Nvidia to Acquire Groq’s AI Chip Assets for $20 Billion in Cash

Nvidia has reached an agreement to purchase the core assets of Groq Inc., a Silicon Valley startup that develops high-performance accelerators for artificial intelligence workloads, for approximately $20 billion in cash. The transaction, confirmed by Alex Davis, chief executive of investment firm Disruptive, marks the largest deal in Nvidia’s history and underscores the company’s strategy of expanding its portfolio through targeted acquisitions and licensing arrangements.

Groq was established in 2016 by a group of former Google engineers, including Jonathan Ross, who helped create the tech giant’s tensor processing unit. The startup has focused on chips optimized for low-latency inference tasks, a segment witnessing rapid demand as enterprises deploy large language models in real-time applications. Groq most recently raised $750 million in September at a valuation of roughly $6.9 billion, attracting investors such as BlackRock, Neuberger Berman, Samsung, Cisco, Altimeter and 1789 Capital.

The newly announced transaction covers Groq’s intellectual property and engineering talent but excludes the company’s early-stage cloud service, GroqCloud, which will continue to operate independently. Jonathan Ross, company president Sunny Madra and additional senior leaders will move to Nvidia to support the integration of Groq’s architecture into the buyer’s existing platforms. Groq’s chief financial officer, Simon Edwards, will assume the role of chief executive and lead the remaining business as a standalone entity.

Nvidia is funding the purchase entirely with cash. The graphics-processing unit maker held $60.6 billion in cash and short-term investments on its balance sheet at the end of October, up sharply from $13.3 billion at the start of 2023. The acquisition eclipses Nvidia’s previous record purchase of Israeli networking specialist Mellanox Technologies for nearly $7 billion in 2019.

In an internal message to employees, Nvidia chief executive Jensen Huang said the company intends to incorporate Groq’s low-latency processors into the broader Nvidia AI Factory architecture, expanding support for inference and other real-time workloads. Huang emphasized that Nvidia is licensing Groq’s technology and onboarding personnel rather than acquiring the operating company in its entirety.

The agreement follows a pattern of similar deals across the technology sector. In September, Nvidia spent more than $900 million to hire executives from Enfabrica and secure rights to the startup’s networking innovations. Competitors such as Microsoft, Google and Meta have likewise opted to bring specialized AI teams in-house through licensing or asset purchases instead of traditional mergers.

Groq’s swift rise reflects surging demand for dedicated AI silicon. The company has set a revenue target of roughly $500 million for the current year, driven by interest from developers seeking alternatives to mainstream graphics processors. Its emphasis on deterministic performance positions the technology as an option for applications that require consistent response times.

The startup’s foundation traces back to late 2016, when it filed an initial fundraising notice with the U.S. Securities and Exchange Commission identifying Ross and Douglas Wightman—another former Google engineer—as principal officers. Wightman departed the company in 2019, according to public records.

Imagem: Internet

Nvidia’s investment activity has accelerated alongside its record earnings from data-center GPUs. The company recently backed AI-focused infrastructure firm Crusoe, language-model developer Cohere and cloud provider CoreWeave. Nvidia has also signaled an intent to deploy up to $100 billion in partnership with OpenAI, and it committed $5 billion to Intel this year to secure advanced manufacturing capacity. Analysts note that such moves reflect Nvidia’s push to reinforce its supply chain and safeguard long-term growth as competitors develop in-house solutions.

As the competition for AI hardware intensifies, several other chip startups have pursued capital markets. Cerebras Systems, which designs wafer-scale processors for generative AI training, raised more than $1 billion in private funding after pausing plans for an initial public offering. According to a filing with the U.S. SEC, Cerebras intends to revisit an offering when market conditions improve.

Nvidia’s latest deal emerged quickly, according to Disruptive’s Davis, who said Groq was not actively seeking a buyer. Financial terms of the non-exclusive license were not disclosed beyond the $20 billion cash consideration, and neither company provided a timeline for closing. Nvidia’s chief financial officer, Colette Kress, declined to comment on the transaction.

Industry observers expect the integration of Groq’s intellectual property to bolster Nvidia’s position in inference, a market segment projected to expand as generative AI systems move from training to widespread deployment. GroqCloud customers are expected to experience no service disruption during the transition, and the independent entity will continue to explore opportunities in specialized AI hardware.

Crédito da imagem: Bloomberg / David Paul