The Global Speed Race Behind AI Infrastructure

Published: September 12, 2026 | Category: Technology | By Mahesh

The Carnegie Endowment for International Peace, an independent foreign policy think tank rather than an industry advocate, published research in June 2026 stating plainly that as of May 2025, almost three-quarters of the world's advanced AI computing clusters sat on American soil, and U.S. projects also move faster than those in most other countries.^[1] That lead is genuine. The same report's next sentence is the one worth sitting with: but that lead is fragile.^[1] Depth Grid's coverage this week has documented the specific domestic constraints straining that lead from the inside, grid capacity, gas turbine backlogs, chip packaging, water, and skilled labor. This final piece in the series looks outward: while the United States works through each of those constraints one at a time, China and the Gulf states are mobilizing around a single competitive dimension the U.S. cannot easily replicate, execution speed, and the gap this creates deserves to be understood on its own terms.

What the Carnegie Endowment's Own Research Actually Found

The Carnegie Endowment's own report frames the current moment with a precision worth quoting directly: domestic constraints, grid capacity, permitting rules, political opposition, are tightening, abroad, China is mobilizing to close the gap, while Gulf states are touting energy and capital to attract developers, and many traditional U.S. allies, meanwhile, risk being left behind entirely.^[1] The scale of that final point is striking on its own: the report finds every major publicly reported AI data center across the whole of Europe, combined, appears to contain less computing power than a single Amazon-Anthropic mega-cluster located at New Carlisle, Indiana, and that some traditional U.S. allies, including Australia, Italy, and South Korea, have no major publicly known operational AI chip concentrations at all as of the report's publication.^[1] That comparison matters for understanding the actual shape of the current global race: it is not a broad, multipolar contest, it is increasingly a race between the United States, China, and a small number of Gulf states moving with genuine speed and capital, while most of the rest of the world, including many of America's closest formal allies, is not currently competing at meaningful scale at all.

The specific constraints Carnegie names as tightening America's own domestic position, grid capacity, permitting rules, and political opposition, map directly onto the exact bottlenecks Depth Grid has documented individually this week. That alignment between an independent foreign policy research organization's own framing and the ground-level industry data examined throughout this series is itself a useful form of cross-validation: this is not merely an industry complaint about permitting friction, it is a conclusion independent geopolitical analysts have reached studying the same underlying constraints from a national competitiveness perspective rather than an individual project economics perspective.

Why China Builds Faster, and Cheaper

Independent industry analysis quantifies China's execution advantage in terms directly comparable to the American constraints documented throughout this series. Data centers in China can pay less than half the rates for electricity that American data centers do, and projects in China can move from planning to operation in months, compared to years in the U.S., a gap the same analysis attributes specifically to faster permitting and fewer regulatory hurdles rather than to any difference in underlying technical capability.^[2] That framing is worth taking seriously precisely because it does not claim China holds any inherent technological advantage in AI chip design, model capability, or engineering talent, the dimensions most Western coverage of the US-China AI competition tends to focus on. The claimed advantage is narrower and, in some ways, more consequential for the specific infrastructure bottleneck this series has documented: the speed and cost at which physical capacity can actually be built and energized.

This is precisely the dimension where Depth Grid's earlier reporting this week found the United States most constrained. The gas turbine backlogs, multi-year grid interconnection delays, and years-long high-voltage transmission line permitting timelines documented in this series are, in aggregate, exactly the category of friction a centrally coordinated, faster-permitting system can more readily bypass. A system that can compress planning-to-operation timelines from years to months is not competing on chip design or algorithmic capability at all, it is competing on the ability to convert capital into physical, energized computing capacity faster than a competitor operating under a more fragmented, multi-jurisdictional permitting and grid interconnection process, which is precisely the process Depth Grid's earlier reporting on the grid capacity bottleneck described as commonly taking three or more years for a single major U.S. project.

