One DOJ Probe Just Put $30 Billion of Nvidia 'Reverse Acquihires' in the Crosshairs

NVIDIA announced the Groq arrangement on December 24, 2025 as a \"non-exclusive inference technology licensing agreement,\" with Groq continuing to operate independently under new CEO Simon Edwards. The deal was reported at roughly $20 billion (the DOJ's working figure, per the Times, is closer to $17 billion) and included the absorption of senior Groq talent and most of its engineering team. Senators Elizabeth Warren and Richard Blumenthal had already pressed CEO Jensen Huang in March 2026 over whether the structure was a deliberate workaround for antitrust review, citing the company's roughly 90% share of GPU shipments.

The DOJ's likely remedies run from a quiet close (least painful) through monetary fines (the Times reports unwinding the deal is unlikely) to a consent decree with behavioral remedies such as mandatory FRAND licensing, royalty-rate caps, or limits on future acquihires. The probe does not need to end in a breakup to matter — it changes the discount rate on Nvidia's entire IP-consolidation strategy.

The Groq deal is the largest line item in a playbook Nvidia has been quietly running since September 2025. Across the past twelve months the company has committed roughly $30 billion of capital through similar structures: an Enfabrica chip-interconnect license plus CEO hire (~$900 million, September 2025), the Groq arrangement (~$20 billion, December 2025), a Poolside Model Factory non-exclusive license plus equity (~$7 billion combined, mid-2026), an equity investment in SB Energy ($1.5 billion, August 2026) to anchor an OpenAI data center, a $3.5 billion convertible bond in MediaTek (August 2026), and ongoing discussions to back Perplexity at a $30 billion-plus valuation (August 2026).

The legal hook is straightforward: a license is not a change of control, and people moving between firms are not assets under HSR. Stitching the two together — IP rights to the company, talent to the acquirer — buys the economic substance of an acquisition without the filing trigger. That is the loophole the DOJ is now probing.

Nvidia is profitable enough to absorb a fine, but the optics matter for a stock trading near $224 against an analyst target of roughly $328 and a market cap of $5.42 trillion. The company posted $81.6 billion of revenue in Q1 FY2027 (period ended April 26, 2026) with data-center revenue of $75.2 billion, up 92% year over year, and a GAAP gross margin of 74.9%. Q2 FY2027 (period ended July 26, 2026) extended the run with $96.2 billion of revenue and $59.7 billion of GAAP net income. On a trailing-twelve-month basis through September 10, 2026, the company booked $303 billion of revenue, $192.9 billion of net income, and $7.95 of diluted EPS.

The rich valuation is itself a transmission mechanism. At 28.6x trailing earnings and a beta of 2.22, the stock discounts years of forward execution; any structural impairment to the IP-consolidation pipeline forces a multiple compression that growth alone may not offset.

The clearest priced consequence of the deal sits in Groq's own valuation. Three months before the December 2025 arrangement, Groq had raised a Series E at a $6.9 billion valuation. The $20 billion check from Nvidia funded large shareholder distributions, but the post-deal operating company — minus its IP, its founder, its president, and most engineers — was marked at $3.5 billion in an August 17, 2026 Series A led by Disruptive. That is roughly half the pre-deal private valuation, with planned NVIDIA participation in the new round underscoring how dependent the shell has become on the acquirer's ecosystem.

Groq has pivoted to a neocloud model — 13 data centers today, scaling from 54 megawatts to 200 megawatts in 2027 — and is now buying Nvidia GPUs rather than selling competing inference silicon. The DOJ's question is whether the original deal was structured to extinguish a competitor while preserving the appearance of an independent rival.

If the DOJ imposes remedies, the impact radiates in three directions. Upstream, ASIC designers Broadcom and Marvell — which together control roughly 95% of the custom AI accelerator co-design market — gain negotiating leverage with hyperscalers that today default to Nvidia. Broadcom carried a $73 billion AI backlog entering 2026 and is targeting $100 billion of annual AI chip revenue by 2027, with OpenAI's first custom accelerator (the \"Jalapeño\" chip) deploying in 2H 2026 on TSMC 3nm/2nm processes. Marvell is projecting up to $11 billion of AI ASIC revenue in 2026, anchored to Amazon Trainium and Microsoft Maia.

Midstream, TSMC's CoWoS allocation becomes more contested. Nvidia today absorbs roughly 60% of TSMC's advanced packaging capacity (about 595,000 wafers), with Broadcom at ~15% and AMD at ~11%. A regulatory cap on Nvidia's acquihire-driven IP build-out frees accelerator capacity for ASIC customers, modestly tightening Nvidia's foundry leverage but supporting TSMC's diversified mix. Downstream, neoclouds CoreWeave and Nebius — both of whom raise capital with Nvidia equity stakes and Nvidia GPU collateral — face renewed scrutiny over circular financing if the legal foundation of those transactions is challenged.

On a days-to-quarters horizon, the next data points are Nvidia's formal response to the DOJ information request (likely 60–90 days), the publication of any consent decree, and the next round of Senate pressure tied to the Warren-Blumenthal track record. Watch the OpenAI-Broadcom \"Jalapeño\" chip launch in 2H 2026 as the first physical proof point that ASIC supply is materially diversifying. A legal outcome that forces Nvidia to spin or re-license the Groq IP into a FRAND pool would reset the inference TAM and accelerate the ASIC mix shift already embedded in customer plans.

On a 1–3 year horizon, the structural question is whether \"reverse acquihire\" survives as a legal structure. The Poolside transaction explicitly used the same non-exclusive license template, and Perplexity talks reportedly included a tech-licensing alternative before settling on equity. If the DOJ wins the test case, every subsequent Nvidia deal above the HSR threshold faces automatic second-request review, and the implicit premium Nvidia pays for IP-and-talent bundles will compress toward market-clearing levels. The probe is small relative to Nvidia's $303 billion revenue base; the framework it sets is not.