Elon Musk’s SpaceX has found a new tenant for its Memphis data center. Reflection AI, an open-source startup, will pay $150 million a month for access to Nvidia’s latest GB300 chips. The contract, which begins July 1, 2026, is valued at $6.3 billion over three years.

This is a massive bet on open-weight models. Reflection AI, founded by two former Google DeepMind researchers, is positioning itself as the primary alternative to closed-source labs like Anthropic. The startup needs raw power to compete. Now, it has it.

The Economics of Colossus 2

The Colossus 2 data center was originally built for xAI’s internal projects. Those efforts have stalled. SpaceX pivoted quickly. It is now renting out its massive stockpile of Nvidia hardware to the highest bidders.

This deal is smaller than previous agreements. Anthropic pays $1.25 billion monthly. Google pays $920 million. Reflection AI’s $150 million monthly commitment is modest by comparison. Still, it is one of the largest infrastructure investments ever made by an open-source AI lab.

SpaceX holds all the cards. The contract allows either party to cancel with 90 days’ notice after the first three months. Musk has publicly downplayed the three-year duration. He views the capacity as liquid. If a better offer arrives, the hardware moves.

Why Open Source Matters Now

The timing is not accidental. The U.S. government recently banned Anthropic’s closed models, Fable and Mythos, citing national security concerns. This created a vacuum. Enterprises and sovereign nations are suddenly wary of relying on black-box systems they cannot audit.

Reflection AI is capitalizing on this anxiety. By releasing its model parameters, the company offers a level of transparency that closed labs cannot match. They are betting that the future of AI is not proprietary, but collaborative.

“More nations and enterprises recognize the risks of depending on closed models,” a Reflection spokesperson said. They want control. Reflection gives it to them.

What This Means for Developers

For the developer community, this deal is a signal. It proves that open-source labs can secure the same tier-one hardware as the tech giants. Access to GB300 chips is the ultimate bottleneck in AI development. If Reflection can scale, the gap between closed and open models will shrink.

However, the volatility remains. Because the contract can be terminated with 90 days' notice, Reflection’s long-term compute runway is not guaranteed. They have the chips today. They might not have them in six months.

Key Takeaways

  • SpaceX is monetizing its idle data center capacity by renting Nvidia GB300 chips to third-party labs.
  • Reflection AI’s $6.3 billion deal is a major validation for the open-weight AI movement following recent government bans on closed models.
  • The contract includes a 90-day cancellation clause, meaning SpaceX retains the flexibility to reallocate hardware if market demand shifts.

The real test for Reflection AI begins in July. They must prove that their open models can match the performance of the frontier systems they aim to replace. If they fail to deliver, the compute will go to someone else. The hardware is ready. The pressure is on.