Here's what happens when the world's most powerful chip company decides your technology is worth $20 billion — but your company isn't?
You raise $650 million, hire replacements, and try to prove you still matter.
That's exactly what Groq is doing right now. Six months after Nvidia executed what the tech press has been calling a "not-acqui-hire" — signing a licensing deal, poaching founder Jonathan Ross, president Sunny Madra, and a bunch of other engineers, and taking control of the LPU chip IP — Groq is back with fresh money and a new plan.
The question nobody's asking loud enough: does any of this actually make sense?
What Exactly Happened Between Groq and Nvidia
Let me back up. In December 2025, Nvidia did something that looked a lot like an acquisition — except it technically wasn't. The company signed what they called a "non-exclusive licensing agreement" for Groq's LPU technology. That's the chip Groq spent years building: something they marketed as a "language processing unit" designed specifically for AI inference.
Except "non-exclusive" is doing a lot of heavy lifting in that sentence.
Along with the licensing deal, Nvidia hired away:
- Jonathan Ross — Groq's founder and CEO. The guy who literally invented the TPU at Google before starting Groq.
- Sunny Madra — Groq's president, who had been running their neocloud business.
- Multiple other engineers — the people who actually knew how the damn chips worked.
End result? Nvidia now owns the core IP for Groq's LPUs. And at GTC in March, they announced their own inference hardware: the Nvidia Groq 3 LPX. Yes. They put both names on it. Which is either the most honest branding in the history of acquisitions-not-acquisitions, or the most passive-aggressive move I've ever seen from a corporation.
If you've been following the AI infrastructure gold rush — like how SK Hynix just overtook Samsung because AI chip demand is so insane — then this story should feel familiar. The hardware layer is where the real power sits. And Nvidia just proved, again, that they'll take what they want.
The $650 Million Answer
So here's Groq, minus its founder, minus its president, minus the IP for its flagship product. What do you do?
You raise $650 million. Obviously.
The round is led by two firms that, honestly, raise some eyebrows:
| Investor | Type | Notable Detail |
|---|---|---|
| Disruptive | Late-stage VC (Dallas) | Founded by Alex Davis, who also happens to be Groq's chairman. Yes — the chairman invested through his own fund. |
| Infinitum | Hedge fund (Fort Lauderdale) | Less public-facing. More Wall Street, less Sand Hill Road. |
Groq declined to share their new valuation. Which — look, I get it. Their last one was $6.9 billion from a $750 million round in September 2025. Announcing a lower number now would be brutal optics. And announcing the same number when you've lost your CEO? Also not great.
But here's the thing the press release doesn't say: Groq's investors apparently did quite well from the Nvidia deal. The article notes they "profited handsomely." So this new round might be less about survival and more about a second bite at the apple.
The Neocloud Pivot — What Groq Actually Does Now
Without its proprietary LPU chips (well, technically Nvidia owns those now), Groq has pivoted hard to what they call their "neocloud" business. If you haven't heard the term, think of it as cloud inference infrastructure — basically selling access to AI processing at scale.
The numbers Groq is throwing around are legitimately impressive:
- 13 data centers spread across North America, Europe, the Middle East, and APAC
- 5 million+ developers using the platform
- Thousands of AI companies as customers
- Trillions of tokens processed every week
Sunny Madra was running this operation before he left for Nvidia. He'd built it up after Groq acquired his previous company, Definitive Intelligence, back in 2024. Now somebody else is running it — the article doesn't specify who — and the question is whether the neocloud can hold its ground without the chip advantage that made it special in the first place.
Think about it this way: if you could get Groq-speed inference from Nvidia's own "Groq 3 LPX" hardware, why would you stay with the original company? That's the existential problem.
The New Leadership Team (or: Who's Left Standing)
Groq's been hiring fast to fill the gaps. Here's who's running things now:
| Role | Name | Previous Experience |
|---|---|---|
| COO | Alan Rice | xAI, Meta, U.S. Navy |
| CTO | Sinclair Schuller | Apprenda, Nuvalence (acquired by EY in 2024) |
| CPO | Rakesh Malhotra | ~10 years at Microsoft cloud products, co-founder Nuvalence with Schuller |
I'll be honest — this is a solid-looking lineup. Rice coming from xAI and Meta means he knows the AI infrastructure space. Schuller and Malhotra built and sold a company together, which is the kind of track record investors like. The Microsoft cloud background is relevant too — neocloud is, at its core, a cloud infrastructure play.
