AI Data Center Build

State of play.

• Hyperscalers are deploying debt at unprecedented scale to fund AI compute infrastructure. Meta has arranged approximately $13 billion in financing for a 1-gigawatt El Paso facility through Morgan Stanley and JPMorgan, while hyperscalers collectively project over $630 billion in AI infrastructure capital expenditures for 2026 alone .
• AI data center demand is now reshaping semiconductor supply chains beyond the data center sector itself. Samsung, SK Hynix, and Micron have reallocated manufacturing capacity toward high-bandwidth memory for AI, producing cascading DRAM and NAND shortages—memory chip prices doubled in Q1 2026 and are forecast to rise another 63% in Q2—with Nintendo raising Switch 2 prices globally and Sony raising PS5 prices by $100 in direct response .
• Wall Street's appetite for AI infrastructure debt has material limits. Blue Owl Capital withdrew from Oracle's $10 billion Michigan data center deal, JPMorgan and other lenders have hit internal exposure limits on Oracle-leased facilities, and Oracle's debt-to-equity ratio has reached 4.3—while Vinson & Elkins is now publishing practitioner guidance on SPVs, asset-backed securitizations, and private credit as the structured capital layer filling the gap .
• Texas is emerging as the defining jurisdiction for AI data center legal and regulatory pressure. ERCOT interconnection queue congestion has created 5-to-10-year delays, forcing Oracle, Google, and OpenAI toward independent on-site generation—and Vinson & Elkins has launched a dedicated practitioner series addressing the permitting, regulatory, and financing architecture that now governs project viability in the state .
• For counsel advising data center developers, hyperscaler infrastructure teams, project finance lenders, or consumer electronics and semiconductor clients, the practical baseline is a simultaneous squeeze: financing capacity is tightening for non-investment-grade borrowers, permitting timelines are lengthening through state and local action, grid interconnection rules are in flux ahead of a June FERC deadline, environmental enforcement is targeting operational power infrastructure, and AI chip demand is now generating antitrust and contract exposure in downstream consumer markets.

Where things stand.

• Debt-driven financing has become the dominant model for AI infrastructure buildout. The AI infrastructure buildout requires an estimated $5.3 trillion through 2030, with a $2.5 trillion funding gap; private credit is expected to provide roughly $800 billion of the $2.9 trillion needed for capital expenditures between 2025 and 2028. Individual data center projects now routinely cost billions of dollars, and SPVs, asset-backed securitizations, and hybrid equity-debt structures have become the standard toolkit for developers minimizing balance sheet impact .
• Private power contracting has become a prerequisite for project financing. Utility interconnection timelines stretching 4-8 years have forced developers to bypass traditional grid connections in favor of direct contracts with independent power producers—virtual PPAs structured as contracts for differences, with objective acceptance criteria and shortage allocation provisions. Lenders increasingly require these arrangements before approving construction financing .
• FERC has committed to finalizing large-load interconnection rules by June 30, 2026. The rulemaking under Docket No. RM26-4-000 will govern how loads exceeding 20 MW connect to the interstate transmission system, addressing cost allocation, co-location standards, and interconnection timing. PJM has already implemented related reforms with a 50 MW cost-shifting threshold, though full activation is delayed to 2027 .
• Texas is positioned to surpass Virginia as the nation's largest data center hub. Over 400 facilities are in planning or construction stages alongside roughly 387 already operational, with ERCOT interconnection queue congestion creating 5-to-10-year delays that are forcing major developers—including Oracle's 1.4-gigawatt West Texas facility—toward independent on-site gas generation to sidestep grid connection delays .
• AI data center demand is structurally reallocating semiconductor manufacturing capacity. Samsung, SK Hynix, and Micron are shifting output toward high-bandwidth memory for AI, creating cascading shortages in DRAM and NAND flash. Memory chip prices doubled in Q1 2026 and are forecast to rise another 63% in Q2. Samsung estimates 30 to 50 percent of planned 2026 data center deployments will slip to 2028 due to fab upgrade delays, helium shortages, and power grid constraints .
• Consumer electronics manufacturers are absorbing AI-driven chip cost inflation through mid-cycle price increases. Nintendo has raised Switch 2 pricing globally, attributing the move to AI data center memory demand, tariffs, and Iran conflict-related shipping costs; Sony has raised PS5 prices by $100 in the US market. Both companies expect elevated component costs to persist into 2027 .
• Nokia and Intel have both reported material revenue growth driven by AI data center demand. Nokia's Network Infrastructure division upgraded its full-year 2026 forecast to 12-14% growth, with AI and cloud net sales surging 49% in Q1; Intel's Data Center and AI segment generated $5.1 billion in Q1 revenue, up 22% year-over-year .
• State-level regulatory divergence is accelerating. Maine has enacted a moratorium on facilities drawing over 20 MW through November 2027. California has introduced legislation eliminating categorical CEQA exemptions and mandating monthly CEC disclosure of data center energy and water metrics. Oklahoma is advancing a cost-causation framework requiring large loads to self-finance grid upgrades and post collateral .
• Environmental justice enforcement has reached the operational power infrastructure layer. The NAACP's federal Clean Air Act complaint against xAI alleges 27 unpermitted methane turbines at the Colossus 2 facility emit nitrogen oxides potentially exceeding 1,700 tons annually without required air permits, in communities already failing federal smog standards .

