NVIDIA Corp. has reached an agreement to acquire critical technology and personnel from AI chip startup Groq in a deal valued at approximately $20 billion, according to reports and company statements released Wednesday.
The Christmas Eve deal’s unusual structure — framed as a licensing agreement rather than an outright acquisition — appears designed to avoid antitrust review.
Under the terms announced by Groq, NVIDIA will receive a non-exclusive license to the startup’s inference technology. Groq CEO Jonathan Ross, President Sunny Madra, and other employees will join NVIDIA as part of the arrangement. Groq will remain nominally independent under CFO Simon Edwards, who will assume the CEO role, though most of the company’s assets appear to be transferring to NVIDIA.
As of Thursday morning, NVIDIA had not publicly commented on the deal. Details about whether the $20 billion figure represents a lump sum payment or includes milestone-based payments remain unclear.
The transaction represents what industry observers are calling a reverse acqui-hire, a tactic that allows companies to obtain talent and intellectual property without triggering merger reviews under the Hart-Scott-Rodino Act. Microsoft Corp. employed a similar strategy in 2024 when it paid $653 million to hire key executives from Inflection AI, including Mustafa Suleyman, while leaving the startup’s corporate structure intact.
CNBC first reported the transaction Wednesday as a full acquisition. However, Groq’s subsequent statement clarified that its cloud business, GroqCloud, would continue operating independently. The discrepancy has fueled speculation that the structure is specifically intended to circumvent Federal Trade Commission oversight of NVIDIA growing market dominance in AI chips.
The deal’s rapid execution, which was reportedly completed within days, underscores both the strategic value of Groq’s technology and the effectiveness of structuring it to avoid regulatory delays that have plagued other major tech acquisitions in recent years.
Founded nearly a decade ago by creators of Google’s tensor processing unit (TPU), Groq develops specialized language processing units designed to run large language models (LLMs) efficiently. The startup was valued at $6.9 billion just three months ago in its most recent funding round, making the reported price tag nearly triple that valuation.
The acquisition signals NVIDIA’s recognition of competitive threats from custom AI chips developed by Google, Amazon.com Inc., and other tech giants. While NVIDIA has dominated the training phase of AI development with its graphics processing units, the industry is increasingly focused on inference — the phase where trained models generate responses to user queries.
Groq’s technology addresses a critical bottleneck in AI inference through its use of SRAM (static RAM) rather than the high-bandwidth memory commonly used in training chips. This approach enables faster, more predictable response times for token generation, the process by which AI models produce text outputs.
Industry analysts suggest Groq’s hardware ecosystem could replicate NVIDIA’s training-era success in the inference market, where hyperscalers like Microsoft, Amazon, and Google generate most of their AI revenue. As companies like OpenAI and Meta Platforms Inc. continue developing frontier models, demand for robust inference infrastructure has intensified.
