AI startup Reflection has signed a deal worth more than $1 billion with cloud provider Nebius to secure access to Nvidia's latest chips and data-center capacity. The agreement, announced Tuesday, highlights how companies racing to build and deploy artificial intelligence are increasingly locking in computing resources years in advance rather than buying them on the open market.
Reflection was founded by two former researchers from Google DeepMind, the Alphabet-owned AI lab behind breakthroughs like AlphaGo. The startup is developing open-source AI models, positioning itself as an alternative to the closed systems of OpenAI and Anthropic. Open-source models are generally easier for businesses to customize and can be cheaper to run, especially as companies look to control their AI spending.
Why compute is suddenly a long-term bet
For AI startups, access to high-end chips and data-center space is a make-or-break factor. Without enough computing power—often called "compute"—training and running advanced AI models slows to a crawl. Demand from businesses adopting AI is rising faster than new data-center supply can keep up, according to Reuters, which is why Reflection is locking in capacity now instead of hoping it will be available later.
This Nebius agreement follows Reflection's earlier deal with SpaceX, which media reports said could cost about $150 million a month through 2029. That pattern shows how quickly compute is starting to look like long-term infrastructure rather than a pay-as-you-go utility. After recent US restrictions limited access to some advanced AI models, reliable and controllable access to both models and the chips behind them is becoming a competitive advantage.
The broader trend is visible across the industry. Data-center operators are racing to expand capacity, with companies like Pure Data Centres planning a €1.5 billion campus in Finland and Switch eyeing a $10 billion IPO. Meanwhile, Chinese AI startup DeepSeek is targeting a $71 billion valuation as it develops its own chip plans, showing the global scope of the compute race.
What it means for investors
When an AI company signs a multi-year compute commitment, it is effectively turning what could be "spot" GPU demand—buying whatever is available today—into a contract with predictable payments. That shift matters most for providers like Nebius, a cloud and data-center firm. Contracted revenue is easier to borrow against, which helps fund new data centers and the Nvidia hardware that fills them.
The more customers reserve capacity up front, the less these operators depend on short-term utilization swings. The AI supply chain is starting to resemble an infrastructure business with visible backlog rather than a purely cyclical tech market. For investors, that means companies that own data centers and cloud capacity may see more stable revenue streams, while chipmakers like Nvidia benefit from long-term demand visibility.
However, the flip side is that startups like Reflection are taking on significant financial commitments. If AI adoption slows or the company's open-source bet doesn't pay off, those contracts could become a burden. For now, though, the race for compute shows no signs of easing, and Reflection's deal is the latest sign that the AI boom is reshaping how companies think about hardware—not as a one-time purchase, but as a long-term infrastructure investment.
