Microsoft–Chevron Deal Broadens the AI Competition Race for Power Generation

Why does the Microsoft–Chevron power deal matter for the future of AI competition?

Microsoft’s agreement with Chevron to power a West Texas data center marks a broader shift in AI competition: control over infrastructure is moving upstream—from chips, cloud access, and data center capacity to the generation of energy itself.

Energy Forge One LLC, a wholly owned Chevron subsidiary, signed a 20-year power purchase agreement with Microsoft to develop a co-located West Texas power facility known as Project Kilby that would provide dedicated electricity to a Microsoft-operated data center.

According to Chevron, Kilby will deliver 2.67 gigawatts of capacity through a phased, modular buildout, with most generation coming from large GE Vernova turbines and associated electrical infrastructure, with more capacity from Solar Turbines, a wholly owned Caterpillar subsidiary. Microsoft says its associated Pecos, Texas, campus is a multibillion-dollar buildout expected to add approximately 2 gigawatts of global data center capacity over five to seven years to support AI and cloud demand.

To put it in perspective, 1 gigawatt can power between 750,000 and 1 million homes.

According to Chevron, co-locating new large-scale power generation with the data center is designed to deliver reliable, dispatchable (essentially on-demand) electricity directly to Microsoft while aiming to mitigate impacts on the regional grid. CNBC reports that the project would rely on natural gas from the Permian Basin, the vast oil-and-gas-producing region in West Texas and southeastern New Mexico. Chevron has described the project as a way to deliver power at a “competitive cost” by leveraging Permian natural gas and its existing infrastructure.

Microsoft says AI and cloud growth require energy infrastructure that can “scale quickly and reliably,” adding that it is funding the new generation and supporting infrastructure needed to serve its own operations.

What AI antitrust issues do economists identify in AI infrastructure control?

Economists have identified different competitive risks that may become relevant as AI infrastructure becomes more vertically controlled.

Carl Shapiro, a University of California, Berkeley economist and former senior official in the DOJ Antitrust Division, is author of “Vertical Mergers and Input Foreclosure: Lessons from the AT&T/Time Warner Case.” In his paper, Shapiro explained that vertical-merger analysis often turns on theories of input foreclosure and raising rivals’ costs, particularly whether control over an important input can increase rivals’ costs without an outright refusal to deal.

Fiona Scott Morton, a Yale economist and former Deputy Assistant Attorney General for Economics in the DOJ Antitrust Division, authored “Digital Platforms: Market Structure and Competition.” Morton wrote in that and related Yale publications on digital-market contestability, that platform markets can be prone to concentration because of scale economies, network effects, data advantages, switching frictions, and conduct that raises entry barriers. That framework is relevant in this discussion because AI infrastructure may create similar chokepoints if access to critical inputs becomes concentrated.

As these and other economists predict, early commitments to particular energy and infrastructure pathways may influence later market structure and investment choices.

How does this deal build on prior AI competition chokepoint concerns?

Recent Mogin Law analysis discussed this trajectory. In prior commentary on Oracle, OpenAI, NVIDIA, AMD, SoftBank, and Stargate, we discussed the “race for supremacy in AI and AI infrastructure,” noting that a small group of dominant firms and global investors have pledged vast sums to projects controlling data center capacity, cloud access, and AI compute.

That analysis described Oracle’s reported role in supplying 4.5 gigawatts of data center capacity to OpenAI, and NVIDIA’s commitment to deploy up to 10 gigawatts of GPU-powered compute capacity.

We have also flagged agency concern that cloud and AI infrastructure control may become a competition issue, including in our prior posts on AI chokepoint deals, the race for GenAI dominance, and AI infrastructure transactions.

The FTC has been scrutinizing partnerships between major cloud providers and AI developers. The European and UK agencies, meanwhile, are paying attention to cloud gatekeeper power, switching costs, interoperability, and access to computing resources.

Could dedicated AI power capacity raise Section 2 concerns?

