largest driver of project value destruction for the data center industry.
Community opposition as a structural industry risk
A meaningful share of community backlash traces to capital structure. Substantial data center land parcels have been assembled by short-hold investment vehicles structured to exit on three-to-five-year time horizons, well before long-term community relationships can mature. The result is a sector-wide externality: community opposition now attaches to the entire “data center” industry rather than the conduct of any specific operator. Long-hold sponsors inherit the political friction generated by short-hold capital, and community engagement has become a shared industry cost that shows up directly in project cap rates, permitting timelines, and interconnection queues. The commercial implication is straightforward, operators who build durable community relationships convert a sector-wide externality into a site-specific advantage.
The same capital dynamic that strains community relations also distorts grid planning. The speculative pipeline cited earlier — 145GW of announced projects held by sponsors with no operational track record — has clogged interconnection queues with filings that may never energize. Grid operators now face a signal-to-noise problem: forecasting which projects will actually deliver has become nearly impossible, and transmission and generation planning is delayed for real capacity alongside the vaporware. Credible operators pay the price — not for their own execution, but because they share a queue and a reference class with projects that will never break ground. Effective policy — sponsor qualification standards, interconnection queue screening, and project-readiness filters — must separate the two, so that committed capacity is not taxed by friction generated elsewhere.
PJM at the forefront of data center energy affordability backlash
Community opposition symptoms are visible at every level of the system — most acutely in PJM. The PJM regional electricity market, historically the heart of US data center development anchored by Northern Virginia, is now at the forefront of rising public and political energy affordability backlash as it struggles with a worsening supply-demand imbalance. PJM's independent market monitor quantified $23.1B in data center-related cost increases across the last three capacity auctions, with wholesale power costs reaching $67B in 2025 — up 54% year-over-year — driven primarily by a 262% surge in capacity costs. In December 2025, PJM's capacity auction cleared at a record $121/kW-year, driving upward pressure on customer electricity bills by as much as 20%. Despite record capacity prices (against a temporary price cap), PJM failed to procure enough resources to meet its 20% reserve margin for 2027–2028 period, threatening grid reliability on top of higher electricity bills.
Data center load growth has been cited as the primary driver of these conditions. PJM expects a 32GW increase to peak demand from 2024 to 2030, with 30GW attributed to data centers. At the individual utility level, projected data center loads are set to dominate system planning. Within PJM, Dominion Energy's data center load is expected to grow from roughly ~3GW today to ~18GW by 2035, AEP from ~2GW to ~10GW, and ComEd from near zero to 6GW (Halcyon). This imbalance is further exacerbated by PJM's inability to unclog its new generation queue (100GW+ backlog of new solar, wind, batteries).
In response, the White House and all 13 PJM state governors issued a bipartisan “Statement of Principles” calling for an emergency PJM backstop capacity auction to address electricity affordability and reliability. Critically, PJM's response included a provision that new large loads that do not supply their own generation will face mandatory curtailment during peak demand periods. The signal to the data center industry is clear, self-supply power or expect to be curtailed.
Within PJM, the PA House of Representatives recently passed HB 1834 (March 2026) as “a data center ratepayer protection bill” directing the state PUC to develop regulations barring electric companies from passing data center infrastructure and energy costs onto ratepayers — including regional transmission, network upgrades, grid reliability, and PJM emergency capacity procurement costs. Opposition to the bill notably centered not on ratepayer protection but on the bill's failure to require data centers to build new generation in-state, signaling that additionality requirements may be added in future legislative iterations. Separately, PA Gov. Shapiro announced GRID (Governor's Responsible Infrastructure Development) standards as part of his 2026/2027 budget proposal — requiring data center developers to fund their own power, hire locally, maintain transparency with communities, and adhere to high water-conservation standards. With GRID, PA will withhold support and expedited interconnection from projects that do not commit to these principles.
The political risk is escalating well beyond PJM
If AI data centers continue to be perceived as contributors to rising electricity costs and worsening grid reliability, especially during extreme weather events when grid stress is most visible to the public, the industry risks becoming a political scapegoat. In its latest Long-Term Reliability Assessment, North American Electric Reliability Corporation (NERC) found that 13 of 23 US/Canada grid regions face higher blackout risks over the next decade, with 5 at high risk by 2029, as demand from new data centers and other large loads is expected to outpace energy supply growth.
The risk extends beyond actual contribution to rate increases. Data centers are increasingly being blamed as a scapegoat for utility rate increases at large, even when increases in recent years have been driven primarily by aging grid infrastructure investments, to which AI load has contributed only marginally thus far. Without a proactive energy and policy engagement plan, data centers will inherit this blame and the policy consequences that follow, regardless of whether the attribution is fair.
The claim that data center loads inevitably result in higher electricity rates misreads utility economics. Rates are determined by total utility capital spend over total demand. New large loads that spread existing fixed costs across more kilowatt-hours (increasing the denominator relative to the numerator) drive utility rate suppression. In capacity-constrained markets like PJM, the co-development model addresses both sides: new generation expands self-supply and reliability, while flexible load expands grid demand — enabling higher grid load consumption without requiring proportionate increases to utility infrastructure upgrade spend.
The backlash is bipartisan and accelerating across every level of government. At the federal level, President Trump announced a formal “Ratepayer Protection Pledge” on March 4, 2026 (signed by Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI) requiring tech companies to build, bring, or buy their own power supply for new AI data centers. Data center electricity costs have emerged as a midterm vulnerability for Republicans after the issue roiled elections in Virginia, Georgia, and New Jersey. Sens. Hawley (R-Mo.) and Blumenthal (D-Conn.) have introduced bipartisan legislation requiring each new data center to have its own power supply independent of the public grid. Rep. Ocasio-Cortez and Sen. Sanders have proposed a nationwide moratorium on data center construction until the industry can make stronger assurances to protect the environment and utility customers.
As of April 2026, at least 11 states have introduced legislation to temporarily ban or restrict data center development, including Georgia, Maryland, Michigan, New Hampshire, New York, Oklahoma, South Carolina, South Dakota, Vermont, Virginia, and Wisconsin. Maine is the furthest along, advancing legislation (with bipartisan support and endorsement by Governor Mills) for the first state data center ban through November 2027 for facilities over 20MW, after electricity costs spiked nearly 60% between 2021 and 2026. Illinois Gov. Pritzker announced a two-year pause on data center tax incentives, with similar rollbacks proposed in Maryland, Michigan, Oklahoma, and Virginia. In Florida, Gov. DeSantis announced an AI “bill of rights” giving communities the right to limit data center construction. In Arizona, Gov. Hobbs has supported pulling the industry's tax incentives.
At the local level, moratoriums are also proliferating. New Orleans passed a one-year ban on new data center construction, Madison, Wisconsin enacted a similar ban after protests, and communities across Georgia, Michigan, and other construction hot spots have enacted local restrictions. In Q2 2025 alone, 20 projects representing $96B in investment were blocked or delayed amid local opposition (Data Center Watch), and ten new moratorium proposals have been filed since (Sightline Climate).
