Nvidia in talks to buy or deepen its stake in Reflection AI

Nvidia may be about to go one step further in its push into open AI models. According to a report by the Financial Times published on Saturday, October 10, 2026, the chipmaker is in talks to acquire or deepen its investment in Reflection AI, a US startup that builds open-weight models. Reuters picked up the report the same day, citing the FT and its people with direct knowledge of the matter.

The talks are at an early stage. Nothing has been signed, no price has been reported, and both companies declined to comment to the FT. Still, the story matters for anyone who follows open models, because Reflection is one of the very few well funded American labs that releases model weights, and Nvidia is already its most important strategic backer.

This article explains what was reported, what the different deal structures would mean, why Nvidia cares so much about open-weight models, and what to watch over the coming weeks.

What the FT reported

The core of the report is simple. Nvidia and Reflection are discussing ways to tie the two companies closer together. According to the FT, as summarized by Mint and Reuters, the options on the table include:

• a full acquisition of Reflection by Nvidia,
• an acqui-hire, where Nvidia hires Reflection's staff and licenses its technology instead of buying the company outright,
• a further equity investment, which would raise Nvidia's existing stake,
• an agreement under which Nvidia supplies additional chips and computing capacity.

The FT reported that an agreement could be reached in the coming weeks. The same people also cautioned that the discussions could still fall apart. Reuters said it could not immediately verify the report, and Nvidia and Reflection did not respond to its requests for comment outside business hours.

That is the full extent of what is on the record. Everything else in this article is context: what we know about Reflection, what Nvidia has done in similar situations, and why the structure of a possible deal is more interesting than the headline.

Who is Reflection AI?

Reflection AI was founded in 2024 by Misha Laskin, who is the chief executive, and Ioannis Antonoglou. Both are former Google DeepMind researchers and, according to Mint, both worked on Google's Gemini models. Reuters describes Reflection as a company that develops tools to automate software development, one of the fastest growing use cases for AI, and that has repositioned itself as a builder of open-weight frontier models.

Open-weight means that the trained model parameters are published, so that companies, researchers and governments can download the model, run it on their own hardware and fine tune it on their own data. That is the opposite of a closed API model such as GPT or Claude, where customers only rent access.

The company is young, but it has raised money and signed deals at a scale that few startups can match.

• In March 2026, the Wall Street Journal reported that Reflection was raising $2.5 billion at a $25 billion pre-money valuation.
• In April 2026, Laskin confirmed on CNBC that the round had closed at that $25 billion pre-money valuation.
• In May 2026, Reflection was named as a model provider for the US government's Genesis Mission, according to its newsroom. The FT also reported partnerships with the Pentagon and the Department of Energy, plus agreements to develop models for US allies including South Korea.
• In June and July 2026, Reflection signed compute deals with SpaceX (worth up to $6.3 billion) and Nebius ($1 billion of AI capacity).
• On October 5, 2026, Reflection launched Beam, its first open-weight model.

Investors include Sequoia Capital, Disruptive and 1789 Capital, where Donald Trump Jr. is a partner, according to the FT as cited by Mint. Nvidia is the most prominent strategic investor.

Beam, briefly

We covered Beam in detail when it launched, in our article Reflection unveils Beam, a 501B open-weight model. The short version: Beam is a mixture of experts model with 501 billion total parameters and 23 billion active per token, built for coding, reasoning and agent work, and trained on Nvidia GB300 systems. Reflection claims it is competitive with Z.ai's GLM 5.2 while using much less compute per answer, and it promised to release the weights under the Apache 2.0 license.

For this story, Beam matters for one reason. It turned Reflection from a well funded promise into a company with a shipping product, and it did so only five days before the FT report. Any buyer now has a concrete model to evaluate, not just a team and a pitch deck.

How much Nvidia already has at stake

According to the FT, Nvidia has already invested $800 million in Reflection. That makes it a major financial backer and a strategic partner, but not a controlling owner.

Put next to the last reported valuation, Nvidia's investment is small: about 3 percent of $25 billion. A full acquisition would therefore be a very different commitment from what Nvidia has done so far. We do not know what price is being discussed, and the valuation of a private company can move quickly in either direction. The $25 billion figure is the last public reference point, not a deal price.

For Nvidia, the money itself is not the obstacle. Yahoo Finance notes that Nvidia generated roughly $127 billion in free cash flow over the past four quarters, so even a purchase near the last valuation would be affordable. The harder questions are about regulation, customers and strategy.

Four ways the talks could end

The most interesting part of the report is the list of possible structures. They are not small variations. Each one would send a different signal to the market and would face a different level of scrutiny.

Full acquisition. Nvidia would buy all of Reflection. This is the cleanest option and the biggest signal: Nvidia would own a US frontier open-weight lab outright. It would also be the option most likely to trigger a long merger review, because Nvidia already dominates AI chips and would be buying a company that is also its customer.

Acqui-hire with a technology license. Nvidia would hire Reflection's researchers and engineers and license its technology, without buying the company itself. Reuters notes that this kind of structure can avoid a lengthy regulatory review. Nvidia has used it before. In December 2025, it struck a deal with chip designer Groq in which it licensed Groq's technology and hired its chief executive, a transaction CNBC valued at about $20 billion. According to the FT as cited by Mint, that arrangement later became the subject of a lawsuit brought by former Groq employees. Acqui-hires are faster, but they are not free of legal or antitrust risk.

