How to Buy a Company Without an Acquisition
NVIDIA recently executed a brilliant move to neutralize one of its most formidable competitors. On Christmas Eve, it was announced that NVIDIA is acquiring key patents and the majority of the leadership and engineering teams from Groq. Crucially, however, they are not acquiring the company itself. Here is why.
The Lesson from ARM
In late September 2020, NVIDIA announced its intention to acquire ARM. Within 24 hours, I published a report claiming this deal would never close. Unfortunately for Jensen Huang and Rene Haas, I hit the nail on the head. My experience with major mergers—including work involving Qualcomm/NXP, Broadcom/Qualcomm, and AMD/Xilinx—taught me valuable lessons.
Because ARM is such a dominant player in semiconductor IP, I knew the industry would revolt and regulatory agencies—especially in China—dissent. Ultimately, the deal collapsed, and NVIDIA had to walk away. That expensive lesson clearly informed their strategy with Groq.
The Inference War
I have had conversations with at least three major AI startups targeting the inference market. Their common pitch is that while NVIDIA is the king of training, they are too expensive for inference. Companies like Tensordyne have been vocal about “tokenomics,” even creating a calculator to highlight these costs.
Groq and Cerebras were the primary exceptions, building chips specifically optimized for inference. Jonathan Ross, the inventor of the Google TPU and CEO of Groq, developed the LPU (Language Processing Unit), which offers the best cost-to-performance ratio for inference within the GroqCloud ecosystem.
A Strategic “Acqui-hire”
As the market matures, the focus is shifting. While NVIDIA still leads in training, business sustainability now depends on lowering inference costs. Currently, generating an image or response in ChatGPT or Gemini consumes massive amounts of power—often requiring multiple “re-runs” that waste even more energy.
By securing Groq’s CEO, leadership team, and a non-exclusive patent deal for $20 billion, NVIDIA found a fast-track forward. While the “shell” of Groq will remain to honor previous contracts, the loss of its core talent makes it a much weaker competitor for external players.
Why This Works
This “reverse acqui-hire” is the fastest possible approach. A traditional acquisition would have taken 18 to 24 months and faced a slim chance of regulatory approval given NVIDIA’s 90% market share. By the time regulators could even rule on a merger, the Groq team—now safely inside NVIDIA—will likely have already finished the work they were hired to do.
NVIDIA gets exactly what they need: one less rival to worry about and the elite talent required to build a more cost-effective inference chip.
Jensen Huang, Ian Buck, and the rest of the NVIDIA team: well played.