Beyond energy, broader socio-political headwinds are building. An emerging de-growth sentiment against AI — driven by societal anxiety over job displacement, privacy, and concentrated corporate power — is spurring an increasingly populist backdrop. Political backlash is the biggest risk to data center development.
Permitting delays cost approximately $10,000 per megawatt per day in lost revenue — a 100MW data center project loses $1M daily, with a 30-day slip in timeline destroying $30M permanently. Community acceptance is among the most significant risks, but it is also a controllable variable in project economics.
In response, AI leaders including Microsoft, Google, Meta, OpenAI, and Anthropic have been publicly committing to pay for their own grid infrastructure and energy costs, to bring net-new generation, and to establish demand response programs. The direction is clear: the industry recognizes it must proactively address affordability, reliability, and community impact. The Playbook provides the framework for how.
The political and regulatory backlash described above is not abstract. It is actively being codified into specific utility tariff structures and interconnection requirements that will directly impact data center economics and development timelines. These standards are being formed right now, across every level: individual utilities, state public utility commissions, RTO/ISOs, and federal regulators. From an energy perspective, regulators and grid operators are most concerned with 1) avoiding cost impacts on general ratepayers; 2) maintaining grid reliability; 3) preventing environmental impacts.
AI infrastructure companies that proactively position to meet these requirements — by bringing their own generation and embedding load flexibility — will be at a structural advantage. Those that wait will negotiate from a position of weakness against increasingly demanding terms.
Three primary categories of “sticks” are being implemented to address these concerns through large-load tariffs:
Take-or-Pay Minimums
Utilities are requiring data center customers to commit to paying for a minimum percentage (typically 75–85%) of their requested load, regardless of actual usage. Contract terms have extended to 10–20 years to align with transmission and capacity planning, materially de-risking utility capital deployment, with collateralized exit fees and early termination penalties. Higher minimum demand charges enhance cost recoverability, while elevated collateral requirements (as high as $1–1.5M per MW) serve as effective filters, to deter speculative demand and improve pipeline quality. The intent is to prevent cost shifting: if a data center requests grid capacity, it pays for that capacity whether or not it uses it, preventing stranded costs from falling on other ratepayers. Multiple utilities including AEP, Duke Energy, and WEC are implementing or proposing these structures.
Additionality Requirements
Regulators and grid operators are increasingly requiring that new large loads demonstrate incremental new generation supply, not simply tap into existing grid capacity that already serves other customers. The FERC PJM Co-Location Case is an early landmark: FERC's continued rejection of Amazon's proposed co-location with Talen Energy's Susquehanna nuclear plant was driven fundamentally by the lack of additionality — contracting with existing generation already serving the grid, without providing new supply, is not politically or regulatorily palatable. Grid system operators are signaling they will only provide interconnection alongside demonstrated additionality.
Curtailment Mandates
Increasingly required as a condition for interconnection, particularly to protect peak grid reliability. Large loads must agree to curtail grid consumption during peak demand periods on notice. Texas SB 6, already passed and being implemented, introduces a mandatory 24-hour curtailment requirement for large load customers, requiring up to 50% of grid consumption to be curtailed on 24-hour notice. The specific parameters — 1-hour, 2-hour, 4-hour curtailment windows — are still being determined, but the direction is set. More broadly, SB6 codifies flexibility as a gateway condition for any large load energy campus in Texas — operators must demonstrate the ability to pull from both grid and on-site generation, making flexibility a regulatory prerequisite for the co-development pathway, not just an operational optimization.
The pace of tariff adoption is accelerating. As of early 2026, more than 65 large-load tariffs have been proposed or approved in over 30 states (SEPA/NC Clean Energy Technology Center). The trend began in 2024, led by early movers Ohio and Indiana, expanded to Kansas, Michigan, and Virginia, and is now being debated in Illinois and Wisconsin. Some utility regulators have included requirements for data centers to “bring their own” generation or grid capacity.
Three tariff models have emerged: 1) published tariffs (transparent and replicable, but limited flexibility); 2) framework tariffs (standardized rules with negotiated economics, increasingly attractive for clean-energy pairing); 3) and bespoke contracts (customizable but opaque and non-scalable). States and utilities are each calibrating their own approach, but take-or-pay, additionality, and curtailment remain the consistent building blocks.
The ratepayer protection trend is accelerating across states. Ohio's PUCO rejected attempts to overturn a protective large-load tariff. Virginia's SCC imposed 14-year commitments and 85% minimum demand payments on data centers. Pennsylvania's PUC split 3-2 on a model tariff. Minnesota passed a law requiring data centers to cover full cost of service with zero stranded asset risk to ratepayers. Georgia voted to freeze rates through 2028 pending resolution. The ratepayer protection signal is clear.
This is a rapidly evolving landscape that the Playbook will track closely.
The Opportunity
What to Do Instead
“Clean flexible power co-development is the fastest, most de-risked, commercially superior path to powering AI at scale.”
The same dynamics that make energy the AI data center industry's biggest risk also make it the biggest opportunity for the companies that proactively lead.
Power has become an executive and board-level constraint for AI and tech leaders, presenting a generational opportunity to not just achieve energy velocity but also to leverage AI-driven load growth as a catalyst to build a better power system for all.
The industry can course-correct to make data centers into the most sought-after grid assets, as early movers are already demonstrating, by bringing their own clean generation, embedding load flexibility, strengthening grid reliability, and suppressing costs for all ratepayers. The policy signals around the opportunity are clear: FERC, PJM, SPP, ERCOT, DOE, etc. are all moving to prioritize large loads that bring both additionality and flexibility. AI data center developments that position for this now will earn accelerated interconnection, favorable utility tariff structures, and the regulatory goodwill to scale at speed.
Clean flexible power co-development is both the risk mitigation strategy and the fastest commercial pathway to scaling firm compute, turning energy from the industry's biggest constraint to its biggest opportunity. Through active co-development of energy alongside compute infrastructure, data center operators can deploy initial power in months, scale to hundreds of MWs of on-site clean generation and energy storage within 12 months from NTP, while pursuing complementary accelerated flexibility-embedded grid interconnection in parallel.
The result: 500MW+ of firm compute powered within 12–18 months per site through a repeatable framework, commercially structured to minimize upfront CapEx, at competitive costs. This is the fastest, most de-risked, commercially superior path to powering AI at scale.
500MW+
Firm compute per site in 12–18 months
Through the Playbook framework
3–5 yrs
Faster grid interconnection with BYOP + flexibility
vs. standard queue timelines
76GW
New load absorbable on existing US grid
(Tyler Norris, Duke)
0.25%
Annual load flexibility to unlock 76GW of absorbable grid capacity
Tyler Norris, Duke
Clean flexible power co-development outperforms the data center industry's current defaults across every strategic priority driving energy velocity to deliver faster, more reliable, more scalable power with lower total cost of compute and mitigated policy and execution risk.
The debate between off-grid gas islands and full grid dependence presents a false binary. The Playbook represents a third path: on-site clean generation and storage behind the meter, paired with flexible grid interconnection. This framework captures the speed advantage of self-supply, the reliability backstop of the grid, the community acceptance of clean power, and the economic upside of bidirectional grid participation. This creates multiple independent pathways to power, staged in phases, with each phase de-risking the next.
