Six months ago, Groq was gutted. Nvidia paid off its investors, hired away its founder and chief executive Jonathan Ross, president Sunny Madra, and a chunk of its engineers. The company looked like a corpse being picked clean.
Now it has $650 million. The money is a bet that Groq can be rebuilt from scratch as an AI inference cloud — and that purpose-built chips can still beat the general-purpose GPUs that dominate the market today.
The raise was led by Disruptive and Infinitum. Existing backers reinvested alongside them. The round was first reported in May.
Now it is closed, and the company has named the team that will spend the money. That team faces a brutal reality.
Nvidia did not just take people. It struck a non-exclusive licensing deal for Groq’s chip technology — an arrangement reported to be worth a considerable amount. So the new Groq does not even own its own hardware exclusively.
It is rebuilding with one hand tied behind its back. At its GTC event in March, Nvidia launched the Groq 3 LP. That is the chip Groq designed, now being sold by the company that dismantled them.
The $650 million bet is that none of that matters. That the AI inference market — the business of running already-trained models, not training them — is big enough and growing fast enough that a scrappy cloud service using custom silicon can carve out a real business.
That GPUs, for all their muscle, are not the best tool for every job. It is a contrarian wager. Nvidia’s GPUs are everywhere.
They train the models. They run the models.
The software stack is mature. The ecosystem is locked in. But Groq’s argument is simple: purpose-built chips are faster and more power-efficient for inference.
They cost less to operate. For companies running AI at scale, that math adds up. The raise lands in Manila, but the implications stretch across Asia.
Groq’s rebuilt cloud will compete for customers in a region where AI adoption is accelerating. Healthcare, education, finance, transportation, logistics — every sector is looking at inference workloads.
Every sector wants lower costs and lower latency. If Groq pulls this off, the impact on families and businesses in the region could be real. Cheaper inference means cheaper AI services.
Cheaper AI services mean broader access. Broad access means more applications that actually touch people’s lives — diagnostic tools in rural clinics, automated translation for cross-border trade, logistics optimization for small suppliers.
If Groq fails, it will be a case study in how hard it is to compete when the largest chip company in the world buys your talent and licenses your technology. Nvidia did not kill Groq. It took everything that made Groq valuable and absorbed it.
What remains is a shell with a thesis and a pile of cash. The company is now a pure bet on execution. The technology is partly licensed away.
The founding team is gone. The market is dominated by a competitor with vastly more resources.
All Groq has is $650 million and the conviction that custom silicon still wins in inference. That is a thin edge. But in technology, thin edges have won before.
Sources
- Groq Raises $650M After Nvidia Deal, Pivots to AI Inference
- AI chipmaker Groq confirms $650M raise, re-staffs after Nvidia's $20 …
