Ebbs and Flows of Merger Control Powers in the US and the EU

The most striking feature of the current moment is that both the US and EU are experiencing the same underlying dynamic: sub-federal actors are asserting merger review jurisdiction using legal tools that were either newly created or newly interpreted to fill gaps left by central-level enforcement.

US/EU Background

Historically, merger control in the US was a federal duopoly between the FTC and DOJ Antitrust Division through the HSR framework. State attorneys general had concurrent authority under the Clayton Act but lacked early access to deal information, forcing them to rely on subpoenas or waiting for federal agencies to share materials. That asymmetry made meaningful independent state review rare except in high-profile cases.

The EU structure is more formally centralized and based on the one-stop-shop principle under the EU Merger Regulation. That means deals above the jurisdictional thresholds are reviewed exclusively by the Commission, removing member state jurisdiction entirely. But below those thresholds, member states retain their own merger control regimes, and Article 22 EUMR has allowed them to refer deals up to Brussels.

Mini HSR and State powers in the US

The mini-HSR wave changes the US architecture. Washington and Colorado now require simultaneous submission of HSR filings to state AGs, with California following in 2027 and several other states in various stages of adoption. The laws are expressly non-suspensory — no additional waiting period, no independent veto — but they deliver something structurally significant: real-time informational parity with federal agencies, at the moment of filing, across an expanding number of jurisdictions.

The practical effect is a multiplication of review eyes. State AGs can now identify transactions of state interest from day one of the federal clock, coordinate with each other through the reciprocal sharing provisions, and — critically — build independent litigation cases in state court under state antitrust law, as Colorado and Washington both demonstrated in Kroger/Albertsons. The filing requirement does not create new substantive powers, but it dramatically reduces the informational disadvantage that previously made state enforcement reactive rather than concurrent. Ironically, the FTC's own information-gathering capacity at the federal level has been clipped precisely as state capacity is expanding.

The new merger control powers of state agencies in the EU

The Illumina judgment was a hard constraint on the European Commission: it ruled that the Commission could not accept referrals of deals over which no member state had actual jurisdiction, foreclosing the strategy of accepting referrals of deals that triggered no national notification anywhere. The Commission's response has been to work through member states rather than around them — encouraging national legislatures to enact call-in powers that create jurisdiction over below-threshold deals on qualitative grounds, then referring those deals upward. The Run:ai/Nvidia case is the first significant legal test of whether that workaround survives scrutiny. Nvidia's core argument — that Italy's call-in jurisdiction was itself created after the deal was signed, making the resulting referral a bootstrapped expansion of Commission power — maps precisely onto the Illumina logic. The Commission's counter — that it is simply accepting the jurisdiction a member state legitimately created under its own sovereign legislative authority — frames the issue as one of national autonomy rather than Brussels overreach.

The Trend

The trend is the decentralization of information access combined with the fragmentation of enforcement authority, driven by dissatisfaction with the perceived gap-tolerance of centralized merger review — particularly for below-threshold tech and AI acquisitions.  The response on both sides of the Atlantic has been to expand the informational perimeter of merger review.

Mini-HSR laws give states the information without the veto. Call-in powers give member states nominal jurisdiction that the Commission can then absorb. In both cases, the architecture is additive rather than replacing the central framework. Merger control is becoming a multi-level, multi-actor system in both jurisdictions, with the center no longer reliably holding exclusive authority over deal scrutiny.

For deals with US and EU nexus, the practical consequence is a growing pre-filing engagement obligation across more jurisdictions, earlier in the deal timeline, against less predictable triggering standards — particularly on the EU side where qualitative call-in criteria introduce genuine ex ante uncertainty.

The Antitrust Federal Agencies seek input on HSR rule post-vacatur

The FTC and DOJ launched a joint public inquiry on March 25, requesting public comment on the effectiveness of the updated HSR form and soliciting feedback on potential improvements to the Hart-Scott-Rodino Act (HSR) rules, which require companies to notify the government before certain transactions. A federal court struck down its new version, restoring the old rules for now.

Background

The “1978 Rule” is the first implementation of the Hart‑Scott‑Rodino (HSR) Act’s premerger notification requirement. The FTC created a standardized Form that parties had to fill out before closing certain mergers. The court emphasized that the 1978 Form asked for a finite, transaction‑focused set of information: who the parties are, what is being acquired, how much, in what industries, and where they overlap, plus key financial and competition‑relevant documents.

