Nvidia just priced out competition
The market flipped from training to inference in 2025. Nvidia paid to own the new bottleneck before it scaled.
Merry Christmas and Happy Holidays!
For the first time, Nvidia paid to remove a competitor rather than bury it with better products.
On December 24, 2025, Nvidia announced its largest deal on record: a $20B non-exclusive licensing agreement with AI chip startup Groq, which includes hiring Groq’s founding CEO, Jonathan Ross, and key executives. [RELEASE]
Forget the “licensing deal” framing. This is a stealth talent and IP acquisition that strengthens Nvidia’s grip on AI, locking down model training (already at a ~90% share) and positioning it to dominate the $255B AI inference1 market by 2030.
Let’s unpack.
Zooming in
Groq continues operating as an independent company under new CEO Simon Edwards (promoted from CFO). GroqCloud remains live. This distinction matters legally. It allows Nvidia to avoid the Hart-Scott-Rodino antitrust review.
Groq’s LPU technology uses deterministic execution and large on-chip SRAM, while Blackwell handles batch inference.
The IP. 74 patents on low-latency2 inference architecture. On-chip SRAM memory running at 80 TB/s versus Nvidia's GPU memory at 8 TB/s. In plain terms: Groq could run real-time AI 10x faster and 10x cheaper than Nvidia's hardware for single-user workloads.
The valuation gap. Groq had just missed 2025 revenue projections by 75%, down from $2B guidance to $500M. In February 2025, it secured a $1.5B commitment from Saudi Arabia to expand its AI chip delivery and infrastructure in the country. In December, Nvidia paid $20B instead of waiting for the price to collapse.
The talent exodus. Jonathan Ross (founder, former Google TPU architect), Sunny Madra (president, three successful exits), and Groq's core compiler team moved to Nvidia.
In July, we mentioned Groq as one of the emerging competitors of Nvidia. And, five months later, it is a part of Nvidia.
Stepping back: Nvidia is following a pattern:
September 2025: Intel partnership ($5B). Nvidia gets foundry access. Ensures TSMC scarcity doesn’t kill supply.
September 2025: OpenAI investment ($100B, structured as a partnership). Nvidia secures the model layer.
Why this matters
Nvidia is buying permanence.
