NVIDIA in early talks to deepen its Reflection AI investment or supply it compute

Bloomberg reports that NVIDIA is weighing a deeper relationship with Reflection AI, either by boosting its equity stake or by striking a compute supply deal. The URL slug references an FT report on acquisition talks, but Bloomberg's framing is more cautious: discussions are early and no terms have been agreed. NVIDIA has already put $800M into the company.

Reflection AI, founded by former DeepMind researchers, is building open-weight frontier models — the same category where Mistral just previewed its 1T-parameter Large 4 and where NVIDIA itself released Nemotron 3 Nano under an open license. Reflection was also a sponsor of this week's open-source AI gathering promoted by Hugging Face.

The pattern is familiar: NVIDIA investing in model labs that in turn buy enormous amounts of NVIDIA hardware. Compute-for-equity arrangements effectively pre-sell GPU capacity and help NVIDIA cultivate an ecosystem of open-model developers who aren't hyperscalers designing their own chips. A U.S.-based open-weight lab is also strategically attractive amid concern that Chinese open models are closing the capability gap.

Skeptics note the circularity of such deals, which can flatter revenue while concentrating risk in a handful of capital-hungry startups. Watch whether talks escalate toward a full acquisition, which would bring antitrust attention, or settle into a structured compute deal.