A start-up, backed by Nvidia, Reflection is leading a push to create freely available US AI systems and is in talks to raise $2.5bn at a $25bn valuation.
Reflection is one of a handful of Nvidia-linked start-ups chasing a network of “open source” AI models that businesses, labs and universities can use and repurpose to suit their needs.
The deal would give Reflection more muscle and keep it central to Nvidia’s attempt to build an open-source ecosystem that runs on its chips and counters China’s increasingly slick offerings.
JPMorgan Chase is in talks to join the round through its newly formed Security and Resiliency Initiative, created in December to back US companies in sectors critical to the economy and national security, the people said.
The bank said at the time that it planned to invest up to $10bn into venture-backed start-ups through the initiative.
Disruptive, an earlier Reflection investor, is also expected to invest.
Reflection has built cosy ties with Nvidia, which invested about $800m in a previous round that valued the young start-up, which has yet to generate meaningful revenue, at $8bn, the Journal has reported.
The $25bn valuation is the figure before the $2.5bn lands. The round sits amid a flurry of cash for neolabs that prioritise long-term research and new AI models over short-term profits.
Nvidia has been introducing Reflection to potential customers, including foreign governments that want homegrown AI infrastructure and control over how the tech gets used.
Reflection recently committed several billion dollars to build models customised for Korean in a deal with South Korean conglomerate Shinsegae Group. Thousands of Nvidia chips will power the data centre behind the project.
Reflection plans more deals like that, aiming to become the open model used by sovereign clouds in US allies around the world. AI boosters have started calling these systems “sovereign AI.”
Investors have been pitching Reflection as the “DeepSeek of the West”, a supposed alternative to open-source models from Chinese companies.
US AI labs have not prioritised open-source models, stoking fears that other countries will end up running Chinese technology instead.
Reflection chief executive Misha Laskin said: “Open models are Trojan horses for the infrastructure they bring with them,” in an interview earlier this month.
For those not in the know. Laskin is a Russian-Israeli researcher who previously worked at Google’s DeepMind.