Latest developments.

• Nintendo raised Switch 2 pricing from $449.99 to $499.99 in the US, cut its FY27 sales forecast to 16.5 million units from nearly 20 million, and attributed both moves in part to AI data center-driven memory chip cost inflation; the stock fell 7.47% and has declined roughly 50% year-to-date .
• Sony and Nintendo publicly attributed consumer hardware price increases to AI data center memory demand—with memory chip prices having doubled in Q1 2026 and forecast to rise another 63% in Q2; new fabrication capacity requires at least one year to operationalize, sustaining cost pressure into 2027 .
• Vinson & Elkins launched its "Powering Progress" practitioner series, with Episode 1 addressing ERCOT grid capacity constraints, interconnection queue congestion, and permitting timelines in Texas .
• Vinson & Elkins released Episode 2 of "Powering Progress", examining structured capital approaches—SPVs, asset-backed securitizations, and private credit—for AI infrastructure portfolios against a backdrop of an estimated $2.5 trillion funding gap through 2030 .

Active questions and open splits.

• Credit quality bifurcation in AI infrastructure debt. Meta, Google, and Microsoft can access debt markets at scale; Oracle cannot. The open question for lenders and counsel is where the line sits—and whether the Oracle episode signals a broader syndication freeze for non-hyperscaler borrowers or is idiosyncratic to Oracle's balance sheet .
• Structured capital adequacy for the $2.5 trillion funding gap. With traditional equity and debt financing insufficient at current deployment scale, the question is whether SPVs, asset-backed securitizations, and private credit arrangements can absorb the gap—and what covenants, collateral structures, and lender protections will become market-standard for projects that depend on unproven power delivery commitments .
• Antitrust scrutiny of memory chip allocation decisions. Samsung, SK Hynix, and Micron collectively dominate global chip production and are prioritizing high-margin AI buyers over consumer electronics manufacturers. Whether antitrust regulators treat allocation decisions favoring AI data center customers as actionable—particularly as consumer prices spike and Oxford Economics documents the first broad computer price increases since the 1980s—is unresolved and increasingly visible to enforcers .
• Contract exposure for consumer electronics manufacturers facing chip allocation shortfalls. As Samsung, SK Hynix, and Micron prioritize AI buyers, device manufacturers who cannot fulfill existing supply agreements face potential liability. Whether force majeure provisions cover AI-driven reallocation decisions—as opposed to physical supply disruption—is an open drafting and litigation question .
• Clean Air Act permitting obligations for on-site data center power generation. The xAI Memphis lawsuit tests whether operators who install on-site generation to bypass grid constraints must obtain air permits before operation. The answer will define the compliance baseline for every hyperscaler using behind-the-meter gas generation—a population that is growing as natural gas becomes the preferred reliability solution .
• FERC's June rulemaking scope on co-location and cost-shifting. The final rules in Docket No. RM26-4-000 will determine whether co-located loads and behind-the-meter generation arrangements—now standard in private power contracts—face new jurisdictional requirements or cost-allocation obligations. The PJM 50 MW threshold may or may not carry into the national standard .
• Federal preemption of state data center moratoriums. The Trump administration's December 2025 executive order on eliminating state law obstruction of national AI policy creates a direct collision with Maine's moratorium and California's CEQA reform. Whether federal preemption arguments succeed in court—and what standard applies—is unresolved .

What to watch.

• FERC's June 30, 2026 deadline for final large-load interconnection rules in Docket No. RM26-4-000—the single most consequential near-term regulatory event for data center developers and utilities .
• Whether antitrust regulators in the US or EU open formal inquiries into memory chip allocation decisions by Samsung, SK Hynix, and Micron as consumer price increases become politically visible through Nintendo and Sony's public disclosures .
• Early motions practice in the NAACP v. xAI Clean Air Act case—particularly whether the court grants injunctive relief requiring turbine shutdown pending permitting .
• Whether Oracle's Michigan collapse triggers renegotiation pressure on its $300 billion OpenAI cloud computing deal, producing contractual disputes that reach public filings .
• Oklahoma Senate action on the cost-causation framework requiring large loads to self-finance grid upgrades—if enacted, the model is immediately exportable to other energy-rich states .
• Whether additional consumer electronics manufacturers beyond Nintendo and Sony make public disclosures attributing price increases to AI chip demand, accelerating regulatory and litigation attention to memory chip allocation practices .