The Next Antitrust Frontier: AI Infrastructure

As hyperscalers race to acquire power, land, and AI infrastructure hardware, antitrust regulators are taking note, and they may be inclined to dust off the same playbook they used against Big Tech for the past 25 years. This article breaks down the four enforcement theories most likely to put AI infrastructure deals under scrutiny: supply banking, killer acquisitions, exclusive dealing, and vertical integration. For more on where the antitrust infrastructure boom may meet antitrust scrutiny, check out Powering Progress Episode 6, “Power, Scale, and Scrutiny,” from Vinson & Elkins.

Antitrust enforcers are watching carefully as AI infrastructure spending ramps up.^[1] Many of the big tech “hyperscalers” funding data centers are the same companies that have faced landmark antitrust scrutiny over the past 25 years. In those previous cases, regulators alleged that these companies used their market dominance to push out smaller rivals or swallow potential threats through acquisitions — sometimes called “killer acquisitions” — before those rivals could become serious competitors.

Enforcers have not yet brought any data center-related cases against hyperscalers, and from our perspective, they likely would find it difficult to do so (not least because of constant evidence of new entry, and that startups continue to get funding). But enforcers appear to be watching for any hint that established AI players may lock up supply under long-term agreements that could foreclose smaller AI rivals from essential inputs. So, despite the vigorously competitive current AI environment, industry participants should spend some time considering how antitrust investigations might arise.

This article draws on the established big tech enforcement playbook to identify the areas most likely to draw antitrust scrutiny: (1) supply banking — the practice of locking up more supply than a company immediately needs, (2) killer acquisitions, (3) exclusive dealing agreements that lock rivals out of essential inputs, and (4) vertical integration. For each area, we identify the core concern and describe what problematic conduct could look like in the AI infrastructure context. We also flag where the legal theories are well-settled and where enforcers would need to break new ground.

Supply Banking: Locking Up More Than You Need

Long-term power purchase agreements, land acquisitions, and reservations of grid interconnection capacity are normal and legal business practices. The antitrust concern arises when a dominant company signs enough of them — or signs deals larger than its operational needs — not because it requires the supply, but to prevent competitors from accessing it. Regulators call this “supply banking” or, in its more aggressive form, “predatory overbidding.”

Enforcers may view this concern as particularly acute in the AI infrastructure context because several key inputs — available land near high-capacity power grids, interconnection queue positions, and certain proprietary hardware — are allegedly scarce and not easily replicated. An enforcer may argue that a dominant hyperscaler that corners these inputs through a pattern of long-term agreements could leave rivals unable to build competitive data centers, regardless of how much capital those rivals are willing to spend.

Theory One: Foreclosure Through Aggregation

The first theory targets the cumulative effect of many supply agreements, even where no single deal is independently troubling. The argument is that a dominant company can quietly corner the market on a scarce input — power, land, interconnection — by signing enough non-exclusive long-term agreements that there is simply nothing left for rivals.

This aggregation theory draws on Section 2 of the Sherman Act, which prohibits a company from willfully acquiring or maintaining monopoly power through exclusionary conduct — conduct that harms rivals through means other than genuine competition on the merits. The leading case is United States v. Microsoft Corp.,^[2] where the D.C. Circuit held that Microsoft Corp. (“Microsoft”) violated Section 2 by entering into a web of contracts with PC manufacturers and internet service providers that, taken together, foreclosed rival browsers from the most efficient distribution channels. No single agreement was independently illegal; the violation lay in the cumulative effect.

Applying Microsoft’s cumulative-conduct framework to a buyer aggregating upstream supply agreements would be a significant extension of existing law. The government has not successfully pursued this theory in the context of input purchasing, and any such enforcement action would face genuine legal uncertainty. That uncertainty reduces — but does not eliminate — the risk of an investigation. And as noted above, an investigation alone is costly and disruptive, regardless of outcome.

Theory Two: Predatory Overbidding

The second theory is predatory overbidding. This applies when a dominant buyer exercises what antitrust lawyers call “monopsony” power — the buying-side equivalent of monopoly power. A monopsonist controls enough of the demand for an input that it can move prices and acquire supply on terms rivals cannot match. When a monopsonist buys quantities well beyond its operational needs, and where supply is scarce, the monopsonist may drive up input costs and squeeze out rivals that lack the financial capacity to compete for what remains.