Beyond new build, flexibility unlocks enormous stranded capacity already in the system on both sides of the meter. Most data centers today operate at just 20 to 30 percent of their available power capacity, overbuilt for worst-case scenarios that rarely materialize. The grid itself mirrors this pattern. NVIDIA CEO Jensen Huang recently noted on the Lex Fridman podcast (April 2026) that the US grid runs at roughly 60% of peak capacity 99% of the time. The real bottleneck to AI power, Huang argues, is not generation, not permitting, not interconnection queues — it is the five- and six-nines uptime standard that forces utilities to reserve capacity above their system maximum rather than using available excess. Data centers designed for graceful degradation — shifting workloads, accepting marginally longer latency during peak hours — paired with utilities offering tiered power products instead of one-size-fits-all contracts, could unlock capacity at a scale that dwarfs any single generation buildout. NVIDIA's DSX Flex platform targets 100GW of unlockable capacity on the existing grid; Tyler Norris's Duke University research corroborates the order of magnitude. The opportunity is not just in what gets built next, but also in what has already been built.
That said, although flexibility is essential, it is not sufficient alone. Flexibility buys time and unlocks faster interconnection, but it does not substitute for new generation or transmission build. This is why the Playbook's architecture is integrated rather than single-lever — each element (on-site clean generation, storage, flexible grid interconnection, and grid-enhancing transmission investment) is complementary — each compensating for the limits of the others.
The Playbook's blended architecture unlocks more total MWs faster. Flexible grid interconnection complemented by on-site clean generation allows operators to scale power from both sources simultaneously rather than waiting on either alone. Because co-developed sites bring new clean capacity to the grid alongside new load, they improve broader system reliability and affordability for all ratepayers, turning data centers from grid burdens into grid assets.
Speed to power, decarbonization, and ratepayer affordability are not tradeoffs, but rather simultaneously achievable objectives when developers co-develop on-site clean energy campuses alongside load and flexibly optimize for existing grid capacity in parallel.
Speed (Time to Power):
The Playbook critically enables compute to be brought online, generating revenue in months rather than sitting idle waiting years for grid capacity. In contrast, the old playbook and current industry defaults deliver power in years. Grid-first interconnection timelines stretch 4–5 years at median and continue to increase, from under 2 years pre-2008. Even the fastest regions (Texas) now require 1–2 years. The Playbook delivers power in staged phases: initial turnkey modular firm-power units (fuel cells, microgrids) within months of NTP, large-scale on-site solar + storage + complementary dispatchable power within 12 months, and accelerated grid interconnection pursued in parallel — not as a prerequisite but as a complement.
Notably, the supply side is not the bottleneck. A single capacity RFP for 400 MW of accredited clean capacity in SPP drew 1.75GW of offers from over 600 developers within one month. Clean generation is ready to deploy — the constraint is a procurement and interconnection framework fast enough to match it to demand. The national supply picture tells the same story. Solar leads all new US generation additions, with ~87GW expected online in 2026 alone. Renewables are the fastest electrons to the ground; gas turbines, by contrast, now require multi-year lead times with deposits locked in years before the first megawatt. Clean capacity is shovel-ready at a scale the current interconnection framework has yet to absorb.
Nor is the existing grid as constrained as it appears — if load flexibility is on the table. In what has become the cornerstone study on data center load flexibility, Tyler Norris — then at Duke's Nicholas Institute, now Head of Market Innovation at Google — found that ~76GW of new large loads could be absorbed on the existing US grid if they curtailed just ~0.25% of annual load hours, roughly 22 hours per year. CPower's operating data across PJM, ERCOT, SPP, and MISO corroborates the same order of magnitude: data center VPP participation concentrates load reductions in the 16–40 tightest peak hours annually. For sites with on-site power resources, those peak hours require no compute trade-off at all — flexibility is simply delivered by shifting to self-consumption. At most, the trade is a rounding error on annual compute output in exchange for years of avoided transmission build and meaningfully faster time-to-power.
Site-level analysis corroborates the national picture. A Duke Energy study found 100GW of generation headroom available on the US grid, and analysis of six real-world 500MW sites showed that just 1 to 2 percent of annual power delivered as flexibility was sufficient to unlock capacity at three of four transmission-constrained locations. The cost of flexibility is trivially small relative to the capacity it frees.
And even the existing transmission system itself is under-utilized. As Jigar Shah — former head of the DOE Loan Programs Office — frames it, the US is not short generation — it is short wires. An estimated 100,000 MW of existing grid capacity sits trapped behind transmission constraints that grid-enhancing technologies (GETs), dynamic line ratings, and better operational data could unlock — without building a single new line or power plant. This is capacity that can come online in months through software and operational reform, not the decade-plus required for new transmission build. Supply, grid capacity, and transmission itself — at every layer, the constraints are procedural, not physical. The Playbook builds from first principles for the system as it actually is — not the one conventional wisdom assumes.
Reliability (Continuity of Power)
The old playbook is grid-dependent with on-site diesel backup designed for limited emergency use, not 24/7 firm capacity. Aging grid infrastructure and increasing extreme weather events make grid-dependent power an increasingly unreliable single point of failure. The Playbook builds a blended on-site power portfolio — fuel cells, microgrid systems, solar, energy storage, complementary dispatchable generation — with flexible modular power conversion (SSTs) removing single points of power failure. Higher reliability of power means higher effective uptime, less stranded compute, and lower stranded revenue.
Energy storage is central to this architecture. AI compute loads create millisecond-scale transients that no thermal generator can follow — storage is the only technology fast enough to buffer these oscillations, sitting between power sources and the load to orchestrate grid interconnection, on-site generation, and compute demand simultaneously. Without it, no co-developed campus can safely operate at AI-grade demand profiles. xAI's Colossus facility in Memphis deploys Tesla Megapacks specifically to buffer training-load volatility, validating battery storage as a core architectural component at gigawatt scale, not a theoretical fix.
Critically, this reliability logic cuts both ways: if grid-only is a single point of failure, so is off-grid-only. Even hyperscalers building dedicated on-site generation acknowledge that grid interconnection remains essential — not just for economics, but for the reliability redundancy that no single-source behind-the-meter configuration can match. Industry consensus is converging around expectations that 95% or more of data center capacity will require grid interconnection.
Capital markets have quietly reached the same verdict. Major financing firms have signaled they will not fund off-grid data center projects beyond two-year horizons — and notably, they are not drawing the line at fuel. Islanded power is treated as a short-term bridge rather than durable long-term infrastructure, regardless of whether the generation is gas, solar, or hybrid. The capital stack's verdict applies to off-grid architectures broadly — gas baseload is simply its most politically and economically exposed version today.
Economics (Total Cost of Compute).
The Playbook's power approach may cost more per megawatt-hour than a simple grid connection. But per-megawatt-hour cost is the wrong way to evaluate power for AI data centers. What matters is the total cost of delivering compute output — including how fast you can start generating revenue, how often your GPUs are actually running, and how much capital you risk on infrastructure that may never arrive. The industry shorthand: LCOE (levelized cost of energy) favors the old playbook; TCOC (Total Cost of Compute) favors the new one.