Any resulting efficiencies will be central to defending these arrangements. Microsoft and Chevron can argue that the project adds new capacity, reduces stress on the grid, supports reliability, and accelerates AI infrastructure deployment. Those are real arguments. But Section 2 of the Sherman Act asks a different question when monopoly power or attempted monopolization is alleged: whether a firm’s conduct tends to maintain or extend market power through exclusionary means rather than competition on the merits.

What does Alcoa have to do with AI infrastructure?

That concern also fits an older Section 2 lineage associated with United States v. Aluminum Co. of America, 148 F.2d 416 (2d Cir. 1945), the Alcoa decision. The issue in that case was not only whether a dominant firm could deny rivals access to a scarce input, but whether Alcoa stayed ahead of demand by continually expanding aluminum ingot capacity before rivals entered the market, making entry appear uneconomic because Alcoa already stood ready to serve the future market.

That theory would face a more skeptical reception under modern antitrust doctrine, which is reluctant to condemn investment, output growth, or capacity expansion absent clearer evidence of exclusionary conduct.

Modern courts would likely require a more specific showing of exclusionary effect. But Alcoa remains useful as a warning about market preemption. Capacity can protect a monopoly not only when a firm locks up existing scarce inputs, but also when it creates or controls so much future capacity that rivals have little incentive or practical ability to build their own.

What legal theories could apply to AI competition power and data-center capacity?

A litigation-focused theory would likely examine both sides of that capacity problem.

One familiar theory is input foreclosure. Long-term, dedicated power arrangements by dominant AI or cloud firms may foreclose scarce energy capacity needed by rivals. This would raise rivals’ costs for compute-intensive AI services; reinforce existing advantages in cloud, model training, and enterprise AI deployment; or create a cumulative infrastructure moat when combined with preferred access to chips, cloud customers, data, and capital.

A related Alcoa-style theory looks different. It asks whether dominant firms are building or reserving so much dedicated AI infrastructure ahead of demand that future rivals cannot justify building competing capacity of their own. The issue would not be the Chevron deal in isolation. It would be whether a pattern of exclusive, long-duration, or preemptive infrastructure commitments either locks up the essential inputs of AI or occupies the future market before smaller rivals can reach scale.

Why could control of the AI physical stack spark antitrust battles?

If the next generation of AI competition depends not only on better models, but on privileged access to power, chips, and data center capacity, then control of the physical stack could spark substantial antitrust skirmishes.

In that environment, Section 2 theories of monopolization, attempted monopolization, raising rivals’ costs, and input foreclosure may become increasingly important—not because infrastructure investment is unlawful, but because infrastructure control can become the mechanism by which today’s leaders become tomorrow’s unstoppable gatekeepers.

Time is a critical factor. As we wrote earlier this month, “Today’s deal activity is shaping competitive conditions before traditional concentration metrics can even register meaningful change. Market structure is being influenced not only by who owns which assets today, but by who has secured priority access to the resources that will determine participation tomorrow.” Courts and enforcers must not be shy about stopping harmful deals.

Editor’s Note: This and more than 220 other deals are included in the interactive Mogin Law A.I. Deal Table.

Edited by Tom Hagy, Editor-in-Chief, Mogin Law Blog. Send comments or questions to Info@MoginLawLLP.com.

Takeaways 

AI competition is moving deeper into the infrastructure stack. The Microsoft–Chevron arrangement shows that power supply, not just chips, cloud platforms, and data centers, may become a competitive chokepoint for AI markets.

Infrastructure buildouts can be both procompetitive and exclusionary depending on context. New dedicated power capacity may improve reliability and expand supply, but long-term control over critical inputs can also raise rivals’ costs or make entry harder.

Older monopolization precedent helps frame the risk, but modern courts would require more. Alcoa supports the concern that dominant firms can preempt future competition by controlling capacity ahead of demand, while current doctrine would likely require a specific showing of exclusionary effect.

If you have questions contact us at Info@MoginLawLLP.com.

Members of the press are encouraged to contact us at Media@MoginLawLLP.com.

FAQ for Business Counsel