Groq is reportedly seeking up to $650 million from existing investors as the Mountain View AI chip startup tries to build a second act around inference cloud services after a major technology licensing agreement with Nvidia.

Axios first reported Thursday that Groq was raising the money from existing backers, after what it described as a $20 billion licensing deal with Nvidia that saw much of Groq’s senior team depart. Reuters, citing a source familiar with the matter, also reported Thursday that Groq was raising up to $650 million, but described the earlier Nvidia licensing deal as worth $17 billion. TechCrunch followed Friday with a report that Groq is leaning into an inference “neocloud” business using its homegrown AI chip and systems.

The new financing has not been announced by Groq in a formal company release, and the reports rely on unnamed sources. Groq’s public newsroom, as of the latest available company release cited in this story, had not posted a new announcement for the reported $650 million raise. The company’s most recent major financing announcement on its website was its September 2025 statement that it had raised $750 million at a $6.9 billion post-money valuation.

The reported raise is notable because it would come after Groq investors already received payouts tied to the Nvidia transaction, according to Axios and Reuters. Reuters reported that a final cash distribution was expected soon through the Nvidia deal, and that existing shareholders would then have the opportunity to invest in a new company structure. Axios described the process as investors being asked to participate in “Groq 2.0,” with existing backers Disruptive and Infinitum backstopping the $650 million raise if it is not fully subscribed.

That structure is what makes the case unusual for Silicon Valley’s AI market. Instead of a straightforward acquisition in which Nvidia buys Groq outright and absorbs the company, the reported arrangement appears to have separated licensing, talent movement, investor liquidity and future operating plans. TechCrunch described it as a “not-an-acquisition” agreement involving senior Groq employees moving to Nvidia and Groq’s hardware technology being licensed to the chip giant.

For Nvidia, the reported licensing agreement strengthens a broader push to control more of the AI compute stack. The company is already dominant in chips used to train large AI models, but inference — the process of running trained models to answer user prompts, execute agentic workflows or power enterprise applications — is becoming a larger infrastructure battleground. Groq has positioned itself around inference performance and cost, saying on its website that it delivers fast, low-cost inference and that its LPU-based stack runs in data centers across the world to support low-latency AI responses.

For Groq, the reported raise suggests a strategic pivot away from selling hardware as the primary business and toward offering cloud access to inference capacity. Reuters reported that the startup has been shifting focus away from hardware toward AI inferencing. TechCrunch similarly reported that Groq’s business is now centered on helping developers and enterprises host inference-heavy applications. That is a narrower market position than competing across the full AI chip landscape, but it also maps to a growing need among companies deploying AI systems at scale.

Inference has become a critical cost and performance issue as AI moves from experimental chatbots into production software. Training a frontier model can require massive clusters and capital expenditure, but inference happens repeatedly every time a user or application calls the model. For companies running customer support agents, search assistants, coding tools, analytics copilots or automated workflow systems, inference speed, reliability and unit economics can determine whether AI products are commercially viable.

Groq’s September 2025 funding announcement framed the company as part of what it called the “American AI Stack,” saying its infrastructure was used by more than two million developers and Fortune 500 companies. The company said that earlier $750 million round was led by Disruptive, with participation from BlackRock, Neuberger Berman, DTCP, a large U.S.-based West Coast mutual fund manager, Samsung, Cisco, D1, Altimeter, 1789 Capital and Infinitum. That investor list is relevant because the latest reported financing is described as coming from existing investors rather than a broad new outside syndicate.

The leadership picture is also part of the story. Axios reported that the new Groq phase is being led by company veterans Adam Winter as CEO and Matt Eng as CFO. TechCrunch described Winter and Eng as interim CEO and CFO. Groq has not issued a public release in the cited materials confirming the final leadership structure for the reported new raise, so the company’s internal organization should be treated cautiously until it is formally announced.

The deal also lands in a sensitive regulatory environment. Large technology companies have increasingly used licensing deals, talent transfers and strategic investments as alternatives to full acquisitions, especially in AI, where antitrust scrutiny has intensified. A traditional acquisition of a prominent AI chip startup by Nvidia could have drawn close review because Nvidia is already the central supplier of high-end AI accelerators. A licensing-and-talent arrangement may achieve some strategic goals while avoiding the optics and legal process of a conventional merger, though regulators can still examine whether such arrangements reduce competition.

The Groq case could therefore become a reference point for other AI startups and investors. Venture-backed companies in chips, AI infrastructure and model development face enormous capital requirements, while large platform companies want technology and talent without always pursuing full takeovers. If investors can receive a large cash distribution from a licensing transaction and still finance a reorganized company, the model may appeal to backers looking for liquidity in a crowded but capital-intensive AI market.

At the same time, Groq’s reported plan carries execution risk. The startup must prove that its inference cloud can win customers in a market that includes hyperscale cloud providers, Nvidia-powered infrastructure vendors, specialized AI infrastructure startups and internal enterprise deployments. It also must do so after reported senior departures and after licensing key hardware technology to Nvidia, which could become both a strategic partner and a competitive force depending on how the licensed technology is commercialized.

The reports leave several important details unresolved, including the new company’s valuation, the exact terms of the financing, how much of the round is formally committed, whether all existing investors will participate, and how Groq will separate its remaining operating assets from the Nvidia licensing arrangement. Reuters reported that existing backers Disruptive and Infinitum would backstop the raise if it is not fully subscribed, while Axios reported that investors were being asked to participate pro rata.

For now, the verified development is that multiple outlets report Groq is seeking up to $650 million after a major Nvidia licensing deal, with the company’s next chapter focused on inference infrastructure rather than a traditional chip-startup path. In Silicon Valley’s AI economy, that makes Groq less a simple funding story than a test of how startup liquidity, strategic technology access and AI compute demand are being recombined in real time.