On paper, the Playbook's blended on-site power portfolio runs ~$80–100/MWh compared to ~$60–80/MWh for grid power + PPA procurement. In practice, the Playbook drives lower total cost through:
- Faster time-to-revenue: GPU utilization starts months earlier, generating revenue while competitors wait for grid
- Higher effective uptime: diversified power portfolio reduces downtime risk
- Avoided stranded costs: no multi-year capital commitments to grid interconnection that may slip or fail
- Orchestrated efficiency: AI-driven controls maximize compute output per watt of constrained power input
The commercial structures that underpin clean flexible power co-development further de-risk economics. Developers have validated delivery of solar+storage through ~15-year PPA structures at ~$100/MWh with 90% firm power guarantee, meaning zero upfront CapEx to the data center operator for a meaningful portion of the power portfolio. The energy developer partners finance, build, own, and operate the on-site power infrastructure. The data center operator pays for power delivered, rather than infrastructure built.
Financing follows the same logic as development, advanced in stages. No single capital check underwrites a multi-GW campus — projects advance in tranches backed by creditworthy offtakers and long-term PPAs. Fifteen-year terms under tariff or energy service agreements are the baseline; longer contracted cash flows unlock better financing terms, aligning developer and operator incentives around duration beyond price.
Policy & Public Perception
Under the old playbook, data centers are perceived as grid burdens, subject to regulatory “sticks” including take-or-pay tariffs, additionality mandates, curtailment requirements, and worst case, interconnection moratoria and de-growth legislation. The Playbook shifts the narrative. Data centers that bring their own generation and embed load flexibility are positioned as grid assets, earning regulatory “carrots” including accelerated “fast-track” grid interconnection (as much as 3–5 years faster) and favorable utility tariff treatment by delivering rate suppression as a public benefit for all ratepayers. This converts policy risk into a competitive advantage.
The same logic applies to community acceptance. Data centers paired with on-site clean generation are consistently better received than standalone facilities, converting a permitting liability into a local economic asset. As Christian Belady, one of the pioneers of hyperscale data center development, put it: “All stakeholders should thrive — you can't have one-dimensional thriving” in regard to how data centers can earn community acceptance and “the license to scale”. The permitting delay math makes this commercially rational - the cost of skipping community engagement dwarfs the investment required to earn it.
A community-benefit permitting framework is emerging: developers who commit to a defined checklist — local hiring, water replenishment, grid flexibility sharing, environmental mitigation — can target six-month permitting timelines versus eighteen months through traditional channels. The framework is being developed by a coalition of large campus developers and represents a 3x acceleration for projects that invest in social license upfront.
Beyond power, water is commonly a flashpoint for community opposition. DigitalBridge's water replenishment model illustrates what this looks like in practice: funding agricultural irrigation upgrades — converting flood to drip systems — that generate two times the water credits consumed by the data center, making the facility a ‘water-positive’ net water contributor to the community.
Scalability
The old playbook scaled through grid interconnection capacity upgrades, meaning each increment of new capacity requires a new interconnection study, new approvals, and new grid infrastructure builds, each gated by multi-year timelines. The Playbook scales modularly through on-site power expansion complemented by grid capacity over time, with the flexibility to integrate additional power resources and emerging technologies (SMRs, geothermal, advanced storage) as they become commercially available and viable — all without needing to re-platform the power supply architecture.
The workload reality demands this modularity: even inference-only deployments now require hundreds of megawatts at minimum, while training clusters operate at gigawatt scale. A well-designed co-development site can scale generation and load together across this full range without redesigning the interconnection. Storage is scaling to meet this architectural requirement. Bloomberg NEF projects US energy storage capacity rising from ~40GW today to ~240GW by 2035 — a 6x expansion driven substantially by data center co-development and grid flexibility requirements.
Execution Risk
The old playbook is binary go/no-go: a single interconnection approval or equipment delivery determines whether the entire project proceeds. One delay stops everything.
The Playbook mitigates this through staged, phased execution with multiple parallel pathways. If grid interconnection slips, on-site generation continues delivering power. If one equipment vendor delays, modular procurement across multiple suppliers limits exposure. If a specific technology underperforms, the blended portfolio absorbs it without full-site impact. Revenue generation begins in Phase 2 (months from NTP) rather than depending on completion of the entire power stack. Each phase de-risks the next — and critically, no single delay creates a total project stoppage.
The supply chain diversity reinforces this. Solar panels, battery storage, fuel cells, and grid-enhancing technologies each draw from independent manufacturing bases and procurement channels. The concentrated equipment risk that defines gas turbine procurement (three OEMs, six-year backlogs, 10% upfront deposits) does not apply to a diversified portfolio.
“Old Playbook” vs. “New Playbook” Comparison
Access
Old Playbook
Single path to power — fully dependent on utility
New Playbook
Multiple paths: on-site self-supply + grid
Speed
Old Playbook
Multi-year (median 4–5 yrs, fastest 1–2 yrs)
New Playbook
Staged phases: 3 months / 12 months / 12–18 months
Reliability
Old Playbook
Grid-dependent + backup power
New Playbook
Modular firm on-site power portfolio
Economics
Old Playbook
Lower LCOE (~$60–80/MWh), higher TCOC
New Playbook
Higher blended LCOE (~$80–100/MWh), lower TCOC
Policy
Old Playbook
Grid burden — public perception & policy risk
New Playbook
Grid asset — driving utility rate suppression
Scalability
Old Playbook
Scaled solely through grid upgrades
New Playbook
Modular on-site expansion + grid in parallel
Execution
Old Playbook
Binary go/no-go risk on single interconnection
New Playbook
Staged/phased portfolio risk management
| Priority | Old Playbook | New Playbook |
|---|---|---|
| Access | Single path to power — fully dependent on utility | Multiple paths: on-site self-supply + grid |
| Speed | Multi-year (median 4–5 yrs, fastest 1–2 yrs) | Staged phases: 3 months / 12 months / 12–18 months |
| Reliability | Grid-dependent + backup power | Modular firm on-site power portfolio |
| Economics | Lower LCOE (~$60–80/MWh), higher TCOC | Higher blended LCOE (~$80–100/MWh), lower TCOC |
| Policy | Grid burden — public perception & policy risk | Grid asset — driving utility rate suppression |
| Scalability | Scaled solely through grid upgrades | Modular on-site expansion + grid in parallel |
| Execution | Binary go/no-go risk on single interconnection | Staged/phased portfolio risk management |
The policy landscape is rapidly developing in a direction that directly validates the clean flexible power co-development opportunity. Across federal regulators, regional grid operators, and state legislatures, the signal is consistent: large loads that bring additionality and embed flexibility will receive preferential treatment.
The posture has shifted materially in the past twelve months. Data center operators are increasingly pursuing on-site additionality and load flexibility, driven by a straightforward recognition: this unlocks faster power access through both self-supply and grid interconnection, which maximizes compute output — and revenue — through faster energization.
