Why Nvidia's $20 billion Groq deal needed no merger filing?

In December 2025, Nvidia signed one of the largest deals in its history: roughly $20 billion for rights to a rival chipmaker's technology. No merger filing was required.

Not because the deal was too small. It was worth about 150 times the dollar threshold that normally forces a company to notify federal antitrust regulators before a deal can close. It went unreported because, on paper, Nvidia never actually bought anything.

Groq's chief executive, its president, and much of its senior engineering team joined Nvidia. Groq itself kept operating, independently, on paper. That distinction, hire the people and license the technology instead of buying the company, is why a deal this size no longer needs to look like a merger to function like one.

A Merger That Was Never Called One

The deal gave Nvidia a non-exclusive license to Groq's inference chip technology, the specialized hardware Groq built for running trained AI models quickly and cheaply. In exchange, Groq's leadership, including founder and chief executive Jonathan Ross and president Sunny Madra, moved to Nvidia along with a large share of the company's engineering staff.

Groq, as a legal entity, didn't disappear. It kept its name, its offices, and, in name, its independence. What it lost was most of the people who built it.

Why $20 Billion Didn't Need a Permission Slip

Under the Hart-Scott-Rodino Act, a deal above a set dollar threshold, adjusted each year and set at roughly $134 million for 2026, normally has to be reported to the Department of Justice and the Federal Trade Commission before it closes. That waiting period exists so regulators can object to a deal before it happens, when unwinding it is still simple, rather than years later, after two companies' operations are already tangled together.

The law's trigger isn't the size of the check. It's the legal form of the transaction: an acquisition of another company's stock or assets. Nvidia's deal was neither. It was a license, plus a set of individual job offers. Nothing about that structure obligated anyone to file paperwork, no matter how large the number attached to it was.

Did You Know?

The Hart-Scott-Rodino filing threshold for 2026 is about $134 million. Nvidia's deal was valued near $20 billion, roughly 150 times that amount, and still required no filing at all. Size alone has never been what triggers the law. Structure is.

What Was Left Behind Was Worth Far Less

Before any of this happened, Groq was already a serious company. A funding round in September 2025 reportedly valued it at $6.9 billion, built on its chip design and the cloud service it ran on top of it. Nvidia's $20 billion license, three months later, was nearly three times that figure.

The number looks like a vote of confidence in Groq's technology. What happened next says otherwise.

A few months after its founders and senior engineers left for Nvidia, Groq reportedly raised a fresh funding round at roughly $3.5 billion, about half of what it had been worth before the deal. It repositioned itself around running an inference cloud service rather than the chip architecture and leadership team that had defined it. Nvidia's $20 billion bought the parts of Groq that mattered enough to license and hire away, and what remained afterward was worth roughly half as much as the whole had been.

The Playbook Nvidia Didn't Invent

Nvidia didn't design this approach. In March 2024, Microsoft paid Inflection AI roughly $650 million combined, most of it to license Inflection's AI models, some of it explicitly so Inflection wouldn't sue over the departure of its own staff, and hired away Inflection's founders along with about 70 employees. Inflection kept existing as a company. Its leadership and technology moved to Microsoft anyway.

Three months later, Amazon ran the same play on a smaller AI startup called Adept. Amazon hired Adept's chief executive and a handful of senior staff, licensed Adept's models and datasets on a non-exclusive basis, and left roughly 20 employees behind at a company that, on paper, kept operating. A U.S. senator, Ron Wyden, publicly named the deal as exactly the kind of acquihire built to dodge antitrust review, and the Federal Trade Commission opened its own inquiry into it soon after.

Antitrust regulators noticed both deals. Britain's competition authority opened an early-stage review of Microsoft's hiring of Inflection's staff, treating it as a substitute for a takeover worth investigating on its own terms. Neither the UK review of Microsoft nor the FTC's look at Amazon produced a formal challenge, but both established the same point: a deal structured this way doesn't automatically escape scrutiny just because it escapes the filing requirement.

Nvidia's version of the same idea dwarfs both. Roughly thirty times the size of Microsoft's Inflection deal, and applied to a company that makes chips competing directly with Nvidia's own hardware rather than AI software several steps removed from Nvidia's core business. By the time Nvidia used the playbook, regulators already had two smaller precedents to compare it against, which is part of why this one drew a response so much faster.

The scale of this deal fits a broader pattern: Nvidia's rise from a graphics card company into the literal backbone of the AI industry has made nearly every major move it makes a matter of competitive concern, whether or not that move takes the legal shape of a merger.

Senators and Regulators Are Paying Attention Anyway

By March 2026, U.S. Senators Elizabeth Warren and Richard Blumenthal had sent Nvidia chief executive Jensen Huang a letter asking for details on the deal, stating plainly that its structure appeared designed to evade antitrust review. By September 2026, the Justice Department had opened its own investigation into the same question: whether Nvidia built this deal specifically to avoid the filing it would otherwise have triggered.

Neither step blocks the deal. Both signal that a filing requirement being technically satisfied, or in this case, technically not triggered, doesn't end the conversation once the deal is large enough and public enough to attract attention on its own.

A Closed Deal Is a Much Harder Deal to Undo

Antitrust agencies keep the authority to challenge a completed transaction after the fact if they conclude it substantially reduces competition, filing or no filing. In practice, that authority is far weaker than the pre-closing review the Hart-Scott-Rodino process exists to provide.

Blocking a merger before it closes means telling two still-separate companies to stay separate. Unwinding one after the fact means untangling people, technology, and contracts that have already merged into daily operations, sometimes years into that process.

The pre-merger filing system exists precisely because that difference is so large. A deal that closes without triggering it gets the practical benefit of that gap, whether or not a challenge eventually follows.

Nvidia is reportedly also facing a separate Justice Department inquiry into whether it uses its dominant position in AI chips to pressure customers against buying from competitors, a wholly different antitrust question running in parallel with the Groq deal review.

Knowlegic Perspective

Antitrust law built around the merger, a formal, filed, reviewable transfer of ownership, assumes that consolidation happens through paperwork regulators get to see in advance. The AI industry's most valuable asset increasingly isn't a company's balance sheet. It's the small number of people who know how to build a given piece of technology, and a license is enough to get most of the practical value without ever filing the form that ownership would require.

That gap between what the law was built to catch and how the most valuable AI deals now actually work isn't unique to Nvidia. Microsoft found it first, at a fraction of the scale, and walked away from an early-stage inquiry with no real consequence. Nvidia's version is large enough, and close enough to its own core market, to force regulators to decide whether that gap needs closing or whether it's simply how AI consolidation works now.

Whatever the Justice Department concludes, the pattern itself has already answered a different question: a deal this size no longer needs to look like a merger to function like one.

Sources & References

• DOJ Probes Nvidia's $20 Billion License Deal With Groq on Antitrust Concerns, Bloomberg (2026)
• Nvidia's $20 Billion Groq Deal Queried by Warren, Blumenthal, Bloomberg (2026)
• FTC Announces Hart-Scott-Rodino Thresholds for 2026, Jackson Walker (2026)
• Big AI Acquihire: Microsoft, Inflection, Amazon, Adept and Antitrust, Fortune (2024)
• Amazon's deal with AI startup Adept faces FTC scrutiny, CNBC (2024)
• Down-round for Groq: AI Chip Pioneer Turns Into an Inference Cloud, Valuation Cut in Half, Trending Topics (2026)