Reflection AI secures $1B compute deal with Nebius

Reflection AI, a U.S. startup founded in 2024 by two former Google DeepMind researchers, has signed a $1 billion（約1600億円） compute deal with Nebius, a European AI infrastructure company. The agreement will provide Reflection access to Nvidia's latest chips for training and deploying its open-weight AI models.

The partnership comes just weeks after Reflection secured a separate computing agreement with SpaceX, demonstrating the startup's aggressive strategy to lock in compute resources at a time when that capacity is fiercely contested. Nebius, formerly the international arm of Russian tech giant Yandex, has positioned itself as a major independent compute provider. The company recently signed a five-year infrastructure deal with Meta worth up to $27 billion（約4.3兆円） and a multi-year agreement with Microsoft worth up to $19.4 billion（約3.1兆円）.

Reflection's pursuit of independent compute infrastructure reflects broader industry dynamics. Last month, the Trump administration pressured Anthropic and OpenAI to restrict their most powerful new models, raising concerns among developers and enterprises about potential overnight loss of access to advanced AI systems. That regulatory uncertainty, paired with growing data retention concerns, has catalyzed mainstream interest in open-source alternatives. China has also released increasingly capable open models, adding competitive pressure.

Reflection is well-capitalized for this battle. Currently valued at $8 billion（約1.3兆円）, the startup has raised close to $2.6 billion（約4200億円） from major backers including Nvidia, Sequoia Capital, and Lightspeed Venture Partners. The Nvidia investment is particularly significant—Reflection secured a $2 billion（約3200億円） commitment from the chip giant, establishing financial and strategic alignment that supports both the compute deal and broader competitive positioning.

Reflection AI's $1 billion（約1600億円） compute deal with Nebius signals a strategic pivot across the AI industry toward securing independent computing infrastructure. The timing is deliberate: just last month, the Trump administration pressured Anthropic and OpenAI to restrict their most powerful models, creating uncertainty about whether access to advanced AI systems could be constrained by government action. That regulatory risk, combined with rising data retention concerns, has made open-weight models—which distribute code and weights publicly—more attractive to enterprises and developers who fear losing access to proprietary systems.

Reflection's dual strategy of signing compute partnerships with both Nebius and SpaceX in rapid succession reflects the broader competitive landscape. Open-weight model developers now face competition not only from each other but from increasingly capable Chinese counterparts. The $1 billion（約1600億円） commitment from Nebius also underscores a structural shift: compute capacity itself has become a bottleneck and a differentiator. Established AI labs that once dominated through proprietary data and scale must now compete with well-funded open-source alternatives that can credibly argue for transparency, sovereignty, and resilience to policy shifts.

No comments yet. Be the first to share your thoughts!