Here's the thing about acqui-hires that most people miss: the target company doesn't get acquired. It gets picked apart. The IP goes one way, the people go another, and what's left behind is a shell with a balance sheet full of cash and a leadership team that just watched its founder walk out the door.
That's exactly what happened to Groq. NVIDIA signed a non-exclusive licensing agreement for the company's LPU (language processing unit) technology — the same silicon architecture that made Groq famous for its lightning-fast inference speeds — and simultaneously hired away CEO Jonathan Ross, president Sunny Madra, and a chunk of the executive team. The whole package was valued at roughly $20 billion. Investors got paid handsomely. NVIDIA got the IP it wanted without the integration headache of a full acquisition.
And Groq? Groq got nothing but a letterhead and a mountain of questions. What do you do when the company that built your moat now owns half the blueprint and is sitting on the other half?
Ross, by the way, wasn't just any CEO. He came from Google, where he helped create the TPU — yes, that TPU — and co-founded Groq roughly a decade ago. His departure wasn't a resignation. It was a transfer of command to the competition.
The deal structure itself is worth staring at for a minute. A non-exclusive license means NVIDIA can use the LPU technology, but so can anyone else who pays. Which sounds generous until you realize NVIDIA has the distribution, the developer ecosystem, and the capital to make that license a death sentence for anyone else trying to sell the same thing. It's not generosity. It's leverage dressed up as openness.
The $650 Million Bet That Said "We're Still Here"
Six months after the NVIDIA deal closed, Groq announced a $650 million funding round. That number alone should tell you something — when investors write that kind of check to a company that just lost its CEO, its president, and its core IP to a rival, they're not investing in what was. They're investing in what's next.
The round was led by Disruptive, a Dallas-based fund founded by Alex Davis — who happens to be Groq's chairman. That's not a coincidence. When your chairman's fund is leading the round, you know two things: first, someone inside the company still believes in the thesis, and second, the people who built this thing are making sure it doesn't die on their watch.
Infinitum, a Fort Lauderdale hedge fund, also came in big. Last valuation was $6.9 billion after a $750 million round in September 2025, so this new money at roughly the same valuation tells you the market hasn't written Groq off — even if NVIDIA thought it had.
But here's where the story gets interesting. Groq didn't just raise cash and hope for the best. They hired a completely new leadership team, none of whom were around when Ross was running things:
- Alan Rice as COO — previously at xAI and Meta, with a U.S. Navy career before that. Someone who's managed large organizations under pressure.
- Sinclair Schuller as CTO — co-founded Apprenda, and his company Nuvalence got acquired by EY in 2024. A builder, not a corporate climber.
- Rakesh Malhotra as CPO — spent roughly a decade on Microsoft cloud products. Someone who understands enterprise distribution at scale.
This isn't a team patching holes. It's a team being assembled for a specific mission. And that mission was something entirely different from what Groq had been doing.
The Pivot: From Silicon to Cloud
The neocloud inference business had been running under Sunny Madra — before he left for NVIDIA, that is. He'd acquired his own company, Definitive Intelligence, back in 2024 and folded it into Groq's infrastructure play. Now the new leadership is doubling down on it as the core strategy.
Let me be clear about what this means. Groq is no longer primarily a chip company. It's an inference cloud provider. They're selling compute, not silicon. And the numbers they're citing are genuinely impressive:
- 13 data centers spread across North America, Europe, the Middle East, and APAC
- 5 million+ developers served
- Thousands of AI companies running on their infrastructure
- Trillions of tokens processed every single week
That's not a startup pivoting on a whim. That's a company that already had significant scale in this business and is now making it the center of gravity. The chip division — what's left of it without Ross and Madra — becomes secondary to the cloud play.
There's a strategic logic to this that most observers are missing. When NVIDIA takes your IP, you can't compete with them on hardware. Not fairly. But if you've already built a cloud platform that runs on that IP, you can compete on service, on reliability, on developer experience — things NVIDIA, for all its might, has historically been terrible at.
NVIDIA sells chips. They sell CUDA. They sell the tools that let you build on their hardware. But they've never been great at being a cloud provider. That's not an insult — it's just how the company is wired. And Groq, by pivoting to neocloud, is playing to a weakness.
