$1B Nebius Agreement Secures GB300 Chips for Reflection AI

Reflection AI signed a multi-year computing agreement valued at over $1 billion with Nebius Group. The deal provides the AI startup with access to Nebius’s cloud infrastructure and Nvidia GB300 chips through 2029. Reflection plans to use the hardware capacity to train and deploy frontier-level open-source foundation models and AI agents.

This transaction brings Reflection’s total infrastructure commitments to nearly $7.3 billion. The company has aggressively secured compute capacity prior to releasing a flagship public model, opting to lock in long-term hardware access before demonstrating generation capabilities.

Hardware and Financial Impact

The agreement guarantees Reflection primary access to Nvidia’s latest GB300 silicon. Following the announcement, Nebius Group shares rose nearly 3% to approximately $216 in premarket trading.

Nebius, a European cloud provider headquartered in Amsterdam, emerged from the 2024 split of Yandex. The company has rapidly expanded its infrastructure backlog to a reported $50 billion. The Reflection agreement follows massive recent contracts with Meta and Microsoft, valued at $27 billion and $19.4 billion respectively. Nebius previously secured a $2 billion direct investment from Nvidia in March 2026 to accelerate its hardware deployment.

Reflection is currently in discussions to raise $2.5 billion at a $25 billion valuation. The lab previously raised $2 billion at an $8 billion valuation in late 2025. Investors in the earlier round included Nvidia, Sequoia Capital, Lightspeed Venture Partners, and Eric Schmidt.

Compute Acquisition Strategy

The Nebius agreement represents the second major hardware deal for Reflection this summer, cementing a capital-intensive strategy to hoard compute resources.

Infrastructure Partner	Deal Value	Hardware Focus
Nebius Group	>$1.0 Billion	Nvidia GB300
SpaceXAI	$6.3 Billion	Colossus 2 Data Center

In June 2026, the startup signed the $6.3 billion agreement with SpaceXAI. That arrangement, which reportedly began on July 1, costs Reflection roughly $150 million per month for access to the Colossus 2 facility in Memphis, Tennessee. Beyond commercial infrastructure, Reflection also partnered with the U.S. Department of Energy’s Genesis Mission in June 2026 to integrate with federal AI science programs.

Reflection was founded in March 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. Laskin previously worked on reward modeling for Gemini, while Antonoglou co-created AlphaGo. The startup positions itself as a U.S.-based alternative to closed-source AI providers.

Analysts view this spending as a defensive maneuver to secure compute hours during ongoing global hardware constraints. The hardware acquisition positions Reflection to compete directly with Chinese open-source releases that currently lead the open-weights market.

If you are building products dependent on open-source foundation models, Reflection’s infrastructure lock-in ensures they have the hardware necessary to train highly parameterized architectures. You should monitor their upcoming open-weight releases as a potential U.S.-based alternative for self-hosted deployments.