The Court's reasoning

The Eastern District of Texas opinion in Chamber of Commerce of the United States of America, et al. v. Federal Trade Commission drew a contrast with the FTC’s new, vacated HSR rule. The point is: historically, the HSR Form has asked for information that (1) is closely tethered to the statutory text and purpose (premerger review to detect anticompetitive effects), and (2) consists largely of existing business information and ordinary transaction documents, not open‑ended policy, compliance, or qualitative narrative submissions. By laying out the 1978 Form item‑by‑item, the court set up an argument about what is “in bounds” for HSR information‑gathering and why the newer rule goes beyond that traditional, statutorily grounded scope. The opinion holds that the FTC exceeded its statutory authority under the HSR Act and that the Final Rule was arbitrary and capricious, particularly due to inadequate cost–benefit analysis and failure to consider less burdensome alternatives.

Practical Impact

Filers may now use the prior HSR form and classic requirements (including the traditional Item 4(c)/4(d) framework) and are not required to provide the expanded narratives, data, and documents the 2024 rule demanded. The FTC has stated it will accept filings on the old form and will also accept voluntary use of the new form from parties that have already prepared it, but there is no obligation to do so. Jurisdictional thresholds and filing fees for 2026 are unchanged; only the expanded content requirements were vacated. Filings made under the new form while it was in effect remain valid; the vacatur does not retroactively invalidate or prejudice those submissions. The FTC's appeal to the Fifth Circuit remains pending on the merits, though on March 19, 2026, the Fifth Circuit denied the FTC's motion for a stay, making the vacatur effective immediately. Additionally,  the FTC may still request information previously required by the vacated rule during the initial waiting period by issuing voluntary access letters. So the practical burden reduction is real but not absolute.

The Public Inquiry

The inquiry goes well beyond relitigating the vacated rule. The Agencies are considering changes to address non-traditional transaction structures, including acquihires, reverse acquihires, transactions involving convertible securities, and certain licensing arrangements. There's also the AI angle: the Agencies are seeking input on the costs associated with requiring filers to disclose their use of artificial intelligence or generative AI tools in preparing HSR submissions. The Agencies are explicitly considering engaging in a new rulemaking process to make necessary updates. Old form operative now, but new rulemaking could emerge in parallel with the appeal.

The FTC scrutiny of AI Input Concentration and the Regulatory Framework

The FTC is sharpening its competition strategy around AI by focusing on two fronts: control over key AI inputs (chips, compute, data) and anticompetitive “acqui-hire” practices that can entrench dominant tech firms.  The FTC looks at whether deals could let a few players control the  essential inputs for developing and scaling AI systems.

The concern about talent acquisition

The FTC is particularly worried about transactions structured primarily to acquire teams of experts rather than products or entire companies. There have been cases where startup founders received large payouts to shut down their companies and join dominant tech platforms, allowing those incumbents to consolidate power and remove nascent competitors. Economic studies show that this behavior can undermine innovation, destabilize job markets, and reduce consumer benefits.

What are the cases the FTC is likely referring to?

Companies might design deals to bypass antitrust review under the Hart-Scott-Rodino Act. Public reporting points to the type of transactions the FTC may have in mind: deals structured as licensing or similar arrangements while bringing over founders or key teams from AI startups. 

Microsoft's 2024 deal with Inflection AI, where Microsoft agreed to pay about $650 million in cash to Inflection AI. Roughly $620 million was for a non-exclusive license to Inflection’s AI models and technology.  Microsoft brought on Inflection’s co-founders and most of its approximately 70 employees into a newly formed “Microsoft AI” division. Inflection itself was not formally acquired, and it retained its intellectual property.

Meta’s hiring of Scale AI’s CEO. Meta agreed to invest about 14.3 billion dollars for a 49 percent, non‑controlling stake in Scale AI. Scale remained a separate company, and Meta secured priority access to Scale’s data capabilities and its top talent without a full acquisition. As part of the same arrangement, Scale’s founder and CEO left to join Meta’s “superintelligence” AI team, with a small group of Scale staff also moving over.

