Key Points
- Federal authorities are examining whether Nvidia crafted its $20 billion Groq agreement to circumvent regulatory antitrust oversight.
- Announced last December, the agreement granted Nvidia licensing rights to Groq’s intellectual property while maintaining Groq’s standalone status, with its top executives transitioning to Nvidia.
- Congressional representatives have denounced the structure, arguing it functions as a disguised acquisition that threatens competitive dynamics in AI semiconductor markets.
- Groq’s latest 3 LPX rack system has entered mass production, achieving speeds of 3,400 tokens per second, with planned installation at Nebius neocloud facilities this year.
- Federal investigators may conclude their inquiry without pursuing enforcement measures, while Nvidia maintains the arrangement fosters technological advancement.
Shares of Nvidia (NVDA) declined 0.91% following disclosure that federal prosecutors are scrutinizing the semiconductor manufacturer’s $20 billion licensing deal involving artificial intelligence chip developer Groq.
Federal authorities are examining whether the arrangement was deliberately structured to evade mandatory antitrust assessment. Initial reporting from The New York Times revealed the investigation, subsequently validated by Bloomberg’s sources familiar with the confidential proceedings.
The arrangement became public last December. Its framework granted Nvidia intellectual property rights to Groq’s technological assets while preserving Groq’s operational independence. Groq’s chief executive Jonathan Ross and chief operating officer Sunny Madra both transitioned to Nvidia positions as components of the transaction.
Groq declined to file the transaction for mandatory antitrust evaluation, triggering regulatory interest. Federal prosecutors initiated their examination of the arrangement earlier this year within a comprehensive antitrust investigation targeting Nvidia.
Congressional members have criticized the transaction framework, characterizing it as a de facto acquisition designed to diminish competitive pressures within AI chip manufacturing. Nvidia has vigorously defended its position.
“The Groq story is a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers,” an Nvidia spokesperson said.
Investigators may conclude their examination without pursuing formal enforcement proceedings.
Groq Production System Launches
From a technical standpoint, Groq’s 3 LPX rack configuration has achieved full-scale manufacturing and is scheduled for installation at Nebius neocloud infrastructure before year’s end. Each liquid-cooled unit contains 256 Groq processors and integrates with Nvidia’s Vera central processing units and Rubin graphics processors.
According to Nvidia, the rack configuration processes 3,400 tokens per second, significantly exceeding the 750 tokens per second that OpenAI has indicated for its Cerebras-enabled Ultrafast capability. Samsung manufactures Groq’s semiconductor chips, contrasting with Nvidia’s reliance on Taiwan Semiconductor Manufacturing for GPU production.
This $20 billion transaction represents Nvidia’s largest acquisition by value. Bernstein’s equity analyst Stacy Rasgon indicated to CNBC that Nvidia possesses sufficient financial resources to complete a transaction of this magnitude without compromising its financial position.
Institutional Investors Maintain Confidence
Hedge fund ownership of Nvidia expanded to 285 funds during Q2 2026, rising from 275 funds in the preceding quarter. Competing chipmaker AMD attracted interest from 164 hedge funds during the equivalent timeframe.
Nebius remains the sole confirmed cloud infrastructure provider scheduled to implement the Groq rack system. Simultaneously, Nvidia is pursuing multiple significant strategic investments, including potential commitments reaching $100 billion to OpenAI and $5 billion to Intel.
The Department of Justice’s examination of the Groq licensing arrangement remains active.


