Groq Raises $350M at Half Its Peak Valuation, Completing Its Reinvention as an Inference Neocloud
Bloomberg reports Groq raised $350M at a $3.5B valuation — roughly half its September 2025 peak — with Nvidia joining a round led by Disruptive, capping the LPU pioneer's pivot from chip rival to AI inference data-center operator.
One of the AI boom’s most instructive corporate sagas reached a new milestone on August 17, 2026, when Bloomberg’s Natasha Mascarenhas reported that Groq has raised $350 million in fresh funding at a $3.5 billion valuation — roughly half of what the company was worth at its September 2025 peak, before Nvidia struck a sweeping licensing deal with the startup and hired away much of its talent, including founder-CEO Jonathan Ross.
The round, set to be announced Monday, was led by Dallas-based investment firm Disruptive, Groq’s most persistent backer through every phase of its transformation. In a twist that says a great deal about how the AI hardware industry actually works, Nvidia — the company that absorbed Groq’s chip team — is also investing in the round, according to a Groq representative. Nvidia did not respond to a request for comment.
The numbers behind the down round
To understand why a $350 million raise at $3.5 billion counts as a landmark event, it helps to trace Groq’s valuation curve. In August 2024, the company was valued at $2.8 billion. In September 2025, Reuters reported that a $750 million round more than doubled the company’s worth to $6.9 billion in just over a year, with BlackRock and Neuberger Berman among the investors piling in. At that moment, Groq was arguably the most credible independent challenger to Nvidia’s inference dominance, with its Language Processing Units famous for serving open models at hundreds of tokens per second.
Then came December 2025, when CNBC reported that Nvidia was acquiring Groq in a deal worth about $20 billion — its biggest ever — structured not as a conventional acquisition but as an extensive licensing agreement, with Nvidia licensing Groq’s technology and hiring away much of its engineering leadership. The deal left the Groq corporate entity intact, but transformed: no longer a would-be Nvidia rival, it became something closer to an inference-focused data-center operator, or what the industry now calls a neocloud.
Monday’s $3.5 billion valuation is the market’s honest re-pricing of that remnant. It is roughly half the September 2025 figure — but it is also a quarter more than the $2.8 billion at which Groq traded before its late-2024 hype cycle, and it comes after the company pocketed billions in licensing proceeds from the deal that hollowed it out.
What the money is for
According to Bloomberg, Groq plans to use some of its recent financing to expand its total data-center capacity to more than 200 megawatts by next year. That tracks with the company’s June 2026 announcement, when it raised $650 million led by Disruptive and Infinitum to scale what it called its global AI inference cloud. At that point Groq was already operating 13 data centers, with a stated goal of reaching 200 MW of capacity by the end of 2027 — a timeline the new capital appears to have accelerated by a year.
The pivot makes commercial sense. Groq’s brand recognition with developers — built on the eye-watering speed of its LPU-based inference endpoints — is an asset that survives the loss of a chip roadmap. Serving models fast, at scale, from owned infrastructure is now a genuine business, as CoreWeave’s rise and the flood of capital into neoclouds like Nebius and Crusoe have demonstrated. Groq’s June disclosures even noted that part of its capacity expansion involves deploying Nvidia’s new LPX platform — meaning the former Nvidia rival is now, in part, an Nvidia customer.
Why Nvidia’s participation matters
The optics of Nvidia investing in Groq less than a year after licensing its technology and recruiting its team are striking, and they cut in several directions.
The cynical reading is that this is consolidation by another name: Nvidia neutralized its most visible inference challenger, took the engineers, and now takes a financial stake in the operator that emerged — aligning Groq’s remaining business with Nvidia’s hardware roadmap rather than against it. Antitrust watchers have already noted that chip giants have participated in AI and robotics funding rounds collectively worth more than $250 billion in 2026 to date, with Nvidia alone involved in 59 known deals. Groq is now one more line in that ledger.
The more charitable reading is that the arrangement is simply rational for both sides. Groq’s licensed technology continues to generate value inside Nvidia’s stack, so Nvidia has a direct interest in the surviving entity’s health. And for Groq’s remaining shareholders — including employees who stayed through the pivot — a down round that locks in $350 million of growth capital and Nvidia’s continued engagement is far better than the alternatives.
What this says about the AI hardware landscape
Groq’s arc is the clearest case study yet of a structural truth in AI infrastructure: competing with Nvidia on chips has proven nearly impossible, but complementary business models around inference — speed, serving, capacity — remain fundable.
The down round also lands in a week when the industry’s finances are under unusual scrutiny. A Wall Street Journal analysis found nine top tech firms carrying roughly $3 trillion in AI-related off-balance-sheet commitments, and investor skepticism about the durability of AI valuations is growing louder. Against that backdrop, Disruptive’s willingness to keep writing checks into a company whose headline valuation just halved is notable — it suggests underwriters are differentiating between AI bets on business fundamentals rather than marking everything down together.
For Groq’s original thesis — that deterministic, software-first silicon could beat GPUs at inference — the verdict is mixed. The LPU design was good enough that Nvidia paid $20 billion for it. But the company built on it no longer exists in the form its founders imagined. What exists now is a well-capitalized inference operator with a famous name, a speed-obsessed developer following, Nvidia’s money in its cap table, and a mandate to hit 200 megawatts by next year.
In an industry that spent 2025 and 2026 rewarding ever-larger rounds at ever-higher valuations, Groq’s $3.5 billion re-pricing is a small dose of honesty: talent walks, narratives expire, and the surviving businesses are the ones that find a way to sell compute. Groq just raised money to do exactly that — at half the price, but on its own terms.
Sources
- [1] https://www.bloomberg.com/news/articles/2026-08-17/groq-valued-at-3-5-billion-in-funding-round-after-nvidia-deal
- [2] https://news.bloomberglaw.com/private-equity/groq-valued-at-3-5-billion-in-funding-round-after-nvidia-deal
- [3] https://seekingalpha.com/news/4633841-groq-valued-at-3_5b-in-new-funding-round-report
- [4] https://groq.com/newsroom/groq-raises-usd650m-to-scale-its-ai-inference-cloud-business
- [5] https://www.cnbc.com/2025/12/24/nvidia-buying-ai-chip-startup-groq-for-about-20-billion-biggest-deal.html
- [6] https://www.reuters.com/business/groq-more-than-doubles-valuation-69-billion-investors-bet-ai-chips-2025-09-17/
- [7] https://siliconangle.com/2026/06/22/inference-chip-startup-groq-raises-650m-grow-cloud-platform/