Why did Nvidia spend $20 billions to acquire Groq?

Nvidia is spending approximately $20 billion to "buy" Groq through a technology licensing deal, with the core intention of eliminating a potential threat in the efficient, low-cost AI inference chip sector and directly absorbing a top-tier team to address its own technological shortcomings. This move is not only a defensive acquisition against competitors but also a key strategic deployment to use its abundant cash to build a wider moat and consolidate its absolute market leadership.

Written by: Ye Huiwen

Source: Wallstreetcn

Nvidia caused a stir in Silicon Valley on Wednesday, agreeing to pay about $20 billion to obtain a technology license from startup Groq and hire its core team.

This massive deal is not only aimed at consolidating Nvidia’s dominance in the field of AI inference computing by acquiring Groq’s proprietary technology, but also adopts a special transaction structure to circumvent increasingly stringent antitrust scrutiny.

According to a person involved in the deal, the specific form of the transaction is a non-exclusive technology license, and Nvidia will simultaneously hire Groq’s founders and executives. Although the details of the deal have not been fully disclosed, the scale of about $20 billion is already about three times Groq’s $6.9 billion valuation from a few months ago. Through this move, Nvidia aims to acquire more cost-effective and faster chip design capabilities to meet the growing demand for running AI applications.

Nvidia CEO Jensen Huang clarified the strategic intent of this deal in an internal email to employees. He stated that he plans to integrate Groq’s low-latency processors into Nvidia’s AI factory architecture, thereby expanding platform capabilities to serve a broader range of AI inference and real-time workloads. This means Nvidia is trying to address its efficiency shortcomings in inference chips, beyond its extremely expensive high-performance training chips.

The structure of this deal is very similar to the model adopted by Microsoft, Amazon, and Google over the past two years, namely, using a "license technology + hire talent" approach to fly under the regulatory radar without formally acquiring the company.

This move not only gives Nvidia access to key intellectual property and talent, but also reflects how the world’s most valuable company is using its $60 billion cash reserves to accelerate the building of defensive barriers in the face of challenges from competitors like Google’s TPU.

Addressing the Inference Shortcomings

The core driving force behind this deal is Nvidia’s battle for the AI inference market.

Although Nvidia’s GPUs and supporting software hold an absolute dominant position in AI model development and training, its existing chips are often too large and expensive for running practical applications such as chatbots (inference). The market has long been seeking cheaper and more efficient alternatives, and Groq’s technology was born for this purpose.

This licensing arrangement gives Nvidia access to Groq’s intellectual property. Groq claims its chips outperform Nvidia’s in data processing speed for specific AI application tasks. In contrast, while Nvidia’s chips maintain flexibility in handling various types of operations, there is room for optimization in processing speed and latency.

Dylan Patel, chief analyst at chip consultancy SemiAnalysis, pointed out that although Groq’s first-generation chips have not yet formed strong competition with Nvidia, its next two generations of products are about to launch. He believes Nvidia may have seen a threat in Groq’s new generation of technology and thus decided to act.

Special Structure of "License + Talent Acquisition"

This deal is not a traditional full acquisition. Groq founder Jonathan Ross, president Sunny Madra, and other employees will join Nvidia to "advance and expand" the licensed technology. Meanwhile, Groq’s original cloud business will remain within the company, with CFO Simon Edwards, who joined in September, becoming the new CEO and continuing operations.

This non-exclusive licensing deal structure is a common way for tech giants to circumvent regulatory scrutiny recently.

Microsoft, Amazon, and Google have all used similar structures to absorb founders and core technology from AI startups without formally acquiring the companies. Although Google’s similar deal with Character.ai once triggered a review by the U.S. Department of Justice, no action was taken. Currently, Nvidia is not facing antitrust scrutiny in the U.S., but has always been cautious in describing its market share in the AI chip sector.

According to sources cited by The Wall Street Journal, as a result of the licensing agreement, Groq’s investors (including BlackRock and Tiger Global Management) will receive returns, including installment payments based on future performance. This deal is similar to Nvidia’s transaction with networking startup Enfabrica three months ago, when Nvidia spent over $900 million to hire the company’s CEO and engineering team and paid a technology licensing fee.

Nvidia’s Unshakable Ecosystem

Despite receiving billions of dollars in venture capital, challengers like Groq have always found it difficult to break Nvidia’s tight control of the high-end AI chip market. Nvidia’s chips, thanks to their proprietary CUDA programming language ecosystem, have created extremely high customer stickiness.

Groq’s recent business performance also reflects the difficulty of challenging a giant. The company recently lowered its 2025 revenue forecast by about three-quarters. A Groq spokesperson said this was due to a lack of data center capacity in regions where chips were planned to be deployed, causing some revenue expectations to be delayed until next year. In July, Groq had expected its cloud business to generate more than $40 million in revenue this year, with total sales exceeding $500 million.

Meanwhile, the competitive landscape is intensifying. Google’s TPU is becoming a strong competitor to Nvidia’s GPU, and major companies such as Apple and Anthropic have used TPU to train large models. In addition, Meta and OpenAI are also developing their own dedicated inference chips to reduce reliance on Nvidia. Among startups, the trend of consolidation is obvious: Intel is in deep talks to acquire SambaNova, Meta has acquired Rivos, and AMD has absorbed the team from Untether AI.

Moat Strategy with Massive Cash Reserves

Nvidia is using its massive cash reserves to consolidate its business. As of the end of October, its cash reserves had reached $60 billion. In addition to funding dozens of cloud providers and startups that exclusively buy or lease its chips, Nvidia is also starting to make larger-scale technology acquisitions.

Previously, Nvidia’s largest acquisition was the $6.9 billion purchase of Mellanox in 2019, which has now become an important networking division for Nvidia, contributing about $20 billion in revenue last quarter.

Although the $20 billion deal with Groq is not a full acquisition, its scale far exceeds previous deals, showing that Nvidia is willing to pay a high price to eliminate potential threats and integrate cutting-edge technology in the face of increasingly specialized chip demand.