Power Scarcity Is the Real Barrier in the Global AI Data Center Race

The 14 largest publicly owned data center operators are tracking close to $750 billion in capital expenditure in 2026, up from roughly $450 billion the prior year, according to BloombergNEF’s analysis of the global artificial intelligence (AI) data center infrastructure surge. The United States, Europe, and the Middle East are each committing capital at a pace that is reshaping energy markets, foreign policy, and real estate strategy at once. But capacity market clearing prices in PJM Interconnection’s grid zone tell a sharper story: those prices jumped from $28.92 per megawatt to $329 per megawatt in two years, a tenfold increase driven almost entirely by AI workload demand.

Every region racing to lead the next generation of AI compute is discovering the same floor: chips, fiber, and sovereign capital are available if you have the right balance sheet. Electrons, delivered reliably at the right cost in the right place, are not.

Compute Geography Is Replacing Cloud Geography

For most of the past decade, data center strategy was about proximity to users and to existing internet exchange points. That model broke when AI training runs began requiring gigawatt-scale sustained load. JLL’s 2026 global data center market outlook projects that nearly 100 gigawatts of new capacity will come online between 2026 and 2030, effectively doubling total global supply. The sector is forecast to expand at a 14% compound annual growth rate (CAGR) through the end of the decade, requiring up to $3 trillion in cumulative investment.

The geography is shifting alongside the scale. Hyperscalers and infrastructure investors are assembling enormous land positions capable of supporting multi-campus ecosystems that scale from hundreds of megawatts toward gigawatt-level footprints. Site selection criteria that once centred on fiber diversity and latency now rank speed to power first, followed by grid resilience, water access, and long-term transmission contracts.

• 23 GW of data center IT capacity was under construction globally at the end of Q3 2025, with roughly three-quarters of it in the United States (BloombergNEF).
• ~$11.3 million per MW is the average global data center construction cost in 2026, up from $7.7 million in 2020 (JLL).
• 14% CAGR is the projected annual growth rate for the global data center sector through 2030.
• 50% of all data center workloads are projected to be AI-driven by 2030, up from roughly 25% in 2025 (JLL).

Northern Virginia’s Grid Is Cracking Under the Weight

PJM Interconnection covers 67 million people across the US Mid-Atlantic and Midwest, and its territory includes Northern Virginia, the world’s densest concentration of data centers. In May 2026, PJM released a white paper declaring that the region “has years, not decades” to fundamentally overhaul how it operates.

The current situation is not tenable.

PJM chief executive David Mills wrote those words in the foreword to that white paper. The stress is specific and structural. Northern Virginia is home to over 600 data centers, and another 50 gigawatts of projects sit active in the region’s interconnection queue. Harvard’s Belfer Center analysis of AI data centers and the US electric grid documents a July 2024 incident in which a voltage event knocked 60 data centers simultaneously offline, producing a 1,500-megawatt power surplus that forced emergency grid adjustments. Capacity market prices for PJM’s 2026-2027 delivery year settled at $329 per megawatt, against $28.92 just two years prior.

Lawrence Berkeley National Laboratory projects that US data center demand will grow from 176 terawatt-hours in 2023 to between 325 and 580 terawatt-hours by 2028, potentially placing data centers at between 6.7% and 12% of total national electricity consumption. Between 2023 and 2024, utilities revised their five-year summer peak demand forecasts upward from 38 gigawatts to 128 gigawatts, a more than threefold jump in a single planning cycle.

• US mid-Atlantic grid capacity prices: from $28.92/MW to $329/MW over two delivery-year cycles (PJM Interconnection data).
• CenterPoint Energy in Texas reported a 700% increase in large-load interconnection requests between late 2023 and late 2024.
• NERC (the North American Electric Reliability Corporation, North America’s grid watchdog) issued a Level 3 alert warning of “significant risks” to the bulk power system from rapid data center demand growth.
• Loudoun County, Virginia, showed grid harmonic distortions more than four times the national county average, correlated with data center concentration in a Bloomberg data analysis.

Europe Reaches for Gigafactories

Much of Europe’s cloud and AI stack still runs on infrastructure owned by US hyperscalers, a dependency that has shifted from commercially convenient to strategically sensitive. The European Commission’s AI Continent Action Plan, launched in April 2025, frames compute capacity as a core element of strategic autonomy, and subsequent legislative work has backed that framing with capital. In January 2026, the EU Council amended the regulation governing EuroHPC JU (the European High-Performance Computing Joint Undertaking), extending its mandate to include AI gigafactories, large-scale national compute clusters equipped with roughly 100,000 advanced processors each. Seven member states, including France, Germany, Italy, Spain, the Netherlands, Finland, and Portugal, are co-financing the programme.

The funding architecture is substantial. The EU’s InvestAI initiative has mobilised up to €200 billion for AI infrastructure and large-scale compute. A parallel Chips Act 2.0 is targeting public-private commitments approaching €80 billion, aimed at raising Europe’s share of global semiconductor output to 20%. In early 2026, Mistral AI raised €830 million in institutional debt, the first time a European AI company financed a hyperscale-level data centre without US venture capital, signalling that sovereign AI infrastructure is being treated as a distinct investable asset class.

