Skip to content
SPY$778.57up +0.60%
Nasdaq$751.27up +0.49%
Nvidia$229.28down −0.52%
SpaceX$162.57up +1.25%
VIX14.84down −3.84%
BTC$82,957.00up +0.60%
Gold$4,183.52up +0.00%
Silver$60.37down −1.14%
Brent oil$104.54up +0.82%
DXY102.19down −0.17%
EUR/USD1.1203up +0.00%
USD/JPY158.30down −0.01%
2Y yield4.76%down −1 bp
5Y yield5.02%up +3 bp
10Y yield5.24%up +2 bp

Search FinPrism

Companies · · 2 min read

Nvidia in talks to buy open-model AI start-up Reflection, FT reports

Nvidia is discussing a takeover of Reflection AI, a US maker of free-to-download AI models it already backs, the FT reported, a step that would put the chip giant into the model business.

Nvidia in talks to buy open-model AI start-up Reflection, FT reports

The short answer

The Financial Times (FT) reported that Nvidia is in talks to acquire Reflection AI, a US start-up building open AI models meant to rival cheap Chinese options such as DeepSeek. Nvidia already has about $800 million invested in Reflection, valued at $25 billion before its latest round. A deal would make Nvidia an owner of a model developer, not just a supplier.

What’s going on here?

Nvidia is in talks to acquire Reflection AI, the Financial Times (FT) reported on Saturday, adding that the Trump administration hopes the start-up can compete with low-cost Chinese models such as DeepSeek. Nvidia has not publicly confirmed any talks. Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. Nvidia already holds roughly $800 million of the company, and the two work together on chip supply and through Nvidia's Nemotron Coalition of open-model developers. On October 5, Reflection released Beam, its first model, with 501 billion parameters.

What does this mean?

Open-weight models are AI systems whose inner settings, called weights, can be downloaded, run on a company's own hardware and customized. Closed models such as ChatGPT or Claude are rented through the developer instead. The strongest open models have come mostly from China, Axios reported, which is why Washington has been keen to see an American alternative. Reflection's pitch is aimed at Western businesses and governments that want a model they can control without relying on one of the big proprietary labs.

For Nvidia, the logic is about chips. When an organization runs an open model itself, it needs its own computing power, and that usually means buying or renting Nvidia GPUs. Reflection's plans already lean that way: it has agreed with South Korea's Shinsegae Group to build a 250-megawatt data center that will run on Nvidia technology, and it has a computing deal worth more than $1 billion with Nebius, according to Crypto Briefing. Owning the lab would let Nvidia shape a flagship open model around its own hardware.

The valuation has climbed fast. Reuters reported in September 2025 that a funding round could value Reflection at up to $5.5 billion. By March 2026 the company was in talks to raise $2.5 billion at a $25 billion pre-money valuation, a phrase meaning the value before new cash comes in. A full purchase would therefore be a large deal, even for a company of Nvidia's size, and any price has not been reported.

There is a strategic risk too. Nvidia sells chips to almost every major AI lab. Buying one of them would turn Nvidia into a competitor of some of its biggest customers, which could make them more willing to back rival chip designs.

The bull case

Owning Reflection could give Nvidia a showcase open model built to run best on its chips, nudging governments and companies that want their own AI systems toward Nvidia hardware. With political support for a US answer to DeepSeek, and existing deals such as the South Korea data center, the purchase could lock in long-term demand for Nvidia GPUs beyond the handful of big labs.

The bear case

Reflection has released only one model, and its $25 billion pre-money valuation is steep for such an early company. Its computing commitments are large, and some, such as a potential arrangement with SpaceX, are not final. Buying a model lab could also unsettle Nvidia's largest customers, which already compete with each other and may not welcome their chip supplier as a rival.

Why should I care?

For markets:

Nvidia shares could react to any confirmation and the price of a deal, while AI infrastructure partners such as Nebius may also move on news about Reflection's future. Broader AI stocks are sensitive to signs of how the open-model race is shaping up.

The bigger picture:

Nvidia is one of the largest weights in major US stock indexes, so its deals ripple into many retirement and index funds. The talks also show how AI is becoming a matter of national policy, not just corporate strategy.

Market impact

Asset (ticker)Potential directionTimeframeConfidenceReason
Nvidia (NVDA) ↔ neutral Long term Low A purchase could boost chip demand but risks friction with major AI lab customers.
Nebius Group (NBIS) ↔ neutral Short term Low Nebius has a computing deal with Reflection worth more than $1 billion.

Potential impact, not investment advice.

Frequently asked questions

Is Nvidia buying Reflection AI?

Nvidia is in talks to acquire Reflection AI, according to the Financial Times. No deal has been announced, and Nvidia has not publicly confirmed the discussions. Nvidia already owns a stake in the start-up after investing about $800 million, and the two work together on chip supply and open models.

What is Reflection AI?

Reflection AI is a US start-up founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. It builds open-weight AI models that anyone can download and customize, aimed at rivaling Chinese models such as DeepSeek. It released its first model, Beam, on October 5, 2026.

Why would Nvidia want to own an AI model company?

Companies that run open models on their own systems need computing power, usually Nvidia chips. Owning Reflection could help Nvidia promote a leading US open model tuned for its hardware. The trade-off is that Nvidia would compete with AI labs that are among its biggest customers.

Sources: Financial Times, Crypto Briefing, Axios via Yahoo Finance, Reuters via Yahoo, Shopifreaks

Back to the latest