Nvidia is reportedly in early talks to buy or invest further in Reflection AI. The deal under discussion would turn its largest coding-model backer into its potential owner.

Nvidia is in early-stage talks to deepen its investment in Reflection AI or acquire the startup outright, the Financial Times reported on 10 October 2026, citing people with direct knowledge of the discussions. A deal could close within weeks, according to the FT’s sourcing, though the talks could still collapse before any terms are reached.

Reflection AI, founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, builds open-weight models aimed at software development and agentic coding tasks. The company raised a round earlier this year that put its pre-money valuation at 25 billion dollars; reporting on the exact month of that round is inconsistent, with some outlets citing March and others April 2026. Nvidia is already one of Reflection’s largest shareholders, having put in 800 million dollars as part of that round.

Why it matters

Nvidia buying or absorbing one of its own portfolio companies would be an unusual move for a chip maker whose business model depends on every AI lab buying its hardware rather than competing with it on models. An acqui-hire structure — hiring Reflection’s staff and licensing its technology without a full merger — is explicitly one of the options under discussion, and would let Nvidia gain the team and the model without the regulatory scrutiny a straight acquisition of a 25-billion-dollar startup would invite.

Nvidia has used this acqui-hire structure before, reportedly with Groq in December, and already runs its own open-model line, Nemotron. A Reflection deal in any form would extend that pattern of Nvidia moving from pure hardware supplier toward also owning pieces of the model layer that runs on its chips.

Nothing here is confirmed by either company

Reflection AI declined to comment on the talks. Reuters, which republished the FT’s reporting on its wire, stated it could not independently verify the Financial Times account. No party named in the story — Nvidia, Reflection AI, or the FT’s own sources — has put a number, a term sheet, or a timeline on record beyond “within weeks” and “could still fall apart.” The entire chain of public reporting on this story, including every outlet covering it since Saturday, traces back to the FT’s single sourcing; no second outlet has reported this independently of the FT.

This matters because the options on the table are structurally very different outcomes for anyone trying to assess what happens to Reflection’s open-weight model, Beam, which the company released as its first public coding-focused model this year. A full acquisition folds Reflection into Nvidia; an acqui-hire could leave the Beam model and its licence in limbo; a further equity round changes nothing about Reflection’s independence at all. Treating any one of these as settled would be a mistake the sourcing does not support.

What this means for buyers

Enterprises currently evaluating or running Reflection AI’s Beam model for coding workloads should not treat this report as a signal to change course yet — the options under discussion range from no change at all (a further investment round) to a full change of ownership, and the FT’s own sourcing says terms have not been reached. The practical move is to wait for either company to confirm something on the record, which neither has done.

The report is also a reminder that Nvidia’s economic interest increasingly extends past chip sales into the software and model layer, a trend 18BYTE has also tracked in Nvidia’s reported server price increases for 2027 shipments. Buyers negotiating long-term Nvidia hardware contracts should factor in that the company’s commercial incentives are no longer limited to selling GPUs.

What would change our reading

This would move toward CLEAR if either Nvidia or Reflection AI confirmed the talks on the record, or if a second outlet reported independent sourcing rather than relaying the FT’s account. It would move toward HOLD, or off our radar entirely, if either company denied the talks or if several more weeks passed with no deal and no further reporting, which given the FT’s own framing of early-stage and collapsible talks is a realistic outcome. We would also revisit this if the structure under discussion narrowed from the current four options (full acquisition, acqui-hire, further investment, or more compute) to one specific term sheet, since that would be the point an unconfirmed talks story turns into an actual transaction to evaluate.