Let's get the number on the table first: $350 million. That's what the AI-chip startup Groq just raised, at a $3.5 billion valuation, in a round led by the Dallas firm Disruptive with Nvidia expected to join in (TechFundingNews).

A $350 million Series A is a big check by any normal standard. But here's the part that makes you look twice: this same company was worth $6.9 billion less than a year ago (Value Add VC). So Groq just raised a fortune at roughly half the price it commanded in September 2025.

How a company loses half its value while investors keep writing nine-figure checks is the whole story. The short version: in December, Nvidia licensed Groq's core technology and hired away the people who built it — including the founder.

The thesis: Groq is being rebuilt from the outside in, and $3.5 billion is the price of betting on the shell that's left.

🧠 Why This Matters

Groq makes an inference chip — the "LPU," a processor tuned to run AI models fast rather than train them. For a while that made it one of the hottest names in the Nvidia-alternative crowd, riding a $750 million round to that $6.9 billion peak (Briefs).

Then Nvidia did something unusual. Rather than buy Groq outright, it signed a non-exclusive license — reportedly worth up to $20 billion — for the rights to Groq's inference architecture, and brought founder and CEO Jonathan Ross, president Sunny Madra, and roughly 90% of the engineering team over to Nvidia (TechFundingNews).

Read that again. The market leader paid for the technology and the talent, but left the company standing. Groq is still, on paper, an independent business. It just no longer employs most of the people who made it interesting.

📊 Deep Dive

The money has been moving fast. In June, Groq pulled in $650 million to fund a data-center buildout. Add this week's $350 million and that's about $1 billion raised in roughly two months (TechFundingNews). The pitch to investors is a pivot: from selling chips to renting them, as a cloud.

Here's how the two chapters of Groq stack up:

  • Peak (Sept 2025): $6.9B valuation, $750M round, founder Jonathan Ross at the helm, chip-maker story.
  • Now (Aug 2026): $3.5B valuation, $350M round, Executive Chairman Alex Davis running the board, cloud-provider story.
  • The build: 13 data centers across North America, Europe, the Middle East and Asia-Pacific, 54 megawatts of capacity today, a target of over 200 megawatts by 2027 (The AI Insider).
  • The reach: more than 6 million developers and enterprise customers already using its inference service.

The person now setting direction is Alex Davis, who also runs Disruptive, the firm leading these rounds. His framing is that inference — running models in production — is where the real volume of AI computing will land.

"Inference will without a doubt become the largest and most critical layer of AI infrastructure. We will be focused on supporting the most important model makers." — Alex Davis, Executive Chairman (Briefs)

⚠️ The Catch

You can't talk about a comeback plan without naming what left. Nvidia didn't just license a patent portfolio; it took the founder, the president, and the bulk of the engineers who designed the LPU (Value Add VC). Rebuilding a hardware company after that is not a hiring problem you solve in a quarter.

There's also the small matter of competing with your own biggest investor. Nvidia is expected to join this round and now holds a license to the very architecture Groq's cloud runs on. When your supplier, your rival, and your cap-table are the same company, "independent" starts to carry an asterisk.

And the cloud pivot puts Groq head-to-head with Amazon, Microsoft, Google and a pack of well-funded inference startups — a market where 54 megawatts is a rounding error next to the hyperscalers' gigawatts.

🎯 What Happens Next

Watch the megawatts. Groq's entire new story rests on getting from 54 MW to 200-plus MW by 2027 and filling that capacity with paying inference workloads. If those data centers light up and stay busy, the $3.5 billion price looks cheap. If utilization lags, it looks generous.

Watch the engineering bench, too. A company that lost ~90% of its chip team has to prove it can still ship silicon — or quietly become a reseller of compute it no longer designs. The next chip roadmap will tell you which.

🧩 Bigger Picture

Groq is a clean example of a new pattern in AI: the license-and-lift. Instead of an acquisition that triggers antitrust scrutiny, the giant pays for the technology and the team, leaves a legal entity behind, and lets outside investors re-fund the husk. Everyone gets a headline. The founder gets liquidity and a big-company job. The investors get to keep believing.

Whether that husk becomes a real cloud business or a cautionary tale depends entirely on execution from here — with most of the original brains now sitting in Santa Clara.

Groq spent a year going from $6.9 billion to $3.5 billion and just raised a billion dollars doing it. In AI right now, losing half your value is apparently still a fundable event.

Sources