The broader neocloud infrastructure wave that Groq is riding has its own supply-chain story: see The Eight-Year Grind That Got Netris a16z Backing for AI Infrastructure, which covers the networking automation companies enabling neocloud operators to deploy GPU clusters faster.
The Irony: NVIDIA Just Became Groq's Competitor
Here's the part that makes this story genuinely delicious. With the LPU IP now licensed to NVIDIA, the company announced its own hardware at GTC in March 2026: the Nvidia Groq 3 LPX inference system. Yes, they literally named it after Groq's technology.
Think about what that means. NVIDIA took Groq's IP, built their own inference chip using it, and is now going to sell that chip — presumably to the same customers Groq was serving. Meanwhile, Groq is running a cloud platform built on that same underlying technology.
This creates a direct competitive tension that most people aren't talking about. Groq's cloud runs on LPU-based hardware. NVIDIA now has its own LPU-based hardware (the Groq 3 LPX) that it can put in its own data centers or sell to other cloud providers. The same IP, two different business models, one of which directly threatens the other.
It's a strange kind of symbiosis. Groq needs NVIDIA to keep licensing the IP — because if NVIDIA stops, they lose the entire technological foundation of their cloud business. But NVIDIA also needs Groq to keep operating, because a dead Groq means nobody's running the LPU architecture in production anymore, which makes the technology look worse than it actually is.
Neither side can kill the other. And that's probably the only reason this whole thing works.
Who's Running the Ship Now
With Ross gone and Madra gone, the company that emerged from this ordeal looks nothing like the one that walked into it. Doug Wightman, Groq's other co-founder (also a Google alumnus), stayed and took over as CEO. He wasn't the face of the company before — Ross was. But Wightman understood the technology, and more importantly, he understood that the company had to evolve or die.
The new C-suite tells you everything about where Groq is going. Alan Rice, the COO, came from xAI — Elon Musk's AI company — and Meta. He's managed massive engineering organizations under extreme pressure. The Navy background suggests he's not afraid of making hard calls.
Sinclair Schuller, the CTO, is a builder's builder. He co-founded Apprenda (a container management platform that got acquired) and his company Nuvalence was bought by EY in 2024. He's not a theorist. He's someone who ships products.
Rakesh Malhotra, the CPO, spent roughly a decade on Microsoft cloud products. That's not incidental — Groq is now positioning itself as an enterprise inference platform, and someone who understands how to sell cloud products to large organizations is exactly the person you want in that role.
This isn't a team trying to recreate what Ross built. It's a team building something entirely new on top of the same technological foundation.
The Bigger Pattern: Acqui-Hire Survivors
Groq isn't the first company to get picked apart by a not-acqui-hire and survive. Scale AI went through something similar about a year earlier, when Meta signed a $14.3 billion deal that extracted talent and IP while leaving the company structurally intact. According to CEO Jason Droege, speaking to Forbes, Scale AI is on track for $1 billion in revenue.
That's the pattern worth watching. These deals — where a dominant player pays a hefty sum for IP licensing and talent extraction without actually acquiring the company — are becoming a feature of the AI landscape. The target company gets cashed out, its best people get absorbed into the acquirer's org chart, and what remains is supposed to wither.
But sometimes it doesn't. Sometimes the remaining company, stripped of its founder and its core technology but still holding onto a functioning business, finds a new direction. Scale AI pivoted toward data infrastructure and labeling at massive scale. Groq is pivoting toward inference-as-a-service.
The investors who profited from the NVIDIA-Groq deal got their return. The employees who left got career moves at the world's most valuable chip company. And Groq, against what should have been overwhelming odds, is still standing — just doing something different than anyone expected.
Whether this pivot works long-term depends on a lot of things: whether NVIDIA actually competes aggressively in inference cloud, whether the new leadership can execute at scale, whether the $650 million buys enough time. But the fact that Groq is still in the game at all, six months after losing its CEO and half its soul to NVIDIA, is itself worth paying attention to.
For context on how other startups are positioning themselves against NVIDIA's dominance in the AI hardware market, see TensorWave to Use $350 Million Funding to Expand Data Centers with AMD Chips.