OpenAI’s $500 Billion Ohio Gambit and the Crunch Reshaping Global AI Infrastructure

The race to build the physical backbone for artificial intelligence is producing numbers that are difficult to grasp. In June 2026, reports emerged that OpenAI is negotiating to lease a proposed 10-gigawatt data center campus on federal land in Ohio . The projected cost: at least $500 billion, based on current prices for chips, labor, and power . That single campus figure matches the entire headline investment of the Stargate project, the $500 billion joint venture between OpenAI, SoftBank, Oracle, and MGX meant to blanket the United States with AI infrastructure by 2029 .

While the Ohio discussions are not yet final, the reported terms illustrate how the industry’s ambition is colliding with physical reality. The proposed deal would give OpenAI control of the equipment under a 20-year lease, with rent payments starting only when the facility becomes operational . The first 800-megawatt phase is expected by 2028, a timeline that already looks tight given the grid constraints detailed below .

The Stargate Project’s Footprint Expands

The Ohio negotiations are happening against the backdrop of Stargate’s broader buildout. Launched at a White House event in January 2025, the project aims to deploy roughly 10 gigawatts of AI compute capacity across the United States over four years . The first site, in Abilene, Texas, is already operational on Oracle Cloud Infrastructure with racks of Nvidia GB200 chips . By September 2025, the consortium had announced five additional sites in Texas, New Mexico, Ohio, and a still-undisclosed location in the Midwest, bringing the total capacity under development to around 7 gigawatts .

Nvidia’s role extends beyond chip supply. The chipmaker is a key technology partner in Stargate and has been engaged in a separate $100 billion equity-plus-chip deal with OpenAI . For the Ohio campus specifically, reports indicate Nvidia is in discussions to use its own balance sheet to provide financial backing—an arrangement that deepens the financial interdependency between AI model builders and the hardware companies that supply them .

Meta Bets on India Through Reliance

Away from the US, Meta made its first significant AI infrastructure push into India. On June 10, 2026, the company announced a partnership with Reliance Industries to build its first AI-enabled data center in the country . Reliance will construct a 168-megawatt facility in Jamnagar, Gujarat—also home to the world’s largest single-site oil refinery—and Meta will lease the capacity to power its products and AI workloads .

This data center deal sits inside a broader Reliance-Meta relationship. The two companies already operate a joint venture, Reliance Enterprise Intelligence Ltd., focused on building Llama-based AI solutions for Indian businesses . The Jamnagar project gives Meta dedicated built-to-suit capacity in one of its fastest-growing markets, while further anchoring Reliance’s pivot from energy and retail into digital infrastructure .

Separately, Reliance Industries, Brookfield Corporation, and Digital Realty have committed $11 billion to develop 1 gigawatt of AI data capacity in Andhra Pradesh, underscoring India’s emergence as a second front in the global AI infrastructure buildout .

The Power Grid as the Real Bottleneck

For all the investment announcements, the binding constraint on AI data center expansion is increasingly not capital, chip supply, or construction labor—but the availability of grid-scale electricity. Academic and industry analyses now frame power delivery as the primary risk to deployment timelines .

The scale of the problem is stark:

• U.S. electricity consumption hit a record 4,195 terawatt-hours in 2025, and demand continues to climb .
• Data centers could account for 9% to 12% of total U.S. electricity demand by 2030, up from roughly 4% in 2024 .
• North America is expected to consume half of global AI and data center electricity demand by 2030, according to DNV’s energy transition outlook .

The largest individual data center campuses now require more than a gigawatt of continuous power—enough to supply approximately 850,000 households . Traditional power grids were not designed for this type of concentrated, 24/7 industrial load. In regions where AI clusters are densest, utilities are already experiencing harmonic distortions, load relief warnings, and near-miss grid incidents .

Concrete financial consequences are materializing. In Q4 2025, U.S. data center developers added only 25 gigawatts of new capacity to their project pipeline—half the previous quarter’s figure—because utilities cannot build generation and transmission capacity fast enough to keep pace . In the PJM Interconnection, which covers 65 million residents from the Mid-Atlantic to the South, analysts project a 49 GW generation shortfall by 2028 . The World Economic Forum has described grid connectivity as the “strategic bottleneck” for AI transformation .

What This Means for AI Capex and Equities

The combination of enormous spending announcements and growing grid friction creates a two-sided narrative for investors.

The bull case for infrastructure spending remains visible in the raw numbers. Stargate alone targets $500 billion in total investment. SoftBank has committed an additional $3 billion just to overhaul an Ohio factory that will produce equipment for OpenAI data centers . In India, billions more are flowing into AI data capacity through both the Meta-Reliance partnership and the Andhra Pradesh consortium.

The risk case centers on execution. If utilities cannot interconnect new campuses on the timelines developers have promised, announced capex figures will become harder to realize . The issue is not whether AI needs more compute—the demand signal is clear—but whether physical infrastructure can be delivered at the pace that equity markets have priced in.

This tension reshapes how to think about sector exposure:

• Chipmakers benefit from the long-term demand trajectory that hyperscale buildouts imply. However, company-level revenue impact depends on projects actually reaching power-on status, not just being announced .
• Utilities and power producers are becoming more central to the AI investment thesis because grid access, not chip availability, is the emerging constraint . Companies with existing transmission infrastructure, generation capacity, or interconnection expertise in high-demand regions may gain relative advantage.
• Grid-connected infrastructure players—those involved in transmission equipment, electrical contracting, and interconnection—are likely to see increased attention as the bottleneck shifts from compute hardware to power delivery .
• The Indian market is gaining visibility as a secondary data center hub. Reliance’s role spans both the Meta deal and the Andhra Pradesh consortium, positioning the conglomerate as a direct play on Indian AI infrastructure .

Key risk to monitor: If grid interconnection delays spread and timelines slip across multiple Stargate sites or other large projects, the market may begin discounting the next round of AI capex announcements until power delivery proves capable of absorbing the load .