Catch and kill? Why Nvidia’s record $20 billion deal with AI chip startup Groq matters

Nvidia has reportedly reached its largest-ever deal, a $20 billion licensing and talent acquisition agreement with AI chip startup Groq (not to be confused with the AI bot Grok from Elon Musk’s xAI). The deal smells like an attempt to gobble up a potential competitor, part of the “catch and kill” moves Silicon Valley’s big tech industry is notorious for. What is the deal, and why does it matter for one of the world’s top AI chipmaking giants? Here is what you should know.

What is in the Groq deal with Nvidia?

Started by former Google engineers in 2016, Groq makes LPU (Language Processing Unit) chips for ultra-fast, low-cost AI inference. The reported $20 billion all-cash deal includes asset purchases and non-exclusive technology licensing. These assets include Groq’s AI inference technology and intellectual property.

Groq’s founder and CEO Jonathan Ross, president Sunny Madra, and other senior leaders will now be part of Nvidia talent force. This deal, if confirmed, surpasses Nvidia’s $7 billion acquisition of Mellanox in 2019. As of its September 2025 funding round, Groq was valued at $6.9 billion. GroqCloud, the cloud operations arm of the startup, is not part of the deal and will continue independently under new leadership, with CFO Simon Edwards likely to be its CEO, as per reports.

Why Groq matters for Nvidia

AI inference is the process of running trained AI models, such as chatbots and other real-time applications, to generate instant responses. Currently, AI inference is becoming a larger share of total AI computing demand as models move from training to deployment at scale.

This possibly explains Nvidia’s growing emphasis on inference optimisation alongside its dominance in AI training.

Therefore, the Groq deal could bolster AI inference leadership for Nvidia, and eventually help integrate low-latency technology into its AI chips.

Why Groq’s tech is a coveted asset for Nvidia: The case for LPU chips

Groq is led by ex-Googlers like Jonathan Ross, who had helped design Google’s Tensor Processing Unit (TPU). TPUs are specialised AI chips that accelerate complex AI and Machine Learning tasks behind Google's Search, Photos, Translate, and Gemini AI interface. It is also used in Google data centres, Google Cloud, devices and recommendation engines.

Groq builds specialised AI accelerator chips called Language Processing Units (LPUs), designed specifically for fast, low-latency, and energy-efficient AI inference.

While traditional graphics processing units (GPUs) are good at training AI models, Groq’s LPUs focus on deterministic, high-speed inference with lower power consumption.

This makes them well-suited to heavier workloads that need quick token generation.

Catch and kill? What Nvidia is saying about Groq deal

Reports citing Nvidia CEO Jensen Huang said the deal’s goal is to integrate Groq’s low-latency processors into Nvidia’s AI architecture to enhance inference capabilities. He insisted that Nvidia is not acquiring Groq as a company.

So, the deal is perhaps an “acqui-hire” or “reverse acqui-hire”, similar to, but not exactly the same as, Microsoft’s arrangement with Inflection AI or Meta’s deal with Scale AI.

Such deals allow access to talent, IPs, and technology while avoiding the regulatory scrutiny that would have occurred in the case of a full acquisition. Note that regulators are scrutinising partial acquisitions, asset purchases, and talent transfers in AI, as firms use these structures to avoid reviews of full mergers.

At least for now, the Groq deal looks different from a catch-and-kill, or the ‘killer acquisition’ policy in which promising startups are acquired mainly to eliminate competition and shelve their technology rather than developing them further.

Even then, one must note that Groq was indeed a direct competitor to Nvidia in AI inference, where the latter faced more rivals than in AI training.

Nvidia, for now, says it is actively planning to integrate and scale Groq’s LPU technology into its own products, including future AI factories and real-time workloads. Huang’s comments highlight an expansion of capabilities, not suppression of talent or technology, even as the deal would end Groq as an independent rival and consolidate Nvidia’s dominance in inference.