Federal & RTO-Level Reform
FERC has ordered PJM to create large load co-location rules with tiered service — firm and curtailable — for expedited interconnection, with BYOP data centers expected to achieve 3–5 years faster timelines. PJM governors from PA, MD, NJ, and VA have jointly backed interconnection priority for self-supplying loads. SPP's FERC-approved HILL program enables interconnection timelines as short as 150 days through paired generation-load studies, and SPP's new Consolidated Planning Process merges transmission planning and generator interconnection into a single framework — cutting study timelines from 400+ days to roughly seven months. ERCOT's large-load queue has tripled to ~226GW (73% data centers), forcing a shift to batch processing that underscores even the fastest US market is hitting capacity constraints. DOE has issued a notice of proposed rulemaking on load flexibility with an April 30, 2026 deadline.
State-Level Action
22 states have already enacted or are actively developing legislation for on-site energy campuses. West Virginia's HB 2014 goes furthest, creating microgrid districts exempt from PUC regulation, zoning, and permitting. Illinois' POWER Act offers fast-track interconnection in exchange for new clean capacity and grid upgrade payments. Oklahoma's SB 480 removes the regulated monopoly bottleneck for behind-the-meter generation. Utah and New Mexico are advancing similar frameworks. This rapid, bipartisan policy diffusion signals broad recognition of co-located generation as the path to both speed and local economic value.
Industry Alignment
The Data Center Coalition is broadly advocating for expedited interconnection for curtailable and BYOP-compliant loads. EPRI's Flex MOSAIC framework, endorsed by 65+ utilities, ISOs, regulators, and hyperscalers, standardizes how data centers describe flexibility — ramp-down speed, duration, and frequency — with an open letter signed by 30+ organizations including Google, Meta, Nvidia, Southern Company, Exelon, and Constellation calling for shared flexibility standards.
The direction across all levels — federal, regional, state, industry — is clear. The regulatory framework is being built around additionality and flexibility as the conditions for accelerated interconnection and favorable tariff treatment. The Data Center Power Playbook is built around the framework regulators are already rewarding.
These aren't hypotheticals. The core attributes of the clean flexible power framework — additionality, load flexibility, ratepayer protection, and accelerated interconnection — are already being demonstrated in signed deals and operational projects.
Additionality — New Clean Power Capacity
Google's partnership with DTE Energy in Michigan represents the most comprehensive example to date. In March 2026, Google confirmed a 1GW data center campus in Van Buren Township near Detroit, structured through DTE's Clean Transition Tariff. The agreement commits 2.7GW of new clean power resources to the local grid — 1.6GW solar, 400MW four-hour storage, 50MW long-duration storage, 300MW additional clean resources, and 350MW demand response. Google covers all electricity costs and infrastructure needs, with $1.7B in affordability benefits to ratepayers over the life of the contract. Critically, every megawatt is new build — directly filling the generation gap left by coal retirements rather than reshuffling existing clean energy claims. The deal touches every pillar of this framework simultaneously: additionality (2.7GW new clean generation), embedded load flexibility (350MW demand response), ratepayer protection (no cost pass-through plus active affordability investment), and accelerated grid access.
Load Flexibility — Data Centers as Grid Assets
Google's Flexible Interconnection Tariff (FIT) establishes the first tariff framework designed specifically for flexible large loads. The blueprint centers on a core trade: data centers receive a mix of firm and conditional grid service while bringing enough accredited capacity through on-site power, PPAs, or VPPs to cover incremental capacity added to the grid. In exchange, data centers receive accelerated grid interconnection — as much as 3–5 years faster than standard queue timelines.
This isn't theoretical. Google has integrated 1GW of demand response into utility contracts across the South and Midwest, scaling beyond initial I&M and TVA pilots to five to six data centers. The operational model routes compute to match available power in real time — a proof point that flexibility is a competitive advantage for interconnection speed, not a concession.
Google's conditions for participating in demand flexibility are instructive for program designers and regulators: adequate financial incentives, clear guardrails on disruption frequency and duration, and integration of flexible resources into utility planning processes. Programs that meet these three conditions will attract hyperscaler participation. Those that don't will be ignored regardless of regulatory mandate.
Ratepayer Protection — Growth Without Cost-Shifting
Anthropic's comprehensive energy policy, announced in February 2026, sets an explicit ratepayer protection commitment in the industry. The policy commits Anthropic to covering 100% of grid infrastructure upgrade costs needed to interconnect its data centers, bringing net-new power generation online to match data center electricity needs, and working with utilities and external experts to estimate and cover any demand-driven price effects on ratepayers where new generation is not yet online. Anthropic is also investing in curtailment systems to reduce data center power usage during peak demand periods. As Anthropic stated: “The US AI sector will need at least 50GW of capacity over the next several years... but AI companies shouldn't leave American ratepayers to pick up the tab.” This is the clearest signal yet that leading AI companies understand the ratepayer question is existential to their license to operate — and are willing to put capital behind the answer.
Accelerated Interconnection — Storage as a Grid On-Ramp
Aligned Data Centers' partnership with Calibrant Energy demonstrates how on-site storage can unlock grid access where transmission constraints would otherwise impose multi-year delays. Aligned funded a 31MW/62MWh on-site battery system developed by Calibrant at an undisclosed Pacific Northwest location, after the local utility required demonstrated capacity additionality as a condition for interconnection approval. The utility studied the grid impact and determined that without new on-site capacity, the interconnection would strain local infrastructure. This is a critical precedent: additionality requirements are not just regulatory theory — they are already being applied as binding conditions for interconnection. The model converts what would be a queue bottleneck into a solvable capital problem.
Across these examples, the common thread is clear: data centers that proactively bring additionality, embed load flexibility, and commit to ratepayer protection earn accelerated interconnection, favorable tariff treatment, and community support. The strategies described in this Playbook are already being executed by leading companies — what's missing is the shared framework to scale them.
The Framework
How to Execute
The Data Center Power Playbook (DCPP) provides a replicable four-phase execution template for powering AI compute at scale through clean flexible power co-development.
Four-Phase Execution Template
Power-First Site Selection
Site selection process led by power potential. Evaluate existing transmission capacity and interconnection infrastructure, available excess generation, and land for on-site energy infrastructure development. Pre-file for grid interconnection and engage partners early to coordinate accelerated interconnection through flexibility and additionality.
→Initial Power Deployment
Deploy turnkey modular firm-power units, such as fuel cells and microgrid systems, to deliver initial MW capacity within months. Enable near-term compute revenue ahead of larger infrastructure development.
→Modular BYOP Build-Out
Develop optimal on-site power portfolio: solar, energy storage, and complementary dispatchable generation scaling to hundreds of MWs within a year. Energy storage supports energy shifting and load curtailment. Flexible power electronics support modular scaling of power sources over time.
→Grid Integration & Flexibility
Finalize accelerated grid interconnection, with on-site power enabling both additionality and load flexibility. Proactively fund targeted grid upgrades to expand interconnection capacity. Implement AI-driven controls to maximize compute per watt by orchestrating power, cooling, and compute workloads.
