According to the Financial Times, citing informed sources, NVIDIA is in advanced negotiations to acquire U.S. startup Reflection AI or to make an additional investment in the company.
Reflection AI focuses on developing open-weight models and launched its first open-weight AI model, Beam, on October 5 local time.
A person familiar with the matter added that negotiations are still in early stages, and the potential transaction could take several forms, including what is known as a "talent acquisition." Under this structure, NVIDIA would hire its core team and obtain technology licensing, thereby avoiding the lengthy antitrust reviews typically associated with a full acquisition.
The report, citing informed sources, states that NVIDIA is currently one of Reflection’s major shareholders and may further deepen its ties with the company through a new equity investment or by offering additional chips and computing resources. NVIDIA has previously invested $800 million in Reflection.
According to several informed sources, the two parties are expected to finalize a deal within the coming weeks, though all sides emphasized that negotiations could still break down. Specific terms of the transaction have not yet been disclosed. However, during its last funding round in March, Reflection was valued at $25 billion. Both NVIDIA and Reflection declined to comment.
Reports indicate that "talent acquisitions" are increasingly becoming the preferred strategy for large tech companies in the face of stringent antitrust scrutiny by regulators on AI-related mergers and acquisitions. In December last year, NVIDIA employed a similar structure in its $20 billion transaction involving the chip design company Groq; this deal has recently prompted a lawsuit from former Groq employees whose offers were not accepted.
NVIDIA has embarked on aggressive investment and expansion over the past two years to boost the overall AI industry ecosystem and solidify its position at the core of the computing economy. Although NVIDIA has already launched the Nemotron family of open-weight AI models, it still lacks its own independently developed cutting-edge large model.
In July this year, NVIDIA CEO Jensen Huang led calls for the establishment of an open-weight ecosystem within the United States, emphasizing its critical importance for “sparking innovation and fostering prosperity across the nation.” The report also stated that in September, the $5.5 trillion tech giant acquired Hugging Face, an open-source model and dataset hosting platform, for nearly $13 billion.
The report states that NVIDIA has invested tens of billions of dollars in leading large model labs such as OpenAI, emerging GPU cloud providers like Nebius, and numerous smaller and mid-sized teams. These investments have further increased market demand for its AI chips—essential core components that determine computing power for any company building or running large models.
Source attribution: IT Home, citing the Financial Times.