Nvidia bought Groq's chip IP for $20B. Then poached the founder. Groq just raised $650M to rebuild — using Nvidia's own hardware.

Nvidia paid $20B for Groq's LPU technology in December.

Then hired away founder Jonathan Ross and president Sunny Madra.

Six months later, Groq just closed a $650M raise to rebuild as a neocloud.

The cruel irony: Groq will now deploy Nvidia's LPX system — which contains Groq's own licensed IP — to scale its inference cloud.

This is what vendor dependency looks like when the vendor decides to become your competitor.

Groq currently operates 13 data centers across four continents.

Serving 5 million developers. Processing trillions of tokens per week.

But Nvidia now owns the hardware playbook. And Groq's new leadership team — ex-xAI, ex-Meta, ex-Microsoft — has to build a cloud business on top of chips their rival controls.

The inference market is projected to demand 15-20x more compute than training.

Groq is betting that operational expertise matters more than chip ownership.

Scale AI survived the same play when Meta poached its CEO for $14.3B.

But Scale AI doesn't depend on Meta for its core infrastructure.

Groq depends on Nvidia for theirs.

Audit your inference vendor contracts now.

If your AI infrastructure runs on a single chip provider, you're one licensing deal away from becoming a customer of your own competitor's platform.

SOURCE: https://groq.com/newsroom/groq-raises-usd650m-to-scale-its-ai-inference-cloud-business
VERIFIED: TechCrunch, SiliconANGLE, Groq official announcement
SIGNAL: The $20B not-acqui-hire deal is the template for how hyperscalers absorb startup innovation. Every enterprise running inference on specialized hardware needs to watch this.