Nvidia weighs more investment in Reflection AI or acquisition, FT reports

Nvidia is in talks to increase its investment in Reflection AI or acquire the startup, according to a Financial Times report carried by Reuters on October 10. The reported discussions would deepen the chipmaker’s involvement with a company developing software-automation tools and introducing Beam, a model aimed at coding and agentic tasks.

The negotiations remain unconfirmed by the companies in the Reuters account. Reuters said it could not immediately verify the FT report, and that Nvidia and Reflection did not immediately respond to requests for comment outside regular business hours. The account establishes neither an agreed investment nor an acquisition.

What the reported Nvidia talks could involve

According to the FT account relayed by Reuters, the discussions are at an early stage and could produce several different arrangements. One possibility is an acqui-hire, under which Nvidia would hire staff and license technology rather than purchase Reflection outright. No staffing decision has been announced in the report.

The FT suggested that such a structure could avoid a lengthy regulatory review. That is a reported rationale for one possible arrangement, rather than a determination by a regulator about how a transaction would be treated. The eventual structure remains unresolved.

Nvidia’s existing investment in Reflection AI

Nvidia is already a major financial backer and strategic investor in Reflection, according to the FT account, which put its investment at $800 million. That figure describes the reported existing relationship; it is not a proposed acquisition price or the amount of a new investment.

Reuters also cited chief executive Misha Laskin’s statement to CNBC in April that Reflection was raising fresh capital at a $25 billion pre-money valuation. That historical fundraising figure does not establish what Nvidia might pay, or show that the fundraising was completed.

Reflection was founded in 2024 by former DeepMind researchers Laskin and Ioannis Antonoglou. Reuters describes the company as developing tools that automate software development, placing the reported talks alongside its efforts to build models for coding and agentic tasks.

Beam’s introduction and planned weights release

Reflection introduced its first open-weight model, Beam, on Monday, October 5, according to Reuters. The report describes the company as seeking to compete in coding and agentic tasks with lower-cost Chinese models, including DeepSeek and Kimi. It provides no independent performance evaluation of Beam.

Axios reported on October 6 that Reflection planned to release Beam’s weights later that month, together with tools for running, evaluating and fine-tuning the model. Its reporting places the planned availability of those materials after Beam’s introduction.

The introduction therefore does not by itself establish that downloadable weights were already available. The dated Axios account describes a release plan, while Reuters describes the model’s launch; neither account cited here establishes that the subsequent weights release had taken place.

Reflection’s open-model strategy

On its company page, Reflection describes its mission as making intelligence “open and accessible to all.” It identifies enterprises, governments and sovereign entities as intended users, and says its systems are designed to give those customers control over their intelligence.

The company sets out three commitments: releasing model weights, publishing research and technical reports, and open-sourcing software that allows others to customise its models. The software commitment includes reinforcement-learning tools and environments. These are stated commitments, rather than confirmation that every component has shipped.

Reflection also argues that opening systems to inspection allows more researchers to identify vulnerabilities and improve safeguards. This is the company’s explanation of its approach to safety, not an independently demonstrated finding that its models are safer than alternatives.

Existing technology ties and unresolved questions

Reflection says it is validated on Dell AI Factory with NVIDIA for enterprise deployment, including proprietary-data customisation and on-premises control. The company’s description supplies another example of its existing technology relationship with Nvidia, separate from the reported investment discussions.

It also says it is building a sovereign AI cloud for South Korean enterprises and agencies around a 250MW data centre using Reflection models and Nvidia GPUs. Its undated company page does not establish that the project is complete or provide an operating date.

The reported talks leave the consequences for those users unresolved. No terms have been established for changes to staffing, customer access, model licensing or ownership. For now, Reflection’s published open-model commitments and product plans remain distinct from the possible transaction described by the FT.