In Weyerhaeuser Co. v. Ross-Simmons Hardwood Lumber Co.,^[3] the Supreme Court set a demanding standard: Predatory overbidding violates Section 2 only if (1) the overbidding drove input costs so high that the company was effectively selling its downstream product at a net loss, and (2) the company had a “dangerous probability” of recouping those losses by exercising monopoly power once rivals were forced out.

This is a demanding standard. Hyperscalers are generating substantial revenues from AI services and are not selling at a net loss, which makes the Weyerhaeuser test difficult to satisfy on its face. Enforcers might argue that the AI infrastructure context — alleged scarcity, unusually long contract durations, and a downstream market that is still forming — warrants a more flexible analysis. But any such argument would face significant headwinds in court. Businesses should treat predatory overbidding as a theory to monitor rather than a near-term enforcement priority, while remaining attentive to the investigation risk that any aggressive input acquisition strategy carries.

Killer Acquisitions

The “killer acquisition” concern is distinct from conventional merger analysis. In a standard horizontal merger, the worry is that the acquirer is eliminating a direct competitor. In a killer acquisition, the worry is that the dominant company is buying a supplier or nascent rival not because it genuinely needs what that company does, but to ensure that the target never becomes a resource for competitors — or never grows into a competitor itself.

This concern has resonance in the AI infrastructure context. A hyperscaler that acquires a promising developer of specialized AI hardware, a proprietary interconnection technology, or a key land portfolio near scarce power capacity may be doing so to prevent that target from supplying rivals — not because the hyperscaler intends to use it productively. From the enforcer’s perspective, the deal extinguishes a potential competitive constraint before it can be exercised.

Regulators will apply the 2023 Merger Guidelines’ nascent competitor and potential competition frameworks. A target’s lack of current revenues or market share is not a defense — what matters is the competitive significance of what is being acquired and what is lost when the target is removed from the market. Enforcers will ask: Would this target, left independent, have constrained the acquirer’s ability to foreclose rivals from key inputs? Would it have become a competitive supplier to those rivals?

Internal documents are the central battleground in these investigations. Emails or board presentations that frame the deal in terms of preventing a rival from accessing the target’s technology or supply relationships — rather than in terms of the acquirer’s own operational needs — are highly damaging and often fail to reflect market realities. Companies considering acquisitions in this space should be thoughtful about how they document deal rationale.

Many infrastructure acquisitions — particularly of asset-rich, revenue-light targets like a specialized technology developer or a land assemblage — will fall below the Hart-Scott-Rodino Antitrust Improvements Act of 1976’s (“HSR Act”) reporting thresholds that trigger mandatory pre-merger review. That does not mean they escape scrutiny. Both the Federal Trade Commission (“FTC”) and DOJ have demonstrated a willingness to investigate and challenge transactions after they close, using their authority to conduct post-consummation reviews. The FTC has initiated retrospective investigations into acquisitions that fell below HSR Act thresholds in the tech sector, most notably in its 2020 study of unreported acquisitions by Amazon.com, Inc., Apple Inc., Facebook, Inc., Google, and Microsoft over a ten-year period.

For hyperscalers, the implication is clear: A deal’s size in dollar terms is not a reliable guide to its antitrust exposure. A $200 million acquisition of a critical cooling technology developer or a key interconnection asset may draw more regulatory scrutiny than a $2 billion acquisition of a conventional target.

Exclusive Agreements

Exclusive dealing arrangements are a longstanding target of antitrust enforcement. The concern is straightforward: If a dominant company signs contracts with key suppliers that prevent those suppliers from doing business with rivals, it can harm competition. In the AI infrastructure context, that means contracts with power utilities, land sellers, interconnection providers, or hardware manufacturers that lock those parties up and leave rival data center builders with nothing to buy.

Unlike the supply banking theories discussed above, exclusive dealing claims do not require proof that a pattern of agreements cumulatively forecloses competition. A single agreement may come under scrutiny if it locks up a competitively significant share of a necessary input.

