Nvidia is in talks to acquire, or substantially increase its investment in, the open-source AI lab Reflection AI. According to reports by the Financial Times and Reuters on October 10, the company, founded in 2024 by two former DeepMind researchers, Misha Laskin and Ioannis Antonoglou, was most recently valued at about $25 billion. Nvidia already invested $800 million in it last October.
Key points:
- An open-source standard-bearer: Last week Reflection released Beam, an open-source mixture-of-experts (MoE) model with 501 billion parameters, of which only 23 billion are activated per forward pass. It claims inference costs of only a quarter to a third of comparable models, with performance matching China's Z.ai GLM-5.2, and Washington sees it as the Western open-source answer to DeepSeek.
- Deal structure: Talks are still at an early stage and may take the form of an acqui-hire (taking the team and licensing the technology) to avoid a lengthy antitrust review, the same approach Nvidia used earlier with the AI chip company Groq.
- Strategic meaning: The chip giant would be taking hold of the open-source route itself, moving from “selling shovels” to “setting standards”. However, the deal could be reached within weeks or could fall through; neither side has commented.
Source: read the original. This text is a summary for reference only.