Amazon hired Adept AI’s co‑founder and CEO, plus all or most of the other co‑founders and several key employees, into its AGI/“AGI Autonomy” team. Amazon entered a non‑exclusive licensing deal for Adept’s agent technology, models, and datasets, giving it access to the startup’s core tech without a full acquisition while Adept continues as a separate company under a new CEO.

Nvidia agreed to a large, non‑exclusive license for Groq’s AI chip technology. Simultaneously, Nvidia hired Groq’s founder and CEO, its president, and a large share of its engineering team, while Groq remains an independent company with a new CEO and an ongoing cloud business.

The enforcement message

The message is clear: the FTC is looking beyond deal structure to assess whether transactions reduce competition or hinder innovation — especially as AI talent becomes a key asset in the technology race.

NVIDIA-OpenAI partnership: no control over OpenAI, so what is NVIDIA buying?

NVIDIA announced in September 2025 its intention to invest up to $100 billion in non-voting, non-controlling OpenAI equity, staged as capacity is deployed. The deal offers NVIDIA equity upside but no governance rights or board seats. On the other side, OpenAI would commit to leasing NVIDIA-powered compute capacity. The arrangement remains contingent on the future deployment of large-scale infrastructure, and it isn't finalized, but it raises some interesting questions.

Circular spending?

The NVIDIA–OpenAI arrangement resembles a capital-backed purchasing loop. NVIDIA injects huge equity capital; OpenAI spends enormous sums (through hyperscalers) on NVIDIA GPU-based infrastructure; OpenAI becomes a massive long-term customer of NVIDIA GPUs, driven by multi-GW deployments. NVIDIA benefits from both sides of the transaction: upside from OpenAI’s equity value, revenue from massive GPU sales to Microsoft, Oracle, etc., to serve OpenAI’s demand.

The FTC highlighted risks about circular spending in AI-cloud service provider partnerships in January 2025 (Microsoft-OpenAI, Amazon-Anthropic, Google-Anthropic). However, it did not assert that these partnerships violated competition laws; it merely described potential concerns. The agreements frequently include “cloud-commitments” (AI developers pledge to spend a large portion of the CSP’s investment back on that CSP’s cloud services) and the sharing of key resources. Overall, these terms often create deeply intertwined commercial and technical relationships between the cloud provider and the AI developer.

OpenAI computing infrastructure reality and the incentive alignment loop

Unlike the cloud-provider structures the FTC examined—where the investor directly operates the infrastructure—NVIDIA's loop runs indirectly. OpenAI is not a hyperscaler. It doesn't own data centers—it relies on partners to build, own, and operate the physical infrastructure. The GPUs OpenAI uses sit in Microsoft Azure, Oracle, CoreWeave, and AWS data centers. These companies purchase the chips from chip suppliers such as NVIDIA or AMD.

Even though OpenAI doesn’t buy GPUs directly, the OpenAI-NVIDIA partnership creates a near-guaranteed, long-term customer, with a deep, multilayered alignment of financial incentives: for OpenAI to grow, it must train and deploy progressively larger frontier models; that requires massive GPU purchases by hyperscalers (Azure, Oracle, AWS), who overwhelmingly buy NVIDIA accelerators. While not all of NVIDIA's equity injection will necessarily flow to GPU purchases, the structure creates a strong bias toward NVIDIA-heavy deployment, effectively aligning the incentives across all three layers: all benefit financially as NVIDIA-powered compute ramps up.

Market foreclosure  and entrenchment risk

OpenAI's commitment to NVIDIA-powered infrastructure creates a demand floor for NVIDIA in a market where OpenAI's ChatGPT is estimated to command the leading share of global AI chatbot traffic, and NVIDIA is the current dominant GPU vendor, leveraging its CUDA ecosystem. The concern is that rivals will be relegated to residual demand segments.

AMD secured a separate 6 GW commitment from OpenAI, which partially mitigates the foreclosure risk. However, with current public announcements indicating at least 10 GW of NVIDIA-based systems and at least 6 GW of AMD capacity, this implies a majority NVIDIA share in OpenAI's near-term deployment roadmap.

Leaving NVIDIA as the majority provider of OpenAI’s near‑term compute, with CUDA as the core abstraction that OpenAI’s highest‑value workloads depend on, can reinforce Nvidia’s position as the default AI infrastructure layer, making it harder for AMD, custom ASICs, or open, vendor‑neutral stacks to move beyond supplementary or hedging roles. Additionally, given OpenAI's existing financial backing and hyperscalers' willingness to invest in AI infrastructure, the $100 billion appears less about filling a capital gap and more about cementing mutual incentives and ensuring supply priority.