Enforcers also look beyond contracts that say “exclusive” on their face. Provisions that achieve the same practical effect are equally at risk, including:

• Volume commitments that consume a supplier’s practical capacity, leaving it with nothing to offer anyone else;
• Most-favored-nation (“MFN”) clauses that require the supplier to offer the hyperscaler the best price and terms it gives anyone — which can deter the supplier from offering rivals better terms and effectively pricing them out; and
• Rights of first refusal that allow a hyperscaler to match any competing offer before the supplier can finalize it — which discourages rivals from investing in supplier relationships they may never be able to close.

Exclusive dealing claims can be brought under several legal theories, each with a different standard of proof.

Under Section 1 of the Sherman Act, exclusive dealing is analyzed under the rule of reason: a fact-intensive inquiry that weighs the agreement’s anticompetitive effects against its legitimate business justifications. Courts will consider market foreclosure, the availability of alternatives, and the duration of the restriction, and then ask whether pro-competitive rationales outweigh the harm. A well-documented business justification — supply security, investment recovery, quality assurance — can be decisive. This is a genuinely contested analysis, and the burden of proof rests with the government. As a result, enforcers tend to not bring Section 1 cases unless the conduct at issue is clearly “unreasonable.”

Under Section 2, the inquiry shifts from the specific arrangement to the broader pattern of conduct. The question is whether exclusive dealing is part of a deliberate strategy by a company that has — or is in the process of acquiring — monopoly power to exclude rivals rather than compete on the merits. Unlike the rule of reason’s structured balancing test, Section 2 does not require a formal weighing of competitive harms against benefits; courts nonetheless ask whether the conduct can be explained by a legitimate business justification or genuine efficiency, or whether its real purpose and effect is simply to suppress competition.

The FTC can also proceed under Section 5 of the FTC Act, which prohibits “unfair methods of competition” and has been interpreted to reach conduct that does not meet the full Sherman Act standard. Section 5 gives the FTC broader discretion than the Sherman Act: It can capture incipient competitive harms before they ripen into full monopolization, and it does not require the FTC to prove all of the elements of a Section 1 or Section 2 claim. In practice, the FTC more commonly invokes Section 5 as a supplement to Sherman Act theories rather than a standalone basis for most enforcement actions, though its precise scope remains contested.

As one example of the FTC using Section 5 for exclusive dealing enforcement in an input market, in 2000, the FTC challenged a pharmaceutical manufacturer’s long-term exclusive supply agreements for an essential drug ingredient that effectively foreclosed rival drug manufacturers from access to that input. The FTC found that the contracts, while individually defensible, collectively locked up the market and constituted an unfair method of competition under Section 5. That precedent is directly relevant to exclusive infrastructure agreements in the AI context.

More broadly, the court found in the DOJ’s recent case against Google that Google’s exclusive or default agreements with device manufacturers and wireless carriers — requiring them to use Google Search as the default search engine — constituted illegal monopolization under Section 2. While that case involved distribution rather than supply inputs, the underlying logic — that exclusive arrangements by a dominant company can foreclose rivals from essential channels — may also apply to a dominant hyperscaler locking up essential infrastructure inputs.

Vertical Integration and Foreclosure

Vertical integration occurs when a company acquires a supplier (or customer) that operates at a different level of the same supply chain — for example, a company that sells AI services buying the business that makes a key component those services depend on. Antitrust enforcers do not generally oppose vertical integration, but they may raise concerns that the integrated company might use its control over that input to choke off rivals by denying them access, offering it only on degraded terms, or locking up enough supply that competitors cannot get what they need to compete.

The “ability-and-incentive” framework is the tool regulators use under the Merger Guidelines to assess whether a proposed vertical merger is likely to harm competition. Because the merger has not yet closed, the inquiry is necessarily forward-looking: Regulators must predict how the combined company will behave once it controls both the input and the downstream product that depends on it. The framework asks two questions. First, does the combined company have the ability to foreclose rivals? In practice, this means the acquired input is competitively significant, rivals cannot readily substitute around it, and the combined company controls enough of the supply to make foreclosure meaningful. Second, does the combined company have the incentive to foreclose? This means the downstream gains from damaging rivals (higher prices, greater market share in AI services) outweigh the upstream revenue the company sacrifices by withholding supply from paying customers. Both prongs must be satisfied. Enforcers recognize that a company that could foreclose rivals, but would lose more than it gains by doing so, is unlikely to act.

