Nvidia in talks to buy Reflection AI or raise its stake, FT reports

Nvidia is in talks to acquire Reflection AI or to increase its investment in the open-weight model developer. The Financial Times reported this on Saturday, citing people familiar with the situation. According to the publication, the talks are at an early stage and may come to nothing. TNW has not independently confirmed the report. For scale: in October 2025, Reflection closed a $2 billion round at a valuation of about $8 billion, and Nvidia, according to media reports, was among the main participants.

What is known about the deal

Public details are scarce, and that is worth keeping in mind when reading the headlines. Neither the amount, the structure, nor the timing has been named. It is not even clear which of the two scenarios looks preferable to the parties. A full acquisition would mean the largest supplier of AI accelerators takes an independent lab under its wing. An additional investment would preserve the current arrangement: Nvidia as shareholder and main compute provider.

The facts in brief:

• Source: the Financial Times, citing anonymous people familiar with the matter.
• Format: a full acquisition or an additional investment.
• Stage: early; the talks may collapse.
• Reflection's last known round: $2 billion at a valuation of about $8 billion (October 2025).
• Price, terms, and timing: not disclosed.
• Confirmation from Nvidia and Reflection AI: not included in the TNW report.

Why Nvidia wants open models

Reflection was founded by DeepMind alumni Misha Laskin and Ioannis Antonoglou. The company positions itself as an American lab that releases frontier open-weight models. It is a response to the wave of Chinese open models such as DeepSeek and Qwen, which have won a notable share among developers over the past couple of years. Reflection's early product focused on coding: Asimov, an agent for working with codebases.

Nvidia's logic is clear even without insider information. Open models are trained and run on its GPUs, and the more widely they are used, the greater the demand for hardware and for the CUDA software stack. The company also already has its own open models, and a strong team for training frontier systems would be a valuable asset. Nvidia has also long invested in labs and startups across the AI chain, so another investment fits its usual strategy.

An acquisition, however, would be costlier and more complicated than an ordinary stake. Nvidia is already under close regulatory scrutiny because of its dominance in the accelerator market. Buying a model lab would give ammunition to those who point to a conflict of interest: a compute supplier would own a model developer that competes with Nvidia's own customers. Such concerns are unlikely to disappear even with an additional investment, although they would be milder in that case.

What developers and companies should watch

For now this is a rumor with caveats, and it is too early to rework plans. But a few things are worth tracking.

First: if Reflection becomes part of Nvidia, its licensing policy and the openness of its weights could change. Teams that build products on open models would do well to review licensing terms in advance and to have a fallback among alternative model families.

Second: hardware optimization. Models released under the control of a chipmaker generally run best on that company's accelerators. For those who deploy inference on other platforms, this is a reason to compare performance on their own workloads rather than on published benchmarks.

Third: follow official statements and regulatory filings. If the talks reach a deal, the details will appear in confirmed form, and until then it is better to draw conclusions with caution.