Power-First Site Selection
Site selection process led by power potential. Evaluate existing transmission capacity and interconnection infrastructure, available excess generation, and land for on-site energy infrastructure development. Pre-file for grid interconnection and engage partners early to coordinate accelerated interconnection through flexibility and additionality.
Initial Power Deployment
Deploy turnkey modular firm-power units, such as fuel cells and microgrid systems, to deliver initial MW capacity within months. Enable near-term compute revenue ahead of larger infrastructure development.
Modular BYOP Build-Out
Develop optimal on-site power portfolio: solar, energy storage, and complementary dispatchable generation scaling to hundreds of MWs within a year. Energy storage supports energy shifting and load curtailment. Flexible power electronics support modular scaling of power sources over time.
Grid Integration & Flexibility
Finalize accelerated grid interconnection, with on-site power enabling both additionality and load flexibility. Proactively fund targeted grid upgrades to expand interconnection capacity. Implement AI-driven controls to maximize compute per watt by orchestrating power, cooling, and compute workloads.
Acknowledging that active clean flexible power co-development is more complex relative to passive utility grid procurement or even on-site gas baseload deployment, the DCPP will be executed through a coalition of best-in-class partners across the energy ecosystem across three interdependent pillars — on-site power, load flexibility, and grid interconnection — working in parallel across all phases to deliver power with unparalleled speed, reliability, and scalability.
The Playbook integrates on-site power, load flexibility, and accelerated grid power into a unified framework. Rather than being treated as procurement, power is treated as infrastructure co-development, pairing data center deployment with proactive energy infrastructure planning and execution. The solution is a dual on-site BYOP + accelerated grid interconnection strategy, prioritizing co-development of energy infrastructure (including co-located energy campuses) to achieve energy velocity and de-risk energy execution.
Four strategic development objectives drive the framework:
- Develop Large-Scale On-Site Power & Storage In Parallel: Power and compute scale together. Partner with energy developers to build large-scale portfolio of on-site generation and energy storage within 12 months, scaled modularly across phases near-term and over-time, while simultaneously pursuing grid interconnection.
- Design Power for Speed, Reliability, Scale, and Flexibility: Each power source serves a specific role in the timeline and grid coordination strategy. Ensure flexibility in design to accommodate modular deployment of power infrastructure to scale alongside compute infrastructure (SSTs), deploy turnkey firm-power modules (fuel cells, microgrid systems) to provide initial power for near-term uptime, establish EMS controls system to manage on-site energy assets with responsive capabilities to support load flexibility/curtailment signals from grid.
- Accelerate and Expand Grid Access Proactively: Use BYOP additionality and load flexibility to earn fast-track queue positioning. Assess grid capacity within site selection through power-first approach, proactively engage utility / grid partners early to embed load flexibility and fund targeted grid upgrades to earn “fast-track” and expanded grid interconnection.
- Maximize Data Center Operations Efficiency: Maximize revenue/compute output vs. constrained energy input. Orchestrate inside-the-fence controls of power, cooling, compute management, and other levers to maximize compute per watt.
The framework addresses both immediate power needs and long-term reliability/profit optimization by enabling data centers to self-supply while coordinating accelerated grid interconnection. This shifts power from a constraint and risk to a competitive advantage, delivering energy velocity to accelerate speed to power and revenue.
Phase 1 begins with initial data center development site selection. Traditional data center site selection prioritized fiber proximity and metro-adjacent colocation. AI shifts data center development to where power can scale. AI training workloads are inherently latency tolerant and the majority of AI inference workloads have demonstrated higher latency tolerance from users. As OpenAI CEO Sam Altman noted, users are “surprisingly willing to wait for a great answer.” Post-training and test-time compute require scale and concentration.
This drives the shift towards power-first site selection to rural, power-rich locations with superior land, power, and community development economics. Power-first site selection is driven by land abundance and access to transmission capacity, not fiber routing.
The Evaluation Framework
The Execution Approach
Phase 2 delivers the first critical milestone: firm MW capacity within 90 days of NTP. This proves “power in months not years” and de-risks the immediate revenue timeline for compute operations.
Install turnkey, modular, firm-power units (ex: fuel cells, microgrid systems, small turbines) to provide immediate power. Priority considerations include modularity, fuel efficiency, and avoid permitting friction (combustion-free power to avoid air pollution impact). Site readiness for equipment delivery is the critical path dependency for timeline.
Initial turnkey modular firm-power units cost >$100/MWh, higher than long-term on-site generation. But this is the premium paid for speed. The value is time-to-revenue. GPUs begin generating revenue within three months while larger scale on-site power development and grid interconnection advance in parallel.
This phase achieves multiple objectives:
These initial turnkey modular firm-power units remain operational as part of the site's power portfolio, providing continued firm capacity as Phase 3 and Phase 4 assets come online and the overall power mix scales.
Phase 3 deploys the optimal on-site power portfolio, solar, energy storage, dispatchable gas capacity, scaling to hundreds of MWs within the 12-month window, in parallel to Phase 2's initial power operation. The layered approach includes:
- Large-scale battery storage at the data center perimeter serves multiple functions: energy shifting of excess intermittent generation (solar production peaks during the day; compute demand may peak at night), dispatchable power for peak demand periods, and load curtailment support. Shifting power consumption to battery and on-site generation reduces peak grid demand and improves efficiency.
- Flexible power electronics (Solid State Transformers / SSTs) enable modular scaling of power infrastructure over time. As compute density increases or new workloads are added, power distribution systems adapt without major re-engineering. SSTs also accommodate integration of emerging power technologies as they become commercially available.
- Sophisticated Energy Management System (EMS) controls manage on-site generation, storage, and demand in real-time. The EMS receives signals from the grid operator (curtailment requests, frequency regulation needs) and responds automatically, adjusting on-site power dispatch, battery charging/discharging, and flexible load management.
- Rack-level supercapacitors inside-the-fence absorb the volatile load oscillations from GPU clusters. Compute loads swing from 10% to 100% utilization with variations at millisecond, second, and minute frequencies. Supercapacitors and battery storage provide frequency regulation, protecting both compute hardware and grid connection, and ensuring compute reliability even under extreme load volatility.
By the end of Phase 3, the data center operates with diverse, modular on-site power assets (including solar, storage, complementary gas-fired dispatchable capacity, and supercapacitor buffers) and the EMS control systems to optimize them. The site is substantially self-sufficient and positioned for grid integration.
Phase 4 transitions the site from self-powered operation to grid-integrated optimization. The data center becomes a dynamic grid asset, providing flexibility to the utility while further optimizing its own operations. The layered approach includes:
- Finalize fast-track grid interconnection with utility partners: On-site BYOP provides additionality (the grid doesn't pay for self-supply, so additional on-site generation doesn't displace grid responsibility). The system is designed to meet curtailable load flexibility requirements, leveraging on-site energy storage and other levers to reduce or shift grid load consumption in response to grid stress events. This additionality and flexibility earns accelerated interconnection.
- Fund targeted grid upgrades to expand capacity: Grid Enhancing Technologies (GETs) - such as Dynamic Line Rating (DLR), conductor coating, and reconductoring - can enhance existing transmission capacity and reduce power losses. Work with utilities to identify and implement necessary grid infrastructure upgrades that bottleneck interconnection.