As an example, imagine that a dominant hyperscaler acquires the leading developer of proprietary high-density cooling technology that is essential for operating the gigawatt-scale data centers needed to train frontier AI models, and for which (hypothetically) no commercially viable substitute yet exists. Rivals denied access to that technology face a hard physical constraint on the size and density of the data centers they can build — regardless of how much power or land they secure. Antitrust enforcers will assess whether the buyer could use that advantage to foreclose its competitors from the market.

Courts rarely block vertical transactions, however. Vertical integration often generates real efficiencies — more secure supply chains, lower coordination costs, better product integration — that courts weigh against the alleged foreclosure harm. The government’s recent record reflects that difficulty. The FTC challenged Microsoft’s acquisition of video game developer Activision Blizzard, Inc. on vertical foreclosure grounds, arguing that Microsoft could use its ownership of Call of Duty to harm rival gaming platforms, and lost in court.^[4] The FTC also challenged Illumina, Inc.’s (“Illumina”) acquisition of GRAIL, Inc., arguing that Illumina could use its dominant position in DNA sequencing to harm rival cancer-test developers. The FTC ordered divestitures, but the United States Court of Appeals for the Fifth Circuit vacated that order in 2023, finding the agency had applied the wrong legal standard for assessing potential competition.^[5] Though the FTC prevailed at the agency level, the appellate loss illustrates the genuine difficulty of winning vertical merger challenges in court. These results reflect courts’ general reluctance to block vertical deals absent compelling evidence of foreclosure.

That said, the FTC and DOJ have shown a clear appetite to investigate vertical transactions, and an antitrust investigation — even one that does not end in a lawsuit — is disruptive, expensive, and time-consuming.

Merger clearance does not immunize retrospective merger reviews, either. A hyperscaler whose transaction clears review under the HSR Act — or that falls below the HSR Act reporting threshold altogether — may still face antitrust exposure for the transaction itself months or years after it closes. In fact, some of the leading hyperscalers have faced post-closing merger reviews in other product areas — the FTC’s attempt to force Meta to unwind its acquisitions of WhatsApp and Instagram years after those transactions closed is one such example.

Conclusion

The antitrust risks surrounding AI infrastructure are not yet fully formed. The legal theories range from well-established (exclusive dealing, vertical foreclosure in merger review) to genuinely novel (cumulative supply aggregation) and include areas in which the government’s recent track record is poor (vertical merger challenges). But enforcement interest is high. Hyperscalers making long-term commitments to power, land, and interconnection capacity — and pursuing vertical integration and acquisitions to secure their infrastructure position — should treat antitrust risk analysis as an integral part of their infrastructure strategy, not an afterthought. That means assessing the competitive significance of key inputs before acquiring or locking them up, documenting legitimate business rationales carefully, and engaging antitrust counsel before deals are structured rather than after they are signed.

^[1] Despite the urgency in the media and the financial markets, enforcers have raised only general concerns without outlining clear enforcement priorities for AI infrastructure. Jonathan Kanter, who led the DOJ Antitrust Division under President Biden, expressed concerns about chokepoints in AI infrastructure, and Gail Slater, his successor at the outset of President Trump’s second term, emphasized effective antitrust enforcement in the electricity sector as critical to domestic AI growth.

^[2] 253 F.3d 34 (D.C. Cir. 2001).

^[3] 549 U.S. 312 (2007).

^[4] Fed. Trade Comm’n v. Microsoft Corp., 136 F.4th 954 (9th Cir. 2025).

^[5] Illumina, Inc. v. Fed. Trade Comm’n, 88 F.4th 1036 (5th Cir. 2023).

This information is provided by Vinson & Elkins LLP for educational and informational purposes only and is not intended, nor should it be construed, as legal advice.