Nvidia considering major move into open AI model

According to Financial Times, reported by Reuters and Seeking Alpha among others, Nvidia is in early negotiations to buy or significantly expand its investment in Reflection AI — a US developer of so-called "open-weight" models, meaning AI models where the underlying weights are publicly available (Seeking Alpha, FT).

The company was founded in March 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, the latter known as co-creator of AlphaGo. Reflection AI has in a short time gone from a valuation of around $550 million in early 2025, via $8 billion late last year, to $25 billion following a round in March 2026 in which JPMorgan Chase, among others, is said to have been involved.

Nvidia in talks to buy AI startup worth $25 billion - Bilde 1

Nvidia already a major shareholder

Nvidia currently holds an $800 million stake in Reflection AI and is considered one of the startup's largest shareholders. According to FT, the talks are still at an early stage, and a potential transaction could come within the coming weeks — or fall apart entirely.

Three different models are said to be on the table:

  • Acqui-hire: Nvidia brings in the key team and secures licensing rights to the technology, without carrying out a full company merger. FT writes that this option is primarily being explored to avoid lengthy antitrust and regulatory processes.
  • Full acquisition: Nvidia buys the entire company at or above the current valuation of $25 billion.
  • Expanded ownership stake: Nvidia deepens its existing investment and gives Reflection AI priority access to chip capacity and cloud services.

    • $25bn
    Valuation March 2026
    $800m
    Nvidia's current stake
    $150m
    Monthly compute cost
    Nvidia in talks to buy AI startup worth $25 billion - Bilde 2

    Enormous computing costs

    Reflection AI spends very large sums on computing. The company has entered into an agreement with SpaceX's Colossus 2 facility worth up to $6.3 billion through 2029, with a reported cost of $150 million per month for use of Nvidia's GB300 systems. In addition, the company has a computing agreement with Nebius worth over $1 billion.

    This expertise and capital requirement is likely a key driver behind Nvidia's interest — a potential acquisition or closer partnership could secure Nvidia a strategic position within open AI models, while also tying a major customer even more closely to the company's own hardware.

    An $800 million investment could grow into an acquisition worth over $25 billion

    Geopolitical dimension

    According to FT, the Trump administration views Reflection AI as an important national resource — an American counterweight to cheap, open Chinese models from the likes of DeepSeek, Moonshot and Z.ai. Reflection AI also has ongoing projects with the US Department of Defense (DoD) and Department of Energy (DoE), which could further complicate a potential transaction from a regulatory standpoint — and at the same time explain why an "acqui-hire" model appears attractive in order to avoid a full antitrust review.

    Nvidia CEO Jensen Huang has previously advocated for a strong American ecosystem of open AI models, stating that such a foundation is necessary to "create opportunities for innovation and prosperity." Neither Nvidia nor Reflection AI has officially commented on the talks.

    Source criticism

    It is important to emphasize that the information stems from anonymous sources cited by Financial Times, and that the talks, according to the paper itself, are at an early stage. Neither the amount, structure, nor timing of a potential deal has been confirmed by the parties. Reflection AI's flagship model Beam — a model with 501 billion parameters trained on around 10,500 Nvidia GB300 chips — nonetheless underscores how capital-intensive the development of competitive open models has become, and provides some context for why a player like Nvidia might see strategic value in a closer bond with the company.

    For the Norwegian market, the story is primarily relevant as an indicator of how concentrated and capital-intensive global AI infrastructure has become, with ripple effects for demand for Nvidia chips and data center capacity — a topic that also indirectly affects European and Nordic technology and energy companies through demand for electricity and cooling capacity for data centers.