More equity. Nvidia could simply write a bigger check and raise its stake. Reflection would stay independent and could keep raising from other investors. This is the least dramatic outcome and probably the easiest one to close quickly.

More chips and compute. Nvidia could provide additional GPUs and computing capacity, either as part of an investment or as a supply agreement. For a lab that trains models on tens of thousands of GPUs, guaranteed access to hardware can be worth as much as cash. For Nvidia, it would deepen the relationship without changing ownership.

The column on regulatory exposure in the chart above is our own reading, not something the FT reported. But it explains why the structure will matter more than the price.

Why Nvidia wants open-weight models

At first glance, it looks odd for a chipmaker to buy a model developer. Nvidia sells hardware to every AI lab in the world, including OpenAI, Anthropic, Google and the Chinese labs that are allowed to buy it. Owning a lab could look like competing with its own customers.

The logic becomes clearer when you look at what Nvidia has done this year.

• Nvidia already builds its own family of open models, Nemotron.
• On September 2, 2026, Nvidia agreed to acquire Hugging Face, the largest hub for open models and datasets, for about $12.93 billion. In its 8-K filing, Nvidia committed to keep the platform open and to keep supporting other silicon vendors. That deal is expected to close in the first half of 2027, subject to regulatory approval.
• In July, chief executive Jensen Huang called for a domestic US open-weight AI ecosystem, saying it was important for innovation and economic growth, according to Mint.

Open models are good for Nvidia's core business. Every company that downloads an open model and runs it on its own servers needs hardware to do so. Closed models, by contrast, concentrate demand in a handful of large cloud providers, several of which are building their own chips. A strong open ecosystem spreads demand across thousands of enterprises, governments and smaller clouds, which is exactly where Nvidia sells most easily.

Reflection fits that picture. It trains its models on Nvidia systems, it pitches what it calls "AI factories" to institutions that want to run models on their own infrastructure, and its first product, Beam, runs on Nvidia chips. Mint sums it up well: a deal would give Nvidia deeper exposure to open-weight models, talent and technology that may generate further demand for its computing infrastructure.

The politics: US open models versus cheap Chinese alternatives

There is also a geopolitical layer. The FT describes Reflection as a developer of open-weight models that the Trump administration hopes will rival cheap Chinese alternatives. Over the past two years, the strongest open models have often come from China, from labs such as DeepSeek, Moonshot AI with Kimi, Alibaba with Qwen and Z.ai with GLM. Laskin himself said on CNBC in April that the best open models were coming from China.

Washington wants an American answer, and Reflection has positioned itself as exactly that. Its government work with the Pentagon, the Department of Energy and the Genesis Mission, plus its deals with US allies such as South Korea, make it more than a normal startup. That cuts both ways for a potential deal. On one hand, the administration may welcome an American champion with Nvidia's resources behind it. On the other hand, regulators and competitors may be uneasy about the dominant AI chip company also controlling one of the few US frontier open-weight labs.

What it means for developers and customers

If you use or plan to use Beam, the immediate impact is limited. Reflection said it would release Beam's weights under Apache 2.0 this month, and once weights are published under that license, the license cannot simply be taken back. The bigger questions concern what comes after:

• Will future models stay open? Under Nvidia, the strategic incentive is to keep models open, because open models sell hardware. That is a reasonable bet, but it is not a guarantee.
• Will they stay hardware neutral? Nvidia promised to keep Hugging Face open to other silicon vendors. It is unclear whether a Reflection deal would come with a similar commitment for Reflection's own software stack.
• What happens to compute deals? Reflection's agreements with SpaceX and Nebius are for Nvidia hardware. A closer tie with Nvidia would probably make supply easier, not harder.
• What about rival labs? Other model developers that buy Nvidia chips might start to worry about competing with a lab that has privileged access to the supplier.

What to watch next

The FT said an agreement could come in the coming weeks, and that talks could also collapse. A few signals will show which way it is going:

1. The structure. A full acquisition would be a strategic statement. An acqui-hire or a bigger stake would suggest that Nvidia wants the talent and the relationship without the regulatory fight. 2. Regulators. Nvidia's Hugging Face deal is already facing questions about merger review in Europe, according to MLex. A second open-model deal in the same autumn would draw more attention. 3. Beam's weights. Reflection promised the weights for October. Whether that release goes ahead on schedule, and under the promised license, will be the first concrete test of its independence. 4. Other investors. Sequoia, Disruptive, 1789 Capital and the backers of the March round would all need to agree to any deal that changes control.

Until there is an announcement, this remains a report about early talks. But it fits a clear pattern. In a little over a year, Nvidia has licensed Groq's technology, agreed to buy Hugging Face and is now discussing a deeper tie with a frontier open-weight lab. The company that sells the shovels is increasingly building the mine as well.

Source: [Financial Times](https://www.ft.com/content/052610c5-22b4-4dd4-932e-b7f9f0628b6a), with additional reporting from [Reuters via Investing.com](https://www.investing.com/news/stock-market-news/nvidia-in-talks-to-invest-further-in-reflection-ai-or-buy-it-ft-reports-4942221) and [Mint](https://www.livemint.com/ai/amazon-nvidia-step-up-ai-bets-inside-the-reported-7b-decart-deal-and-reflection-ai-talks-11791994753830.html).