Capitalism without risk is antithetical to innovation

Via the NVIDIA-OpenAI partnership, NVIDIA is essentially attempting to buy future revenue. The staged, contingent nature of NVIDIA's investment—tied to actual infrastructure deployment—shifts significant demand risk back onto OpenAI and the hyperscalers. At the same time, NVIDIA captures upside through both equity and near-term GPU sales. In other words, NVIDIA effectively de-risks its position relative to a conventional merger or long-term supply contract.

The antitrust laws are in place to keep markets open to competition, which, in turn, means no guarantee of revenue for market participants.

The NVIDIA-OpenAI deal structure functionally mimics aspects of vertical integration between an upstream input provider and a downstream AI developer, even though corporate control remains separate. The deal structure, based on the information available, seems to aim to achieve the perfect incentive alignment that comes with acquiring control, without bearing the risk of such an acquisition.  As the alignment of financial incentives increases lock-in, path dependence, and switching barriers, it might affect competitors such as AMD, NVIDIA's most direct rival, new entrants, and alternative architectures. For example, several startups have raised substantial capital to challenge NVIDIA, and the major cloud providers are designing their own chips. Deals like NVIDIA-OpenAI that lock in demand could make it harder for these entrants to compete, even if they have technically viable products.

Conclusion

The structure does not, on current public information, fit neatly within traditional antitrust frameworks. But the deal raises an important policy question: whether financial arrangements that achieve outcomes similar to those of vertical integration should be subject to the same scrutiny, even when they lack traditional indicia of control.

Minority Investments & Merger Control: Focus on Big Tech & AI

U.S./UK/EU policy and enforcement: convergence, misalignment, and gaps

Google/Anthropic, Microsoft/OpenAI, and Meta/ScaleAI deals share a common feature: the perceived regulatory gap in merger control thresholds and triggers. Big Tech is gaining effective control over AI startups. Still, regulators are struggling to apply the current regulatory framework to these deals because they do not confer control over the invested entity or otherwise meet the regulatory requirements. 

There seem to be two main issues at two different levels:

Level n. 1) - The deals do not meet the jurisdictional requirements for merger control.

Level n. 2) - Absent control, regulators cannot prove the causal connection between the deal and the effects on competition.

This initial report focuses on level n. 1).

Let's first unpack the different types of deal structures faced by regulators so far:

1) Partnerships in combination with equity (Google/Anthropic)

2) Profit sharing and exclusivity (Microsoft/OpenAI)

3) Non-voting shareholding (META/ScaleAI)

The jurisdictional requirements for merger control in the UK, the U.S., and the EU

The UK material influence is the lowest level required to trigger merger control. The agency looks for evidence of the ability to materially

influence the invested entity's behavior in the marketplace and its ability to meet its commercial objectives.

The U.S. and the Hart-Scott-Rodino (HSR) Act gap: an investment that provides no voting rights is typically not considered a reportable event under the HSR Act, regardless of its value.

In the EU, the European Commission (EC) has no jurisdiction to review a merger when the transaction does not confer decisive influence over another business.

Google/Anthropic

The CMA in the UK launched a formal inquiry into whether Google’s investments in Anthropic amount to a relevant merger situation under UK merger control law and concluded that material influence was not established.

The FTC in the U.S. is not acting under the Clayton Act and the HSR. Section 6(b)  of the FTC Act, which empowers the agency to conduct studies, issue orders to firms for information, and scrutinize business practices for broader market impacts—even if the deals are not formal mergers or acquisitions.

No investigation under the European Union Merger Regulation (EUMR).

Microsoft/OpenAI

The CMA conducted a Phase 1 review; ultimately, it decided not to open a Phase 2 investigation because the transaction did not meet the legal standards for review.

No investigation under the Cayton Act and the HSR. The FTC issued demands for information from Microsoft, OpenAI, and others about whether the structure circumvents merger control.

The EC considered the transaction not notifiable under the EUMR.

META/ScaleAI

No public statements of formal proceedings by the CMA, the FTC/DoJ, or the EC so far.

Minority investments in AI firms may face heightened antitrust scrutiny; businesses should review deal structures and disclosure strategies carefully.