Groq Acquisition Clears Path For Cerebras
OpenAI's side chip
Nvidia’s Groq deal removed the strongest independent rival in low latency inference and left Cerebras as the clearest remaining pure play alternative. The reason the comparison matters is that Groq was not just selling chips, it had already turned its hardware into a cloud product with pay per token pricing, projected $500M of 2025 revenue, and a licensing agreement that shifted key Groq talent and technology into Nvidia while Groq kept operating.
-
Groq was effectively folded into Nvidia at the technology layer, not through a standard merger. The core pieces were a non exclusive inference technology license signed on December 24, 2025, $17B of cash payments tied to that deal, and the move of founder Jonathan Ross and other key operators to Nvidia.
-
That matters because Groq had become a real commercial benchmark for inference. It had grown from about $90M of 2024 revenue to a projected $500M in 2025, selling OpenAI compatible API access, enterprise contracts, and hardware racks for customers that wanted very fast token generation with tight latency.
-
Cerebras is now differentiated less by being the only startup with custom silicon, and more by its go to market shape. Where Groq centered on inference cloud and GroqRack, and SambaNova sells a fuller enterprise stack with services, Cerebras is turning its wafer scale hardware into an API business aimed at coding agents and other speed sensitive workloads.
From here, the independent AI chip field should narrow around companies that can convert hardware advantage into recurring software like revenue. With Groq’s best assets now feeding Nvidia, Cerebras has a clearer opening to own the premium speed segment, especially if inference and agent loops keep growing faster than training.