Nvidia Explores Acquisition or Expanded Stake in Reflection AI
Nvidia is reportedly in early discussions to buy, invest in, or acqui-hire Reflection AI, the maker of the 501-billion-parameter Beam open-weight model valued at $25 billion.
Nvidia Corp. has entered early discussions regarding open-weight artificial intelligence developer Reflection AI Inc., weighing possibilities that include an outright purchase, an expanded equity investment, or an acqui-hire, according to reporting from the Financial Times. Citing individuals familiar with the matter, the report notes that an agreement could materialize within weeks, though talks remain preliminary and could still fall apart.
The potential transaction could take several distinct forms. Nvidia could choose to acquire Reflection outright or increase its existing investment, potentially pairing capital with additional compute resources and processor allocations. A third consideration is an acqui-hire structure, in which Nvidia would bring on Reflection's technical personnel and license its intellectual property without acquiring corporate control—a model that would likely circumvent lengthy regulatory scrutiny. Neither company has publicly commented on the discussions.
Precedents in Nvidia's recent dealmaking strategy
The talks align with Nvidia's active approach to expanding its software and platform capabilities. In December, the chipmaker executed an acqui-hire and licensing agreement with inference chip specialist Groq Inc. in a deal reported at $20 billion. That structure saw founder and Chief Executive Jonathan Ross along with senior leadership transition to Nvidia, while Groq retained its independent cloud operations. Separately, Nvidia opted for a direct acquisition on September 3, agreeing to purchase model hosting platform Hugging Face Inc. for $12.9 billion.
Reflection already maintains financial ties with the semiconductor giant. Nvidia led the startup's $2 billion funding round in October 2025 and has invested $800 million into the company in total. In April, Reflection Chief Executive Misha Laskin stated to CNBC that the company's latest round closed at a $25 billion pre-money valuation. Other venture backers in the startup include Sequoia Capital Operations, Lightspeed Venture Partners LP, and DST Global LP.
Reflection AI's compute commitments and the Beam model
Reflection was founded in 2024 by Laskin and Ioannis Antonoglou, both former researchers at Google DeepMind. The reported discussions arrive just five days after Reflection introduced Beam, its first open-weight architecture. Beam is a mixture-of-experts model containing 501 billion parameters, engineered primarily to handle coding agents and autonomous developer workflows. Reflection has positioned Beam to compete against lower-cost open models such as DeepSeek and Kimi.
Developing Beam has required massive computing commitments. The model underwent training on Nvidia GB300 clusters leased from Space Exploration Technologies Corp. under a June contract valued at up to $6.3 billion. Reflection pays SpaceX $150 million monthly to secure capacity at the Colossus 2 data center in Memphis through 2029. In addition, Reflection entered an agreement with Nebius Group N.V. in July to purchase another $1 billion in AI computing capacity.
What it means for developers
For engineering teams, Nvidia’s interest in Reflection highlights the escalating importance of high-parameter, open-weight models designed specifically for coding tasks. Reflection’s Beam was engineered to challenge models like DeepSeek and Kimi directly on agentic software development tasks, giving teams more open options for code generation and programmatic problem solving.
If Nvidia absorbs Reflection or licenses its technology, developers will need to watch how the startup's open-weight release strategy evolves and whether future models remain accessible for self-hosting. In the meantime, teams building production workflows must constantly evaluate cost, latency, and code accuracy across different systems. Developers can try top AI models cheaply through one API at https://apixoai.online, making it straightforward to compare commercial and open architectures side by side.
Regardless of how the talks resolve, the scale of Reflection's infrastructure commitments—spanning billions of dollars across SpaceX and Nebius—underscores that frontier open-weight models for coding agents now demand computational investments comparable to the largest proprietary AI labs.
Source: Nvidia reportedly in early talks to buy, acqui-hire or invest more in Reflection AI — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
