Groq, the AI inference startup that became known for running large language models faster than anyone else, confirmed a $650 million funding round on June 22 — its first major raise since NVIDIA struck a $20 billion licensing deal that took most of Groq’s founding team with it.

The round was led by Disruptive, a Dallas-based late-stage investment firm whose founder Alex Davis also serves as Groq’s chairman, alongside Infinitum, a Fort Lauderdale hedge fund.

What Happened With NVIDIA

In December 2025, NVIDIA signed a non-exclusive licensing agreement to acquire rights to Groq’s language processing unit (LPU) architecture. On the surface, it was a technology licensing deal. In practice, it was something closer to a talent acquisition.

Founder and CEO Jonathan Ross left to join NVIDIA. President Sunny Madra went with him. So did roughly 90% of the engineering team that built the LPU — the chip design that gave Groq its speed advantage, running inference 10 to 100 times faster than traditional GPU setups for many common model tasks.

Groq retained its intellectual property and its corporate independence. But the people who built the company were gone.

The Rebuild

What’s left is a company at a genuine fork in the road, and the new leadership has made a clear choice: Groq is going all-in on becoming an AI inference cloud, what some are calling a “neocloud.”

Doug Wightman, who stayed through the NVIDIA deal, stepped up as CEO. The company has since added Alan Rice as COO — previously at xAI and Meta, with a background in the U.S. Navy — along with Sinclair Schuller as CTO and Rakesh Malhotra as CPO. The $650M will go toward expanding data center capacity and scaling the cloud service.

The pitch is simple: Groq still holds the LPU technology license. Even though NVIDIA now has the right to use it, so does Groq. The startup can still offer inference speeds that traditional cloud providers can’t easily match — and it plans to compete on exactly that.

What This Means for Business

For businesses running AI workloads, this story matters because it signals where the AI infrastructure market is heading.

More competition in inference. Most of the conversation about AI compute focuses on training — who has the most GPUs, who can build the biggest clusters. But for businesses deploying AI in production, inference is where the cost lives. Every time your AI agent answers a question, summarises a document, or processes a customer request, you’re paying inference costs. More competition means more pricing pressure, which means lower costs over time.

Speed still matters. Groq’s core claim has always been that latency matters for real applications. A voice AI employee that takes four seconds to respond isn’t useful. A customer service agent that hesitates noticeably breaks the interaction. The “neocloud” model Groq is building is a direct bet that enterprise buyers will pay a premium for fast, reliable inference — not just cheap inference.

The talent story is instructive. NVIDIA was willing to spend $20 billion to acquire Groq’s technology and bring its founding team in-house. That’s not a peripheral investment. NVIDIA is betting that inference speed will be a defining constraint in the next wave of enterprise AI deployment, and they wanted Groq’s approach for themselves. Businesses building AI strategies should be watching the inference layer closely.

Groq still has customers. Despite losing most of its engineering team, the company retained its customer base and its cloud service. That continuity matters — it means the $650M raise is a real second act, not a wind-down.

The AI chip race is no longer just about training. Inference infrastructure is becoming the competitive battleground for enterprise AI, and Groq’s survival and re-capitalisation is one more data point confirming it.