Reflection AI Signs $1 Billion Chip Deal With Nebius, Its Second Compute Buildout in Weeks

Reflection AI, a San Francisco startup building open-source AI models, has signed a computing deal worth more than $1 billion with Nebius Group, the Amsterdam-based cloud provider that split off from Russia's Yandex in 2024, according to TechCrunch and Techstrong.ai. The agreement gives Reflection access to Nvidia's GB300 chips through 2029.

The startup's second major infrastructure deal came in a matter of weeks. Reflection signed a multibillion-dollar arrangement with SpaceX last month reportedly worth roughly $150 million a month through 2029, also for access to Nvidia's latest processors, according to Techstrong.ai.

"The need for open models is clear, and this additional compute capacity will allow Reflection to continue to build and train frontier AI models at scale," said Ioannis Antonoglou, Reflection's co-founder and chief technology officer, in a statement reported by Techstrong.ai.

Who's Behind Reflection

Reflection AI was founded in 2024 by Antonoglou and Misha Laskin, both former Google DeepMind researchers. By late 2025, the company had raised $2 billion at an $8 billion valuation, with backers including Nvidia and Sequoia Capital, per TechCrunch and Crypto Briefing. TechCrunch also lists Lightspeed Venture Partners among Reflection's investors.

Techstrong.ai reported Reflection is now in talks to raise another $2.5 billion at a $25 billion valuation, citing The Wall Street Journal. That would be a significant jump from its current $8 billion valuation.

Why Open Models, Why Now

The deal lands amid a real shift in how AI developers think about dependency on closed, proprietary models from firms like OpenAI and Anthropic.

TechCrunch reported that the Trump administration pressured Anthropic and OpenAI last month to restrict their most powerful new models, a move that rattled developers who suddenly realized access to frontier AI could be pulled with little warning. Techstrong.ai echoed this, noting that "recent U.S. regulatory curbs on Anthropic's advanced models have exposed the geopolitical risks of relying on closed-door providers that can be restricted overnight."

Any business betting its product on a single company's model, subject to a government that can order restrictions with no advance notice, is exposed to a risk that has nothing to do with the model's quality. Open-weight models, which developers can download, modify and run on their own hardware, don't carry that same off-switch risk. That's the pitch driving demand for Reflection and competitors building open alternatives, including increasingly capable Chinese labs.

None of the source reports here identify exactly what the Trump administration restricted or why, so the scope and justification for that intervention remains an open question. But the market reaction, startups stockpiling their own compute and betting on open weights, is a real, measurable response to it.

Nebius's Bigger Picture

For Nebius, this is one contract among several. The company's contracted backlog has reportedly reached $50 billion, according to Crypto Briefing. Nebius signed a five-year infrastructure deal with Meta worth up to $27 billion, and last year inked a multiyear agreement with Microsoft worth up to $19.4 billion, per TechCrunch and Techstrong.ai.

Nvidia itself invested $2 billion in Nebius in March 2026, making the chipmaker both an investor in and a supplier to a company that then turns around and sells that same chip capacity to firms like Reflection. Crypto Briefing called this dual role effectively making Nebius "both a customer and a strategic partner" of Nvidia, a structure that's become common across the AI infrastructure sector but raises questions about how independent these valuations really are when the same players are investing in each other up and down the supply chain.

What's Unresolved

The exact terms of the Reflection-Nebius deal, including GPU quantities and payment structure, have not been disclosed by either company, according to Crypto Briefing. Neither Reflection nor Nebius has provided additional public detail beyond confirming the agreement, per TechCrunch's outreach to both companies.

Whether the broader AI infrastructure spending spree is sustainable is a separate and unresolved question. Other reporting referenced in the same news cycle noted Oracle's stock fell roughly 47% since June 1 amid investor worries about heavy AI capital expenditure, and Meta announced a $50 billion expansion of its Louisiana data center to 5 gigawatts while also planning to sell excess AI computing capacity it doesn't need. That contradiction, building more while also trying to offload capacity, suggests even the biggest players aren't certain how much compute the market actually requires versus how much they're stockpiling out of fear of being caught short.

Reflection's next move is the reported $2.5 billion funding round at a $25 billion valuation. If it closes, it will be a clear signal that investors believe the open-model bet is paying off. If it doesn't, the billions in locked-in compute contracts through 2029 will still be sitting on the books regardless.