- Implement AI-driven orchestration controls to manage inside-the-fence power, cooling, and compute systems simultaneously: The goal is to maximize compute per watt. Given constrained power input (defined by on-site generation + grid interconnection capacity), the data center should operate to produce maximum revenue and token output from every megawatt available. AI models predict workload patterns, optimize cooling efficiency in real-time, and distribute compute loads across hardware to minimize power waste.
- Full load flexibility and curtailment capability become operational: The data center integrates with grid interface platforms to respond to grid signals in real-time, in compliance with its flexible interconnection agreement. When the grid experiences stress (peak demand, generation shortfall, frequency drop), the data center reduces non-critical workloads or shifts consumption to on-site energy storage resources. This flexibility earns accelerated interconnection and preferential utility tariff agreements, creating a virtuous cycle: flexibility → faster interconnection → lower costs → higher margins.
By Phase 4 completion, the site is fully grid-integrated, operationally efficient, and financially optimized.
Energy co-development is more complex than traditional utility procurement. The Playbook is executed through partnerships with companies across the energy ecosystem, organized across three interdependent pillars to achieve energy velocity at scale.
The Sijbrandij Foundation's Data Center Energy Initiative serves as an agnostic coordination layer — convening partners across all three pillars, architecting the framework, and aligning around shared timelines and energy velocity objectives. The partner landscape has been mapped across each category below; coalition formalization is underway. Further details on our energy partner coalition will be shared near-term.
On-Site Power
Deploy rapidly across multiple phases, developed in parallel to compute, for greater speed and reliability of power
Load Flexibility
Enable data center to be a dynamic grid asset to earn faster grid interconnection and deliver rate suppression aligned with policy priorities
Grid Interconnection
Proactive utility planning starting with power-first site selection; focus on expanding grid capacity faster for accelerated access and scalability
| Pillar | Core Objective |
|---|---|
| On-Site Power | Deploy rapidly across multiple phases, developed in parallel to compute, for greater speed and reliability of power |
| Load Flexibility | Enable data center to be a dynamic grid asset to earn faster grid interconnection and deliver rate suppression aligned with policy priorities |
| Grid Interconnection | Proactive utility planning starting with power-first site selection; focus on expanding grid capacity faster for accelerated access and scalability |
Successful execution of the Playbook depends on coordinated partnerships across these three pillars, aligned around shared timelines and energy velocity objectives.
Pillar 1: On-Site Power Infrastructure Development
Develop the on-site energy campus in parallel to compute. The objective is to deliver reliable power faster and more flexibly, providing self-supply in complement to grid interconnection.
Energy Infrastructure Developers
- Strategic Purpose: Foundational partner and executor for the development of the large-scale on-site BYOP portfolio, managing the complex integration of solar, storage, complementary dispatchable power, and other resources behind-the-meter.
- Key Criteria: 1) Leading utility-scale power developer with demonstrated track record of developing GW-scale power generation; 2) On-site power integration expertise — capability to manage complex integration of large-scale power on-site for BYOP; 3) Financing capability, likely structured as longer-tenor PPA (15-years+) with embedded buy-out option.
Initial Turnkey Modular Firm-Power Equipment Providers
- Strategic Purpose: Deliver the first firm MW capacity within 90 days of NTP through modular, rapidly deployable power units (fuel cells, microgrid systems, small turbines) to enable immediate compute operations while permanent on-site generation develops in parallel.
- Key Criteria: 1) Speed of deployment — proven ability to deliver 50+MW within 90 days of site readiness; 2) Modularity and scalability — units that can be incrementally added as demand grows; 3) Equipment availability — manufacturing capacity available to deliver on orders; 4) Fuel efficiency and avoidance of permitting risks — combustion-free power to avoid local air pollution and air permit risks.
Power Electronics (SSTs)
- Strategic Purpose: Enable modular scaling of power infrastructure over time alongside compute, provide flexible improved power conversion and distribution system design (natively supporting 800V DC power distribution architecture), and accommodate integration of emerging power technologies as they become commercially available.
- Key Criteria: 1) Proven commercial readiness — demonstrated technical capability prior to deployment; 2) Equipment availability — manufacturing capacity available to deliver on orders.
Energy Storage
- Strategic Purpose: Provide energy shifting of excess intermittent generation, dispatchable power for peak demand, load curtailment support, and frequency regulation to protect compute hardware and grid connection. Large-scale battery storage at data center perimeter for energy shifting and peak demand; rack-level supercapacitors inside-the-fence as essential buffer to absorb highly volatile load oscillations from GPU clusters.
- Key Criteria: 1) Equipment availability — manufacturing capacity to deliver on orders at required scale; 2) Technology readiness — proven commercial deployment (not pilot/demonstration stage); 3) Integration readiness — commercially ready for behind-the-meter integration with EMS and on-site power portfolio.
Pillar 2: Load Flexibility Integration
Enable the data center to operate as a dynamic grid asset — incorporating sophisticated EMS controls to manage on-site energy assets with responsive capabilities to support load flexibility and curtailment signals from the grid. Orchestrate power consumption, cooling, and compute workloads to respond to grid conditions and maximize compute output relative to constrained power input. Flexibility provides policy currency to enable fast-track interconnection, tariff benefits, and rate suppression.
Energy Management Systems (EMS)
- Strategic Purpose: Central control layer for the on-site power portfolio — manages real-time orchestration of on-site generation, storage, and demand to serve data center load demand. Meet load flexibility and curtailment requirements by receiving dispatch signals from grid interface partners and determining how to meet curtailment across controllable on-site assets.
- Key Criteria: 1) Capability to manage portfolio of multiple on-site power resources simultaneously in real-time; 2) Integration with grid interface partners to receive and act on dispatch signals; 3) Integration with data center operational orchestration layer for coordinated power-compute optimization.
Grid Interface Layer
- Strategic Purpose: Interface between the grid operator/utility and the data center's EMS — receiving dispatch signals (curtailment requests, frequency regulation, demand response events) from grid operators and translating them for the EMS to execute against.
- Key Criteria: 1) Established relationships and integration with RTOs/ISOs and utility dispatch systems; 2) Real-time signal processing and communication capability; 3) Track record operating in demand response and curtailment markets.
Data Center Operational Orchestration
- Strategic Purpose: Maximize compute per watt by orchestrating inside-the-fence power, cooling, and compute systems simultaneously — AI-driven controls that predict workload patterns, optimize cooling efficiency in real-time, and distribute compute loads across hardware to minimize power waste.
- Key Criteria: 1) Data center specialized (vs. broader building management systems); 2) AI/ML-driven optimization of compute workload scheduling relative to power availability; 3) Ability to shift non-critical workloads or redistribute inference tasks to battery-backed resources during grid stress events; 4) Integration with EMS for coordinated power-compute optimization.
Pillar 3: Proactive Grid Planning & Interconnection Acceleration
Integrate power-first site selection with proactive utility engagement — embedding grid capacity analysis within initial site evaluation, proactively engaging utility and grid partners to embed load flexibility and fund targeted grid upgrades to earn “fast-track” or expanded grid interconnection.