Electricity pricing complicates the sovereign ambition. European industry pays roughly double the electricity rates of US counterparts, according to data from ACER, the EU’s Agency for the Cooperation of Energy Regulators. With electricity costs compounding quickly at the scale of AI training runs, developers are gravitating toward secondary markets where renewable generation is abundant and cheap. Scandinavia, Finland, and Norway attract investment because of hydropower access, natural cooling, and low-carbon intensity, drawing capital away from historically dominant markets like London and Dublin, which are now facing grid constraints of their own.

Private developers are moving independently of policy programmes along the same geographic logic. Nebius, the Netherlands-based AI cloud provider, announced a 310-megawatt facility in Finland and a 240-megawatt site near Lille, France, each comparable in scale to a planned EU gigafactory. Europe’s data center capacity is projected to grow 70% by 2030, but AI workload demand is expected to outpace that build rate, leaving sovereign compute access as a structural rather than a cyclical question.

The Gulf’s Sovereign Bet Faces a Wartime Test

Before regional conflict erupted in February 2026, the UAE, Saudi Arabia, and Qatar were executing some of the most aggressive AI infrastructure programmes anywhere. The UAE’s Stargate project, a joint development involving G42 (Abu Dhabi’s state-linked AI company), OpenAI, NVIDIA, Oracle, and SoftBank, is designed to deploy a 5-gigawatt AI campus in Abu Dhabi spanning 10 square miles, the largest AI infrastructure project outside the United States. Saudi Arabia’s HUMAIN, a state-backed entity under the Kingdom’s Public Investment Fund (PIF), is pursuing a $77 billion infrastructure strategy targeting 1.9 gigawatts of data center capacity by 2030. Qatar’s investment authority stood up Qai, its own AI company, in partnership with Brookfield Asset Management.

Then Iranian drone strikes hit AWS data centers in Bahrain and the UAE in the spring of 2026. An Oracle facility in Dubai was also struck. Amazon Web Services chief executive Matt Garman said the company worked around the clock to keep Middle East services operational after the attacks. The conflict introduced a variable that sovereign wealth modelling had not fully priced: the physical vulnerability of multi-billion-dollar campuses to targeted infrastructure attacks in a region with active hostilities nearby.

Sovereign investors have not changed course. The major Gulf wealth funds, including the PIF of Saudi Arabia, the Abu Dhabi Investment Authority, and the Qatar Investment Authority, collectively deployed $66 billion into AI and digitalisation in 2025, according to Global SWF. G42 said publicly in April 2026 that its direction “remains unchanged” and that AI “will become as foundational to economies and societies as electricity.” Projects continue with renewed emphasis on resilience and redundancy rather than single-site concentration. Per the Middle East Institute’s primer on Gulf and US AI strategy, the near-term customer base for Gulf compute is primarily sovereign, regional, and inference-at-the-edge demand rather than global hyperscale workloads choosing between Abu Dhabi and established Western interconnection hubs; some rollout delays are expected in 2026, with a rebound anticipated in 2027.

Three Regions, One Decade-Long Race

Each corridor brings a structurally distinct combination of strengths and constraints to the same competition. No single region holds uncontested advantage across all four dimensions that determine long-term AI infrastructure leadership: compute scale, energy position, governance architecture, and geopolitical stability.

The United States retains the deepest hyperscaler ecosystems, the most mature capital markets for digital infrastructure, and the largest existing compute base. Europe has the governance architecture that attracts regulated, sovereignty-sensitive workloads and is now building the compute capacity to match. The Gulf can do something neither region replicates quickly: design a gigawatt-scale AI campus from the ground up, with integrated energy contracts, on a timeline measured in months rather than the seven to ten years a US transmission infrastructure upgrade typically takes.

Energy as the New Geopolitical Variable

Goldman Sachs Research projects that global data center electricity consumption will surge 165% by 2030 compared with 2023 levels. The World Economic Forum and Bain, in a January 2026 white paper, forecast that annual investment in AI-dedicated infrastructure will exceed $400 billion per year by 2030. Those projections are downstream of a structural shift already visible in real-time pricing: data centers have stopped behaving like real estate assets and started behaving like energy assets, with site selection driven by megawatt access ahead of fiber diversity or tax incentive packages.

The regions best positioned for long-term AI leadership, as the World Economic Forum’s analysis of shared AI infrastructure and sovereignty argues, are those capable of combining reliable baseload power with governance systems trustworthy enough to host sensitive workloads and talent ecosystems deep enough to operate them. That combination requires American scale, European governance, and Gulf energy operating in the same supply chain. No single region has all three, which is why the decade-long trajectory looks less like a winner-take-all race and more like a structured negotiation between complementary strengths, with energy the variable that connects them.

If Gulf infrastructure weathers the geopolitical pressure and scales on its planned timeline, the global AI compute map by decade’s end will look genuinely multiregional rather than American-dominant. If the US grid crisis deepens faster than transmission investment can follow, and Europe’s gigafactories slip their build schedules, America’s compute lead could narrow in absolute terms precisely at the moment its energy constraint tightens most.