The Gulf States' Bet: Speed and Power as the Pitch Itself

The Gulf states have built their entire pitch to AI infrastructure developers around exactly the two variables the United States currently struggles with most: available power and execution speed. Saudi Arabia's HUMAIN program, backed directly by the Public Investment Fund, has committed approximately $100 billion across eleven data centers targeting 2.2 gigawatts of capacity, and current tracking places Saudi operating data center load at 440 megawatts and the UAE's at 340 megawatts, together representing close to 80 percent of the entire Middle East's roughly 1 gigawatt total installed base.^[3] One analysis of the region's strategic positioning frames the underlying advantage directly: Persian Gulf nations offer unconstrained build zones, cheap solar, and integrated gas and nuclear hybrid grids capable of delivering power around the clock, precisely the firm, continuous power supply Depth Grid's earlier reporting identified as the specific requirement most straining U.S. grid capacity.^[4]

The execution speed differential in the Gulf is documented with specific, concrete examples rather than general characterization alone. G42's first phase of a 1 gigawatt project broke ground just months after its underlying bilateral deal was signed, and Saudi Arabia's HUMAIN program is described as accelerating site selection through direct royal decree, an executive mechanism with no close analog in the more fragmented, multi-agency, multi-jurisdiction U.S. permitting environment.^[4] This regulatory velocity, unlike the years-long data center zoning timelines common in the United States and the European Union, allows these sovereign states to compress project timelines specifically because a small number of empowered decision-makers can authorize major infrastructure commitments directly, without navigating the multiple layers of state, local, and often multi-state regulatory approval processes Depth Grid's earlier reporting on U.S. transmission permitting described as routinely stretching past a decade for major interstate transmission lines.^[5]

What One Operator on the Ground Actually Says the Constraint Is

Perhaps the most directly useful data point in this entire international comparison comes from an on-record statement by an actual regional industry executive rather than from a policy analyst's secondhand characterization. Mehdi Paryavi, founder and chief executive of the International Data Center Authority, told Asharq Al-Awsat in an interview published July 21, 2026 that energy, not chips or capital, is the biggest constraint on AI growth, followed by workforce availability and public policy.^[3] That statement, from an executive with direct visibility into the actual bottlenecks facing Gulf AI infrastructure development, produces a ranking genuinely worth pausing on: energy first, workforce second, public policy third, with chips and capital explicitly placed below all three.^[3] That hierarchy is a striking, independent confirmation of the exact pattern Depth Grid has documented across this entire series in the U.S. context, power, then labor, are the binding constraints, not chip supply or available investment capital, even in a region widely assumed to be defined primarily by its access to sovereign capital.

Saudi Arabia's own case illustrates that even a well-capitalized, chip-authorized program faces the identical underlying power constraint documented throughout this series. Saudi Arabia does not lack authorized chips, with HUMAIN holding a U.S. authorization for the equivalent of up to 35,000 GB300-class accelerators and phase-one hardware already confirmed at 18,000 Nvidia GB300 systems, yet the same analysis concludes the electricity actually running Saudi AI data centers in 2027 and 2028 will be burned, meaning generated from natural gas, not harvested from the country's renewable energy programs, since the gap between the kingdom's installed renewable capacity and its stated 130 gigawatt renewable target remains among the widest in its entire Vision 2030 development plan.^[3] This detail matters because it confirms the power constraint is not a uniquely American structural failure, it is a genuine, near-universal physical limitation on how quickly any country, however well-capitalized or fast-permitting, can bring firm, reliable electricity generation online to match AI infrastructure's growth rate.

Why America's Structural Advantages Are Real, and Not Permanent

A rigorous accounting of this competitive picture requires taking the case for continued American advantage seriously rather than treating this piece's international comparison as purely a story of decline. Current industry analysis of the American position notes real, durable structural strengths: capital markets and experienced developers can still finance multi-gigawatt campuses relatively quickly by global standards, and despite genuine grid and permitting friction, the United States still has more available high-voltage power and fiber infrastructure than most competitors, with Texas alone hosting 25.8 gigawatts of tracked data center capacity, followed closely by Indiana, Louisiana, Pennsylvania, and Ohio.^[6] That same analysis notes the combined 2026 capital expenditure guidance across the four largest American hyperscalers runs near $725 billion, with the bulk of that spending directed specifically at AI-ready capacity, a scale of committed capital that few competing nations or regional blocs can currently match in aggregate.^[6]

