Reflection AI locks in $1B+ Nebius compute pact through 2029
TL;DR
- Reflection AI signed a more than $1 billion multi-year compute deal with Nebius running through 2029, with access to Nvidia GB300 chips.
- The pact stacks on a June SpaceX agreement reportedly worth about $150 million a month, roughly $6.3 billion over its life through 2029.
- Reflection, launched in 2024 by two former Google DeepMind researchers, is valued at $8 billion on about $2.6 billion raised from Nvidia, Sequoia, and Lightspeed.
Reflection AI, the year-old open-weights lab, has committed to more than a billion dollars of computing capacity from European cloud provider Nebius through 2029, with access to Nvidia's GB300 chips as part of the package. TechCrunch reported the deal on July 14, and it sits on top of a June arrangement with SpaceX that CNBC pegged at about $150 million a month starting July 1, 2026, running through 2029 for a total of roughly $6.3 billion.
For a startup founded in 2024 by two former Google DeepMind researchers and valued at $8 billion on around $2.6 billion raised from Nvidia, Sequoia Capital, and Lightspeed Venture Partners, that is a lot of chip-hours to underwrite. The pitch is that Reflection is one of the few US labs seriously trying to compete on open-weights models, positioned as an alternative to the closed offerings from OpenAI and Anthropic and, implicitly, to the Chinese open-source releases that have been setting the pace on that side of the market.
The Nebius side of the story is quieter but arguably the more interesting corporate arc. Nebius is the entity that spun out of the old Yandex, and it has been aggressively signing infrastructure contracts, including a reported $27 billion five-year deal with Meta, a $19.4 billion multi-year deal with Microsoft, and a $2 billion investment from Nvidia. The Reflection agreement nudged Nebius shares up nearly 3% in premarket trading, according to TipRanks, small compared with those hyperscaler contracts but consistent with the trajectory.
The honest caveat is that these are commitments to spend, not proof of demand for the resulting models. Reflection has not shipped a flagship open-weights release at the scale the funding implies, and both the Nebius and SpaceX terms are single-sourced in press reports rather than filed in detail. What the reporting does not give you is the delivery schedule for the GB300s, how much of the $1 billion is fixed versus contingent, or how the two contracts sequence together.
The signal worth watching is whether a US open-weights lab with this much compute can actually put a competitive checkpoint out inside a year. That is the bet Nvidia, Sequoia, and Lightspeed are underwriting, and it is the reason Nebius's stock ticked up on the news rather than shrugging.
Originally reported by techcrunch.com
Read the original article →Original headline: Reflection AI Signs $1B+ Multi-Year Compute Deal With Nebius for Nvidia GB300 Access — Follows June SpaceX ~$150M/Month Contract, Cements Ex-DeepMind Startup's $8B Valuation as Leading US Open-Weights Challenger to Chinese Models