Nvidia-backed Reflection AI in talks to raise $2.5 billion at $25 billion valuation

Reflection AI, a startup backed by Nvidia that is building freely available AI models, is in talks to raise $2.5 billion at a pre-money valuation of $25 billion, the Wall Street Journal reported on March 25.

The company is central to Nvidia's strategy of fostering an open-source AI ecosystem that runs on its chips and competes with increasingly capable Chinese alternatives. Investors have described Reflection as a Western counterpart to the open-source models coming out of Chinese labs, a gap that has raised concerns about U.S. allies defaulting to Chinese AI infrastructure.

JPMorgan Chase is in talks to participate through its Security and Resiliency Initiative, a fund created in December to back U.S. companies in industries considered critical to the economy and national security. Disruptive, an earlier investor, is also expected to join the round.

Reflection has yet to generate meaningful revenue, according to the Journal. Nvidia invested approximately $800 million in a previous round that valued the startup at $8 billion. The company has raised more than $2 billion to date, not including the latest financing, from investors including Lightspeed Venture Partners, CRV, Sequoia Capital, and 1789 Capital.

The startup recently committed several billion dollars to build Korean-language AI models in a deal with South Korean conglomerate Shinsegae Group, powered by Nvidia chips. Reflection plans to pursue similar arrangements with the goal of becoming the default open model used by sovereign cloud deployments in U.S.-allied countries.

Reflection is also part of the Nemotron Coalition, a group of startups Nvidia has assembled to develop open AI models. Other members include Cursor and Thinking Machines Lab.

What Reflection AI does

Reflection AI is building what it describes as superintelligent autonomous systems, starting with coding. The company's thesis, laid out in a March 2025 blog post, is that the breakthroughs needed to build a fully autonomous coding system, such as advanced reasoning and iterative self-improvement, extend naturally to broader categories of computer work.

The company was founded by two former Google DeepMind researchers. Co-founder and CTO Ioannis Antonoglou was a founding engineer at DeepMind and helped create AlphaGo. CEO Misha Laskin led reward modeling for the Gemini project. Their approach centers on scaling the autonomous capabilities of large language models with reinforcement learning.

The open-source framing is core to Reflection's positioning and to why Nvidia is backing it. U.S. AI labs have largely prioritized closed models, leaving a gap that Chinese labs like DeepSeek have filled with capable open alternatives. Reflection is betting that Western allies would prefer a U.S.-built open option.