The geographic diversification within the U.S. market itself is also a genuine structural advantage worth naming directly: spreading major projects across Texas, Indiana, Louisiana, Pennsylvania, and Ohio reduces single-point regulatory or grid failure risk while keeping every project inside one broadly consistent national regulatory and capital environment, a genuine benefit relative to a multinational developer trying to coordinate projects across several entirely separate sovereign jurisdictions, each with its own distinct legal, currency, and political risk profile. The honest synthesis of this entire international comparison is that the United States retains real, substantial structural advantages in capital depth, existing infrastructure base, and market coherence, while genuinely trailing specific competitors, particularly China and the Gulf states, on the narrower, but increasingly decisive, dimension of raw execution speed, the single variable this piece has argued is now the actual contested terrain of the global AI infrastructure race.

What This Means for Anyone Allocating Capital or Capacity

Treat execution speed as a distinct, trackable metric separate from chip access or available capital, when evaluating any country's or region's AI infrastructure competitiveness. Given that the International Data Center Authority's own chief executive ranks energy and workforce availability above chips and capital as binding constraints, and given the well-documented planning-to-operation timeline gap between the U.S. and China, an investor or strategist assessing where global AI compute capacity is likely to concentrate over the next several years should weight demonstrated execution speed, actual energization dates rather than announced capacity, at least as heavily as headline capital commitments or chip allocation figures.

Recognize that the Gulf states' advantage is structural and durable, not a temporary capital-driven anomaly. The compressed permitting timelines enabled by direct executive authority in Saudi Arabia and the UAE reflect a fundamentally different governance structure than the United States' multi-jurisdictional regulatory environment, meaning this specific speed advantage is unlikely to erode over time the way a purely capital-driven advantage might, and should be treated as a durable feature of the competitive landscape for planning purposes extending well beyond the current AI infrastructure buildout cycle.

Watch US policy responses to grid and permitting friction as the single most consequential lever available to preserve America's current lead. Given that the Carnegie Endowment's own research explicitly identifies domestic permitting rules and political opposition, not a lack of capital or chip access, as the primary forces tightening around America's current advantage, any meaningful policy reform specifically targeting grid interconnection speed and transmission permitting timelines, rather than additional chip export controls or further capital incentives, is likely to matter more for preserving America's competitive position than any other single category of federal or state action available.

Common Questions

This analysis is editorial commentary based on publicly available sources cited above. It is not financial, investment, or geopolitical policy advice. Infrastructure capacity figures, timelines, and national comparisons cited reflect data and analysis available as of publication and change frequently; verify current figures with the cited research organizations and companies before making decisions based on this information.

Sources

1. Carnegie Endowment for International Peace, "The Compute Coalition: How to Build the Future of AI in the Free World," June 17, 2026. Link
2. IEEE ComSoc Technology Blog, "China vs U.S.: Race to Generate Power for AI Data Centers as Electricity Demand Soars," February 16, 2026. Link
3. Vision2030.ai, "Saudi AI's Real Constraint Is Power, Not Chips," citing Mehdi Paryavi/Asharq Al-Awsat interview of July 21, 2026, July 31, 2026. Link
4. Sanie Institute, "From Oil to Algorithms: Trillion-Dollar Capital Shift in Global AI Geopolitics," 2026. Link
5. Quartz, "America's Power Grid Can't Keep Up With AI Demand," citing US Department of Energy and Transmission Agency of Northern California data, May 17, 2026. Link
6. SuccessKnocks, "AI Infrastructure Investment by Country 2026," September 2026. Link

Read More on Depth Grid

• The Compute-and-Power Bottleneck: Why AI's Next Decade Will Be Decided by Electrons, Not Algorithms
• Microsoft's Own President Says Electricians, Not Chips, Are the Biggest Constraint on AI Expansion
• Water, Not Power or Chips, Is Killing the Most AI Data Center Projects Right Now

Article by Mahesh | Depth Grid