Nvidia in talks to buy or deepen its stake in Reflection AI after Beam model launch

Nvidia is in talks to either buy Reflection AI or pour more money into it, according to a Financial Times report published on Saturday, October 10. The Nvidia Reflection AI talks are described as early, and they could still fall apart, but the news lands just days after the Brooklyn startup released Beam, its first open-weight model. If a deal happens, the world's most valuable chipmaker would own a lab that builds the kind of large AI models its customers usually build on its hardware.

Reuters, which picked up the FT story, said it could not independently verify the discussions. Reflection declined to comment on the report, and Nvidia did not immediately respond to requests for comment. So treat everything below as reported talks, not a signed agreement.

What happened

According to the FT, which cited people with direct knowledge of the matter, Nvidia is weighing three paths:

• A deeper investment in Reflection, on top of what it has already committed.
• A full acquisition of the company.
• An acqui-hire, in which Nvidia would hire Reflection's staff and license its technology instead of buying the company outright.

The report said an agreement could come within weeks. It also said the talks are at an early stage and could collapse.

Nvidia is not a stranger here. The FT reported that Nvidia has already invested $800 million in Reflection and is one of its major financial and strategic backers. That makes this less of a surprise bid and more of a question about how close the two companies want to get.

Who is Reflection AI?

Reflection was founded in 2024 by Misha Laskin and Ioannis Antonoglou, two former Google DeepMind researchers. The company started out building tools that automate software development, and it has since moved into training its own large models.

According to Silicon UK, Reflection has raised about $4.7 billion from investors including Nvidia, Sequoia Capital and Lightspeed. Laskin told CNBC in April that the company was raising new money at a $25 billion pre-money valuation. That price tag matters, because it helps explain why an acqui-hire is on the table at all. Buying a company at that size invites a long antitrust review. Hiring the team and licensing the technology is a faster route that has become common in the AI industry.

The Beam model

Reflection launched Beam earlier this week. It is the company's first open-weight model, which means developers can download the model weights and run them on their own hardware instead of only reaching it through a paid API.

Silicon UK reported these details about Beam:

• It is a text-only mixture-of-experts model.
• It has 501 billion total parameters, with 23 billion active at any one time.
• It is aimed at reasoning, coding and agentic tasks.

Reflection claims Beam matches the leading open-weight models from China and beats the leading Western open models while using three to four times less inference compute. The company says it performs on par with Z.ai's GLM-5.2, a model with roughly 744 billion parameters and 40 billion active. Those claims come from Reflection itself and have not been independently verified, so they are worth watching as outside testers get their hands on the model.

If you are new to how agent-style models work, our explainer on what AI agents can and cannot do covers the basics.

Why the Nvidia Reflection AI deal matters

Nvidia wants a seat in the model race

Nvidia sells the chips that almost every major AI lab trains on. Owning a frontier model team would push it further up the stack, into the software layer where companies like OpenAI, Anthropic and Google compete. Reuters noted that a deal would expand Nvidia's reach into AI software development beyond its dominant position in processors.

There is a business logic to this. Open-weight models are downloaded and run by companies on their own servers. Every company that runs a large model locally needs serious hardware to do it. A strong American open-weight model, tuned to run efficiently, is a good reason for businesses to buy more GPUs.

The open-weight fight with China

Reflection has openly pitched Beam as an American answer to low-cost Chinese open models such as DeepSeek and Kimi. Chinese labs have released many of the most capable open-weight models over the past two years, and that has worried US policymakers and some American tech companies. A well-funded US lab shipping competitive open weights changes that picture a little.

For developers, more strong open models usually means lower prices and more choice. We have seen the same pressure play out in closed models too, from Anthropic's steep Claude Haiku 5.5 price cut to OpenAI shipping its cheaper Sol model.

Why an acqui-hire keeps coming up

Big tech firms have leaned on acqui-hires in AI for a reason. A traditional takeover of a company valued in the tens of billions would likely face months of regulatory scrutiny. Hiring the founders and engineers and licensing the technology gives the buyer most of what it wants while skipping much of that review. The FT specifically flagged this structure as an option that could avoid a lengthy regulatory process. Regulators in the US and Europe have started looking harder at these arrangements, so the structure is not a free pass.

Who is affected

Developers using Beam. Beam is open weight, so the copies people have already downloaded stay usable no matter what happens to the company. The bigger question is future releases. A new owner could keep shipping open weights, change the license on new versions, or fold the team into internal work.

Other AI labs. Nvidia supplies chips to nearly everyone in the industry. A chipmaker that also owns a competing model lab could make some customers uneasy, especially if they worry about access to scarce hardware.

Investors. Reflection's backers, including Sequoia and Lightspeed, would be affected by whatever price or structure is agreed. Nvidia's share price barely moved on the report, according to Investing.com, which showed the stock down about half a percent at the time.

What you can do now

If you build with open models, nothing changes today. Beam is out, and you can test it against the models you already use. A few practical steps:

1. Run Beam on your own tasks instead of relying on launch claims. Coding and agent workflows vary a lot from one team to the next.
2. Check the license terms for the version you download and keep a copy of the weights you depend on.
3. Avoid building critical systems around a single model provider while ownership is uncertain.

Businesses that are watching AI costs closely may also want to read how Google recently limited Gemini free users to Flash-Lite, a reminder that access terms can change quickly. For broader coverage of the AI industry, visit our AI news section.

What to watch next

• A signed deal. The FT said an agreement could come within weeks. Watch for which structure Nvidia chooses.
• Regulatory response. An acqui-hire of a $25 billion lab would draw attention from antitrust officials, even if it avoids a formal merger review.
• Independent Beam benchmarks. Outside testing will show whether Reflection's efficiency claims hold up.
• Nvidia's broader spending. Nvidia has been funding partners across the AI supply chain, and its chips sit behind deals like SpaceX's $40 billion borrowing for Nvidia hardware.

Frequently asked questions

Is Nvidia buying Reflection AI?

Not yet. The Financial Times reported that Nvidia is in early talks to acquire Reflection, invest more, or do an acqui-hire. No deal has been announced, and the talks could fall apart.

How much has Nvidia invested in Reflection AI?

According to the FT, Nvidia has invested $800 million in Reflection and is already a major strategic backer of the startup.

What is Reflection's Beam model?

Beam is Reflection's first open-weight model, launched this week. It is a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active, aimed at reasoning, coding and agentic tasks.

What is an acqui-hire?

An acqui-hire is a deal where a large company hires a startup's staff and licenses its technology instead of buying the whole company. It can avoid some of the lengthy regulatory review that a full takeover would face.