Nvidia — the chip company at the center of the AI boom — is in talks to buy a startup called Reflection AI, or at least put a lot more money into it. The Financial Times reported it on October 10, and Bloomberg confirmed the talks are real.
Here’s the part worth understanding: Reflection AI makes “open-weight” AI models. That phrase gets thrown around a lot, so let me translate.
Most AI you use — ChatGPT, Claude — is closed. You type into someone else’s app, on someone else’s computers, and you pay whatever they charge. You can’t see how it works under the hood, and you can’t run it yourself.
Open-weight is the opposite. The company publishes the model’s “weights” — basically the recipe — under a license that lets anyone download it, tweak it, and run it on their own machines. Reflection’s new model, Beam, is going out under the Apache 2.0 license, which is about as open as it gets.
Why does that matter? Money, mostly. Running your own copy of an open model can be much cheaper than paying a closed service — especially for a business running millions of queries a day. And you can customize it: teach it your company’s data, your way of talking, without sending everything to someone else’s servers.
So why does Nvidia — a company that sells the chips — want to own the recipe too? Because right now Nvidia sells the ovens but doesn’t own a frontier-scale recipe of its own. It has an open model family called Nemotron, but nothing at the very top tier. Reflection’s Beam is a 501-billion-parameter model aimed at coding work, built by two former Google DeepMind researchers — and the company was valued at $25 billion earlier this year, roughly triple what it was worth last October, when Nvidia first put in $800 million.
There’s a bigger backdrop too. The White House has been pushing American open models as the Western answer to cheap Chinese ones like DeepSeek. Nvidia’s CEO Jensen Huang wrote an essay in July calling for a homegrown open AI “ecosystem,” arguing openness might be one of the most important paths to AI safety.
The deal itself might not even be a normal buyout. One option being discussed is an “acqui-hire”: Nvidia hires the researchers and licenses the technology without formally buying the company — a structure that can dodge a long regulatory review. Nvidia did something similar in December with the chip startup Groq, in a deal reported at $20 billion.
And it might not happen at all. The talks are early, and the FT’s sources say they could still fall apart. As of today, nothing is signed.
But the direction is clear either way. The chip giant thinks the future isn’t just selling shovels — it’s owning a piece of the gold mine too. And whether AI stays open or gets locked up behind paywalls is a question that affects everyone who’ll ever use it — which is to say, everyone.
Sources: Reuters via Financial Times (Oct 10, 2026); Startup Fortune (Oct 11, 2026); SiliconANGLE (Oct 11, 2026); The Gulf Gazette via Mobile World Live (Oct 14, 2026); EM360Tech (Oct 14, 2026); PYMNTS (Oct 11, 2026).