Reflection AI, a New York-based artificial intelligence startup founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, is in advanced talks to raise approximately $2.5 billion at a target valuation of $25 billion, according to reporting by the Wall Street Journal and TechStartups citing people with knowledge of the discussions. If completed, the round would be one of the largest raises in history for an open-source AI project. NVIDIA had previously invested approximately $800 million in Reflection AI at an $8 billion valuation, meaning the new round would more than triple the company’s value from that reference point in a short period. JPMorgan Chase is reportedly considering participating in the new round through its venture unit — the same institution that co-led Shield AI’s $2 billion raise in the same week. Reflection AI is building open-source large language models and agentic AI systems specifically aimed at software development automation. The company’s open-source positioning is strategically significant: it explicitly targets the same market segment as Meta’s Llama, Mistral’s open-weight models, and the Chinese laboratory DeepSeek — all of which have demonstrated that competitive AI performance can be achieved at dramatically lower cost than the frontier closed models from OpenAI, Anthropic, and Google. The $2.5 billion raise, if completed, would fund the compute infrastructure needed to train models at a scale where open-source systems can compete directly with closed frontier models. The capital will also be used to attract leading AI research talent in what Reflection’s founders describe as an extremely competitive market for top researchers. TechStartups noted that the fundraising push comes explicitly amid a U.S. government effort to develop homegrown, efficient AI models as a strategic counterweight to foreign AI systems, particularly from China.
NVIDIA-Backed Reflection AI Eyes $2.5 Billion at $25 Billion Valuation — The Open-Source AI Lab Built to Rival DeepSeek and Meta’s Llama
37