Grid Analytics & Site Selection Software
- Strategic Purpose: Embed grid capacity analysis within the initial site selection process to optimize for higher starter capacity and faster interconnection timelines, driving site ranking based on transmission capacity, queue positioning, and interconnection potential.
- Key Criteria: 1) Specialized focus on grid capacity analysis for large load customers; 2) Comprehensive transmission and interconnection data across target geographies; 3) Modeling capability for queue positioning and interconnection timeline scenarios; 4) Integration with utility planning data.
Utility & Transmission Partners
- Strategic Purpose: Coordinate proactive grid interconnection planning starting before or immediately after land acquisition, communicate BYOP additionality and flexibility plans to position for fast-track queue placement, and identify and execute targeted grid infrastructure upgrades (Dynamic Line Rating, conductor coating, reconductoring) that bottleneck interconnection.
- Key Criteria: 1) Willingness to engage in proactive interconnection planning ahead of standard queue timelines; 2) Openness to co-funded grid upgrade investment as part of interconnection acceleration; 3) Alignment on large load tariff structures that reward additionality and flexibility.
Policy & Regulatory Partners
- Strategic Purpose: Collaborate at state and federal levels to establish and refine regulatory structures that provide the framework for data centers to deliver capacity additionality and load flexibility in exchange for accelerated interconnection and favorable tariff treatment.
- Key Criteria: 1) Demonstrated engagement with FERC, RTOs/ISOs, and state-level regulatory proceedings; 2) Understanding of evolving large load tariff landscape; 3) Ability to represent data center energy interests in policy development aligned with ratepayer protection and grid reliability.
Emerging Energy Technology Radar
The Playbook is designed for technology flexibility. The modular architecture supports the integration of new power sources and storage technologies as they reach commercial viability. The following technologies are on active radar, with limited commercial availability for data center scale deployment today due to location-specific constraints or technological readiness:
- Enhanced/Advanced Geothermal: Growing interest and improving technology. Limited deployed capacity at data center scale today, but potential for high-capacity, low-footprint baseload power at specific geothermal-rich site locations (geographically constrained).
- Small Modular Reactors (SMRs): Promise firm baseload power with minimal land footprint. Earliest commercial deployments expected early 2030s. Potential to provide 24/7 firm carbon-free power at data center scale without combustion or intermittency constraints.
- Advanced Energy Storage (iron-air, flow batteries, compressed air, gravity): Could complement lithium-ion for longer-duration energy shifting and seasonal storage. Commercial viability for data center scale expected within 2–5 years.
- Fusion: Long-term potential. No near-term commercial pathway for data center deployment.
The Playbook's modular architecture does not depend on any of these technologies reaching commercial readiness on a specific timeline. But it is designed to integrate them as they do — without re-platforming the power supply stack. This is deliberate: the framework that works with today's commercially available technologies should also work with tomorrow's.
Industry energy defaults are forming now. Policy frameworks are actively being codified. The companies and frameworks that define the standard now will determine how the next 50GW of US data center capacity gets powered.
Following extensive research, industry engagement, and direct conversations with operators, developers, regulators, and technologists across the AI data center energy ecosystem, the DCPP represents our best current understanding of how to solve this challenge. We are now moving to execution — convening the coalition to put this framework into practice.
The Sijbrandij Foundation's Data Center Energy Initiative was created to solve the data center energy challenge to accelerate the AI transition. We are building an agnostic operational platform to convene and coordinate a coalition of energy partners to execute clean flexible power co-development for the data center industry.
The DCPP is an open living framework. We encourage independent adoption, and will also facilitate select reference project deployments with partners to demonstrate the model in practice. The Playbook will be publicly iterated regularly as we receive new information including as reference project deployments deliver real-world data, as the regulatory and market landscape evolves, and as the partner coalition expands.
The coalition is organized around specific partner categories including:
- On-site power developers (solar, storage, complementary dispatchable generation)
- Initial turnkey modular firm-power equipment providers (fuel cells, microgrid systems)
- Flexible power electronics (solid-state transformers, power conversion systems)
- Energy management and controls systems (EMS, grid interface platforms)
- Grid analytics and interconnection software
The Foundation is facilitating select reference project deployments to validate the model in practice — directly coordinating energy partners with AI infrastructure partners to demonstrate the Playbook framework in the field. Partners who engage early help define the standard that the broader market will adopt.
We evaluate partner capabilities, track record, commercial readiness, and alignment with the Playbook framework. This is not a vendor directory — it is a coordinated coalition built around shared execution standards and timelines.
Next step: Email aric@sijbrandijfoundation.org with an overview of your solution (technology, deployment track record, commercial readiness, and relevant capabilities).
Engagement starts with a scoping call. We assess whether opportunities within your development pipeline fit the Playbook framework and determine where the Foundation's partner network adds the most value. The Foundation's role is agnostic — we are not selling a specific technology or partner, but coordinating across the coalition to match the right capabilities to each project's needs.
The Foundation is facilitating select reference project deployments alongside early participants. These engagements validate the model in practice, de-risk commercial execution with energy providers, and generate shared learnings that strengthen the framework for all participants. You retain control of your energy strategy; the Foundation accelerates the convening, vetting, and partner coordination.
This engagement is designed for companies building GW-scale AI infrastructure pipelines that recognize power as their primary bottleneck and want to move faster than assembling the energy ecosystem alone. The Foundation brings the coalition, framework, and deployment coordination.
Next step: Schedule a scoping call to discuss your pipeline. Email aric@sijbrandijfoundation.org with an overview of your development plans (target geographies, timeline for power decisions, where coordination support would be most valuable).
The Data Center Power Playbook provides a framework through which data centers deliver measurable public benefit: local energy affordability improvement, grid reliability strengthening, economic development, and avoidance of negative environmental impacts including local air pollution and water resource strain.
The Foundation's position is direct: data centers that bring additionality and embed flexibility should earn accelerated interconnection and favorable tariff treatment. This alignment between data center development priorities and public interest is embedded in the Playbook's execution model.
Many regulators may be unaware that co-development models exist as an alternative to conventional grid service or on-site gas baseload turbine procurement. Proactive engagement before filing interconnection requests is essential. Developers who educate regulators on rate-suppressive load structures and engage host communities early will shape the rules.
Large-load tariffs and interconnection standards are actively being developed across utilities, states, RTOs, and federal regulators. The Foundation is engaged in this landscape and available as a resource — not as an advocate for a specific company or technology, but as a source of framework-level analysis and evidence-based input on topics including tariff design, interconnection policy, grid impact modeling, and rate suppression potential.
If this framework aligns with your policy objectives, we welcome the opportunity to work together to ensure these standards incentivize the best outcomes for both industry and the public.
Next step: Email aric@sijbrandijfoundation.org to discuss how the Playbook framework intersects with your regulatory developments.
Stay Updated
The Playbook is a living document.
We iterate publicly as reference projects deliver real-world data, the regulatory landscape evolves, and the partner coalition expands. Get updates when we publish.