But here's what nagged me. None of these people are chip designers. They're platform people. Operations people. Product people. And the core competitive advantage of Groq was always the chip — a chip that Nvidia now controls.
It's like losing your quarterback and hiring three really excellent wide receivers. The offense might still look good on paper, but who's throwing the ball?
The Scale AI Comparison
The most optimistic reading of this situation comes from — oddly enough — another company that went through something similar. Scale AI watched Meta execute a $14.3 billion not-acqui-hire about a year ago. Scale's CEO Jason Droege told Forbes that business actually rebounded after the initial shock. Scale is now on track for $1 billion in revenue.
So there IS a playbook. You lose key people, you lose some IP, and you come back leaner and more focused. The market is big enough that you don't need to be the only player — you just need to be good at one specific thing.
For Groq, that one thing might be inference-as-a-service at global scale. For other companies dealing with leadership exodus, the recovery path looks different. But the principle holds: losing people doesn't kill companies. Losing purpose does.
Why This Story Actually Matters
Look, I know this seems like a lot of insider baseball. Billion-dollar chip companies, venture capital rounds, executive reshuffles. But there's a pattern here that goes way beyond Groq.
The AI infrastructure stack is consolidating. Fast. Nvidia is at the center of it. And every time a smaller company builds something that threatens Nvidia's dominance in inference, one of two things happens:
- Nvidia builds its own version and outcompetes them on scale, or
- Nvidia "partners" with them in a way that absorbs their IP and talent
Groq got hit with option two. The fact that they're still standing — with $650 million in the bank and a global neocloud network — is actually kind of remarkable. Most companies don't survive losing their founder, their president, and their core intellectual property in a single deal.
But "surviving" and "thriving" are very different things. And I'm gonna wait to see which one this turns out to be before I start writing the victory lap.
What to Watch Next
A few things that'll tell us whether Groq's bet pays off:
- Customer retention numbers. Are those 5 million developers actually sticking around, or slowly migrating to Nvidia's own inference platform?
- The next earnings or revenue disclosure. $1 billion in revenue is the benchmark that Scale AI hit. Can Groq get there?
- New chip development. Is Groq working on something new that doesn't rely on LPU IP? Because if they're just running other people's silicon, the moat is gone.
- Nvidia's pricing. If Nvidia prices the Groq 3 LPX aggressively, it becomes very hard for the original Groq to compete on cost.
Comparison: AI Chip Companies vs. The Nvidia Problem
| Company | Strategy | Nvidia Threat Level | Current Status |
|---|---|---|---|
| Groq | Neocloud inference service | Extreme — Nvidia owns their LPU IP | Rebuilding, $650M fresh |
| Cerebras | Wafer-scale chips for training + inference | High — different architecture but competing for same customers | IPO attempts, legal battles |
| SambaNova | Reconfigurable dataflow units | Medium — niche approach | Pivoting to enterprise |
| AMD (MI300X) | Direct GPU competitor | Direct competition, massive scale | Gaining cloud data center share |
| Custom ASICs (Google TPU, Amazon Trainium) | In-house silicon for own clouds | Low — they ARE the competition | Scaling aggressively |
Bottom Line
Groq raising $650 million after effectively selling its soul to Nvidia is either the most audacious comeback story in AI, or it's the most expensive way to watch a company slowly become irrelevant. Maybe both simultaneously.
The money is real. The infrastructure is real. The customers are real. What's not clear — what might never be clear for another year or two — is whether any of that adds up to a company that can compete in a world where the biggest player in GPU computing now owns your best technology.
I'll believe it when I see the revenue numbers. Until then, I'm filing this under "interesting but unproven." And keeping my eyes on whether those 5 million developers start drifting toward Nvidia's own platform.
Because that's what actually decides this story. Not the funding round. Not the executive hires. Just: do developers choose Groq's cloud when they could use Nvidia's cloud running the same hardware?
Sources
- TechCrunch: "AI chipmaker Groq confirms $650M raise, re-staffs after Nvidia's $20B not-acqui-hire deal" — Julie Bort, June 22, 2026
- Groq Official Website
- Nvidia Data Center Solutions
- Scale AI — referenced for not-acqui-hire recovery comparison
Related Reading
Frequently Asked Questions
