Data-Center Power Shortages Will Delay AI Rollouts by 2027—Here’s Why

The AI infrastructure race is colliding with grid capacity in real time. US data center power demand is projected to more than double in two years—from 31 GW in 2025 to 66 GW by 2027—but the electrical grid isn’t expanding fast enough to keep up. The result: AI TechForecast predicts that power shortages will restrict 40% of planned AI data centers by 2027, forcing companies to choose between geographic relocation, renewable energy investments, or shelving non-critical projects. The market is already pricing this in: capacity clearing prices in the PJM grid jumped 11-fold in a single year.

This isn’t speculative. It’s infrastructure math, and it will reshape where AI gets deployed and who wins the race to scale.

The 11x Price Spike That Signals Real Scarcity

The PJM Interconnection—the grid operator serving 13 states and DC—runs an annual capacity auction where utilities bid to guarantee power availability during peak demand. In 2024–25, the clearing price was $28.92 per megawatt. In 2026–27, it cleared at $329.17 per megawatt.

That’s an 11-fold increase in a single year.

This isn’t noise; it’s a market signal. Morgan Stanley’s energy research directly attributes the spike to data center growth outstripping grid investment. When capital markets price in this kind of scarcity, it reflects real constraints that utilities, operators, and corporate planners are already factoring into their decisions.

The PJM market covers roughly 65 million people across the Northeast, Mid-Atlantic, and parts of the Midwest. If capacity prices are spiking this dramatically in one of America’s largest grids, the implication is clear: power is becoming a binding constraint on data center deployment. Companies bidding for power contracts in 2026 are paying 10 times what they paid two years ago, and they’re still uncertain whether they’ll secure enough capacity for 2027.

The Demand Cliff: 31 GW to 66 GW in Two Years

Why is power suddenly scarce? Because AI data center demand is accelerating faster than the grid was designed to handle.

Goldman Sachs projects US data center power consumption will grow as follows:

2025: 31 GW
2026: 41 GW (+32%)
2027: 66 GW (+61% year-over-year)

In absolute terms, that’s an additional 35 GW of demand in two years—roughly equivalent to the total electricity consumption of a mid-sized state.

To contextualize: data centers consumed 4.1% of US peak summer power demand in 2025. By 2027, that share is projected to reach 8.5%—more than doubling in just two years. For comparison, the entire residential sector (all homes in America) currently consumes about 20% of peak summer power. Data centers are on track to become a major grid load, second only to industry and HVAC.

The grid wasn’t built for this acceleration. Transmission lines take 7–10 years to construct. Power plants take even longer. Utilities historically move cautiously on capacity expansion because they’re penalized for over-investment—if they build generation that sits idle, they lose money. So they plan conservatively, based on historical trends.

But historical trends didn’t account for the AI boom. Every major tech company—Google, Microsoft, Amazon, Meta, Apple—is racing to build data center capacity simultaneously. They’re bidding against each other for power contracts in a market where supply is suddenly constrained. The result: a capacity crunch that’s already visible in energy markets.

Gartner’s 40% Constraint: What “Power Shortage” Means in Practice

Gartner’s forecast, cited through Enki AI’s analysis, predicts that power shortages will restrict 40% of AI data centers by 2027. This isn’t a minor bottleneck—it’s a structural limit that will force visible delays and prioritization.

In practical terms, “restrict” means:

Delayed project launches. Data center builds that were scheduled for 2027 will slip into 2028 or later.
Power rationing. Facilities that do come online may operate under capacity constraints, with workloads queued rather than executed immediately.
Prioritization by revenue impact. Companies will run AI inference for revenue-critical applications first; research, experimentation, and non-essential services get queued or shelved.
Higher power costs. Bidders in constrained markets will pay premium prices for available capacity, compressing margins on AI services.

For a startup with a 2027 AI launch roadmap, this is potentially catastrophic. For a hyperscaler like Google or Microsoft, it’s a problem—but one they can manage through geographic diversification, renewable investments, and negotiating power.

The hidden story: this shortage won’t be evenly distributed. It will create winners and losers based on who has power contracts locked in right now.

Regional Winners and Losers: Geography as Infrastructure Destiny

Not all US regions face the same grid stress. The distribution of available capacity will reshape where AI data centers get built over the next 18 months.

Regions with spare capacity:

Texas. The state has invested heavily in renewable generation (wind, solar) and operates a deregulated market that attracts power generation investment. The ERCOT grid has more flexibility than many regional grids. Major tech companies are already announcing data center projects in Texas, and this trend will accelerate.
Renewable-rich zones. Areas with access to hydroelectric, wind, or solar generation—parts of the Pacific Northwest, parts of the Mountain West—will attract data center investment because companies can secure long-term renewable power contracts.
Virginia. Already a data center hub with relatively stable grid conditions and existing infrastructure, Virginia will remain attractive for companies seeking to expand in the East.

Regions facing constraints:

The Northeast. The PJM grid is already tight. New Jersey, Pennsylvania, and New York will see elevated power costs and constrained capacity for new data center projects.
Parts of California. Despite renewable investment, California’s grid faces peak demand challenges and transmission bottlenecks. New data center projects will face permitting delays and power availability questions.
The Midwest. Grids in Illinois, Indiana, and Ohio face capacity constraints similar to PJM. Companies planning data center builds in these regions will encounter higher costs and longer timelines.

