On July 14, 2026, TechCrunch reported that Reflection AI, a startup building open-weight models, signed a $1 billion compute deal with European AI infrastructure company Nebius, gaining access to Nvidia’s latest chips. It is Reflection’s second mega-scale compute transaction in three weeks — in late June it signed a contract with SpaceX worth $150 million per month, up to $6.3 billion in total.
For a company founded in 2024, that pace explains the rules of the game in 2026: models can be open, compute cannot. To train frontier-grade open-weight models, you first need a place in the supply chain.
What a Billion Dollars Buys
The core of the contract is access to Nvidia’s latest chips, delivered through Nebius’s AI cloud; the specific chip models and counts were not disclosed. But Nebius’s existing contract structure suggests the scale: in March it signed a five-year agreement with Meta worth up to $27 billion, committing to deliver one of the first large-scale deployments of the Nvidia Vera Rubin platform starting in early 2027.
In other words, Reflection is not buying “servers.” It is buying a multi-year guarantee of training and inference capacity. In a market where GPU lead times are measured in years, that kind of long-dated contract is itself a moat.
Reflection: Second Mega-Deal in Three Weeks
Reflection was founded in 2024 by two former Google DeepMind researchers, positions itself as an open-weight lab, and has raised nearly $2.6 billion from investors including Nvidia, Sequoia, and Lightspeed, at a last-known pre-money valuation of $25 billion.
Its compute posture now stands on two legs. The SpaceX deal announced on June 22 rents Nvidia GB300 chips at the Colossus 2 facility near Memphis, Tennessee, at $150 million per month through 2029, with either party able to exit on 90 days’ notice after the first three months; Reflection called it one of the largest announced open AI infrastructure commitments to date. Add Nebius’s $1 billion on top. For scale: SpaceX’s contracts on the same compute resources run $1.25 billion per month with Anthropic and $920 million per month with Google. Reflection’s commitments are far smaller, but for a startup they are already a heavy bet — and the company’s own framing is that “more compute means more runway to build the world’s best open models at scale.”
Nebius: The New Giant in GPU Cloud
Nebius was spun out of Yandex and is now a publicly traded AI infrastructure company. Its customer list has swollen within a year: a five-year agreement with Meta worth up to $27 billion, a multi-year Microsoft deal signed last year worth up to $19.4 billion, and a $2 billion investment from Nvidia announced in March, with a shared goal of deploying more than 5 GW of Nvidia systems by the end of 2030. Even after the Meta agreement, Nebius said its 2026 guidance remained unchanged — a signal of how much headroom it believes it has. The Nvidia partnership spans the full stack, from AI factory design to fleet management, with early access to platforms like Nvidia Rubin, Vera CPUs, and BlueField storage.
The model is worth noticing. Nebius does not sell chips and does not build closed models; it plays the “neutral GPU cloud,” in CEO Arkady Volozh’s framing one of the first large-scale clouds purpose-built for AI developers. When frontier labs compete with one another and hyperscale clouds each have their own allegiances, an independent infrastructure provider becomes the supplier everyone can live with.
Signals for the Compute Market
Three observations. First, mega compute contracts are becoming standard issue — billion-dollar scale, multi-year terms, specified chip generations — and the buyer profile now extends from giants to valuation-backed startups. Second, the competitiveness of the open-weight route depends directly on compute: Reflection is converting raised capital into GPU hours, betting that the quality of open models is ultimately decided by scale. Third, the supply side keeps diversifying — SpaceX rents out its own facilities, Nebius expands the neutral cloud, hyperscalers build more of their own — and buyers, for the first time, genuinely have options. The reminder for infrastructure teams is simple: these contracts lock in capacity for 2027 through 2029. Today’s shortage was decided two years ago.
Sources
- Reflection inks $1B compute deal with Nebius — TechCrunch
- SpaceX inks compute deal with Reflection AI — TechCrunch
- NVIDIA and Nebius partner to scale full-stack AI cloud — Nebius
- Nebius signs new AI infrastructure agreement with Meta — Nebius
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.