Nvidia is in early-stage talks that could result in a deeper investment in Reflection AI or an outright acquisition of the startup, according to reports from the Financial Times and Reuters.
The discussions could produce several outcomes, including another equity investment, additional chips or computing capacity, or an “acqui-hire” in which Nvidia would hire Reflection employees and license its technology instead of buying the whole company. The reports said an agreement could come within weeks, but also cautioned that the talks may still fall apart. Reflection declined to comment, while Nvidia did not immediately respond to Reuters.
What Nvidia already has at stake
Nvidia has invested $800 million in Reflection and is one of the startup’s largest shareholders, the FT reported. Reflection was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou. It develops AI systems for software development and more recently has positioned itself as a provider of open models that customers can adapt and run in their own environments.
That strategy became more concrete with Beam, which Reflection describes as a 501-billion-parameter open-weight model. “Open-weight” means customers can inspect, fine-tune and deploy the model rather than relying only on a closed provider’s hosted service. Reflection says Beam can run in private clouds, on premises, in air-gapped systems and at the edge. Axios reported that the company’s broader goal is an “AI factory” product that combines its models with customers’ private data and rented computing power.
A fast-rising valuation
Reflection’s valuation has climbed quickly. The company raised $2 billion at an $8 billion valuation in October 2025, up from a previous $545 million valuation. In April 2026, Laskin said the company was seeking fresh capital at a $25 billion pre-money valuation.
For Nvidia, a larger stake or acquisition would deepen a relationship that already links its chips to a US open-model developer. For Reflection, the talks arrive only days after Beam’s launch. The company is positioning the model against lower-cost Chinese systems in coding and agentic tasks, but there is no announced transaction, and the range of structures under discussion shows that the outcome remains unsettled.