Open scale models are becoming increasingly popular due to their affordability. Nvidia may absorb one of the American developers of such models – the startup Reflection AI, which is trying to compete in this segment with Chinese companies.
Image source: Reflection AI
As has been noted more than once, in the face of rising costs for companies to use closed AI models, many of them began to turn to open source models or open scales, which are actively promoted by Chinese developers. US startup Reflection AI is also betting on open weights, making it a potential takeover target for Nvidia.
Edition Financial Times reported that the latter company is in negotiations either to increase investment in Reflection AI capital or to take it over. The administration of President Donald Trump believes that American models with open scales can compete with cheap Chinese alternatives.
Considering how Nvidia founder Jensen Huang and the current American president have grown closer in the wake of the AI boom, the company’s interest in such acquisition targets is understandable. Nvidia some time ago tried to organize something like an “AI bank”, the funds of which, totaling up to $500 billion, should be used to finance projects in this area. Startups do not have a serious credit history, while the development of AI models requires large expenses, and large investors like Nvidia and its partners can greatly help them with money.
Negotiations with Reflection AI are in the early stages, so a potential deal could take one of several forms, including poaching the startup’s staff and licensing its technology. This would avoid increased regulatory attention to the Reflection AI acquisition deal. True, a similar scheme implemented in the case of the startup Groq is already being challenged by its former employees in court, so it still entails certain legal risks.
An alternative would be for Nvidia to invest more, since the company is already one of Reflection AI’s largest shareholders. In addition, Nvidia’s own AI accelerators are the “capital” that is highly sought after by startups, and in times of scarcity, it is valued almost more than cash. Nvidia also has access to ready-made computing power through its many partners, and can share this resource with Reflection AI.
Nvidia has already invested $800 million in Reflection AI, and among other institutional investors there is the 1789 Capital fund, in which the son of the American president, Donald Trump Jr., is a partner. If negotiations between Nvidia and Reflection AI lead to any deal, it will be concluded in the coming weeks. In March of this year, the startup’s capitalization reached $25 billion. How much Nvidia is ready to pay for its assets is not specified.
Image source: Nvidia
For Nvidia, the segment of open-weight models is not something new; it has developments from the Nemotron family, but cannot boast of its own advanced AI models. Open scale models allow enterprise customers to run them on their own computing infrastructure. This not only saves money, but also helps protect corporate data from potential leakage through the cloud.
Jensen Huang made a call for American companies to create a national ecosystem of open weight models back in July. He stated that she would allow “create opportunities for innovation and prosperity across the country”. In September, Nvidia agreed to buy Hugging Face, a repository that stores millions of AI models and data sets. The deal cost Nvidia almost $13 billion, but having made a fortune from the AI boom, it now does not hesitate to actively absorb promising startups. The capitalization of Nvidia itself is steadily moving towards $6 trillion.
The company has already spent tens of billions of US dollars supporting projects from large AI labs like OpenAI, neo-cloud providers like Nebius and smaller market participants. Such investments are driving demand for its own AI chips, so it is even interested in expanding such activity. Reflection AI presented its Beam model with open scales this past week. She claims that in tests for complex logical reasoning, Beam shows results at the GLM-5.2 level of the Chinese company Z.ai and outperforms leading Western open-source models, while consuming 3-4 times less computing resources for inference operations.
The Reflection AI startup, founded by people from Google, has already made a name for itself thanks to its cooperation with the Pentagon and the conclusion of agreements with US geopolitical allies such as South Korea. Nvidia and Reflection declined to comment to the Financial Times on the ongoing negotiations.
If you notice an error, select it with the mouse and press CTRL+ENTER.
