What happened
Nvidia is in early talks to acquire open-weight model developer Reflection AI outright, take a bigger stake, or do an acquihire that hires its staff and licenses its technology, per the Financial Times. An agreement could come within weeks, though the talks could collapse.
Why it matters
If a deal happens, Nvidia would deepen its control over a startup it already backs as a major investor, and an acquihire structure would likely avoid a lengthy regulatory review. Nvidia has done both an acquihire and a straight purchase in the past year.
WHO IT HITSThe talks land on Nvidia's deal-making team and on Reflection AI's investors and staff, who could see the startup absorbed, expanded with more Nvidia chips and computing power, or restructured as a talent-and-license deal. Rivals in open-weight models such as DeepSeek and Kimi may face a better-funded competitor if Nvidia takes a bigger stake.
Summaries like this, in your inbox every morning.
The talks follow a pattern Nvidia has already set in the past year. In December, inference chip startup Groq licensed its technology to Nvidia in a deal reported at $20 billion, with founder and CEO Jonathan Ross moving to the chipmaker while Groq kept running its cloud business on its own. Nvidia also agreed Sept. 3 to pay $12.9 billion for Hugging Face, a straight purchase. That history gives the Reflection talks three ready-made shapes: a full buyout, a bigger stake possibly paired with more chips and computing power, or an acquihire.
Reflection is not a stranger to Nvidia. The chipmaker led the startup's $2 billion round in October 2025 and has put $800 million into the company in total, according to the Financial Times. CEO Misha Laskin told CNBC in April that Reflection's latest round had closed at a $25 billion pre-money valuation. Laskin and Ioannis Antonoglou were both researchers at Google DeepMind before founding the company in 2024, and its backers include Sequoia Capital, Lightspeed Venture Partners and DST Global.
The timing sits close to Reflection's first open-weight model. Beam, a mixture-of-experts model with 501 billion parameters, was unveiled five days before the report and is pitched against lower-cost Chinese open models such as DeepSeek and Kimi on coding and agentic work. Beam's training ran on Nvidia GB300 systems that Reflection rents from SpaceX under a June agreement worth up to $6.3 billion, with $150 million monthly payments covering capacity at the Colossus 2 data center in Memphis through 2029. Nebius Group agreed in July to sell Reflection another $1 billion in AI capacity. Neither Nvidia nor Reflection has commented publicly on the talks.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
India's technology buildout is deepening across semiconductors and AI infrastructure, with chip IP, packaging…
Cerebras Systems announced it is powering the Ultrafast mode for OpenAI's GPT-5.6 Sol, delivering up to 750 ou…
Amazon is in talks to acquire AI startup Decart in a deal that could be valued around $7 billion, the Wall Str…
A walkthrough uses a custom four-signal reward (correctness 0.6, parseability 0.2, format 0.1, reasoning 0.1)…
A Fortune commentary argues AI is repeating a 50-year pattern called the Innovation Paradox
A self-described data analyst argues on Zenn that feeding large language models past analysis reports and simi…