What this means: Data center projects will migrate geographically based on power availability. A company planning an AI data center in New Jersey might shift the project to Texas or Virginia. A hyperscaler might split its 2027 buildout across multiple regions to avoid hitting capacity constraints in any single grid.

This geographic shift is already happening. When you see announcements of new data center builds in Texas or the Mountain West, you’re not seeing random site selection—you’re seeing companies respond to power constraints in their original target regions. Power access is now a first-order factor in data center location strategy, alongside fiber connectivity and real estate cost.

The Competitive Moat: Energy as Structural Advantage

Here’s the forecast that matters most for investors and tech leaders: the companies that secure power contracts or build renewable generation capacity in 2026 will own a structural advantage in AI infrastructure for the next five years.

Hyperscalers have leverage. They can:

Negotiate long-term power contracts with utilities, locking in rates before scarcity prices rise further.
Invest in or partner with renewable generation projects, securing dedicated power supply.
Build data centers in low-constraint regions, spreading capacity across multiple grids.
Absorb higher power costs through scale and pass some of those costs to customers.

Startups and mid-size companies face the opposite situation:

They’re bidding in a constrained market with limited leverage.
They pay premium prices for available capacity.
They’re geographically limited to regions with available power.
They may not be able to build the infrastructure they need on their desired timeline.

This creates a durable advantage. If Microsoft secures a long-term power contract for 5 GW at $X per megawatt in 2026, and a startup tries to bid for capacity in 2027 at $Y per megawatt (where Y >> X), the startup faces a structural cost disadvantage that’s hard to overcome. The hyperscaler can train models faster, deploy inference cheaper, and scale services that competitors can’t afford to run.

This isn’t temporary. Power becomes a moat—a structural competitive advantage that persists as long as grid capacity remains constrained. For companies in AI infrastructure, this creates a 5–10 year window where early moves on power contracts and renewable investments compound into durable advantages.

Timeline: When Power Constraints Become Visible

The power shortage isn’t a 2027 problem; it’s a 2026 decision problem.

Now (mid-2026): Companies are negotiating power contracts for 2027–2028 data center builds. Capacity prices are elevated; bidders are competing aggressively. Hyperscalers are locking in long-term deals; smaller players are facing higher costs and uncertainty.
Late 2026–early 2027: The first wave of capacity constraints becomes visible. Projects that couldn’t secure power contracts get delayed. Regional disparities in data center deployment become obvious.
Mid-2027: Gartner’s 40% constraint becomes visible in deployment timelines. Companies announce delays. The AI roadmaps that assumed 2027 launches slip to 2028. Investors start pricing in infrastructure constraints as a factor in AI company valuations.
2028+: Utilities begin completing new transmission and generation projects that were fast-tracked in response to 2026–27 demand. Capacity begins to ease. But the competitive winners—companies that secured power early—have already built durable advantages in cost and scale.

FAQ: Power Shortages and AI Deployment

Q: Will power shortages actually stop AI development?

No. AI development will continue, but it will be geographically concentrated and more expensive. Companies will build data centers where power is available, pay higher costs in constrained regions, or delay non-critical projects. The 40% restriction means 40% of planned data centers will face delays or constraints—not that 40% of AI work stops. But for companies with tight timelines and limited geographic flexibility, the impact will be real.

Q: Is this just a US problem?

No. Europe and Asia face similar grid constraints. But the US is the largest AI infrastructure market, and the data here is most concrete. Similar dynamics are playing out in the UK (grid strain), parts of Europe (renewable intermittency), and Asia (rapid data center buildout in China and India). The global pattern is the same: demand is outpacing grid investment.

Q: Can renewable energy solve this?

Partially, but not immediately. Renewable generation (wind, solar) is growing, but it’s intermittent and requires battery storage or grid balancing. Building new renewable capacity takes 2–5 years. Hyperscalers are investing in renewables, but they can’t build fast enough to offset the entire 35 GW demand increase by 2027. Renewables will be part of the solution, but the shortage will persist through 2027 and likely into 2028.

Q: What should companies do now?

If you’re planning AI infrastructure: secure power contracts now, before prices rise further. Evaluate geographic options; Texas and renewable-rich zones will be cheaper and more available than constrained regions. If you’re investing in AI companies: factor power access and geographic location into your due diligence. Companies with secured power or renewable partnerships will have structural advantages.

The Bottom Line: Power Is the New Constraint on AI Scaling

The AI infrastructure race is real, but it’s hitting a hard physical limit: the electrical grid. The 11-fold jump in capacity prices, the 66 GW demand projection, and Gartner’s 40% constraint forecast aren’t speculation—they’re market signals and expert predictions backed by tier-1 sources.

AI TechForecast predicts that power shortages will visibly delay 40% of planned AI data center rollouts by 2027. Companies that secure power contracts and renewable capacity now will own a structural advantage in cost, speed, and scale for the next five years. Geographic location will become a first-order factor in data center strategy. And the AI race will increasingly be won by companies with the foresight to lock in energy access before the grid gets tighter.

The decisions are being made right now, in 2026. By 2027, the winners and losers will be obvious.

Meta description: AI TechForecast predicts power shortages will restrict 40% of data center rollouts by 2027. Demand is doubling; the grid isn’t. Here’s what it means for AI deployment.

Data-Center Power Shortages Will Delay AI Rollouts by 2027—Here's Why