SpaceX signs computing deal with Reflection AI

According to CNBC, SpaceX has signed a multi-year agreement to provide the open-source AI startup Reflection with access to Nvidia GB300 GPUs and related hardware housed at the Colossus 2 data center in Memphis. CNBC reports Reflection agreed to pay $150 million per month starting July 1, 2026, through 2029, which would total about $6.3 billion over the contract term if it runs to completion. CNBC also reports the contract contains a 90-day termination clause exercisable by either party after the first three months.

Editorial analysis - technical context: Access to high-end inference and training accelerators such as the GB300 materially shortens wall-clock time for large-model experiments and production runs. Companies that cannot source immediate GB300-class capacity often face multi-quarter delays; leasing colocated capacity at established supercomputing sites is a common workaround in 2026. The reported month-to-month price point in this deal reflects both raw chip scarcity and the premium for colocated rack space, power, and networking that supports large-scale model training.

Public reporting by Quartz and Yahoo highlights that Nvidia is both an investor in Reflection (reported at roughly $800 million) and the supplier of the GB300 hardware the startup will run, creating a capital-compute loop frequently observed in the sector. Quartz also notes prior multi-year compute commitments tied to SpaceX, reporting roughly $30 billion for Google and $45 billion for Anthropic in earlier arrangements, which frames this Reflection deal as another data point in SpaceX's broader commercialization of Colossus capacity. Quartz reports Reflection has not yet released a public frontier open-source model at scale, and a Reflection spokesperson told Quartz, "Recent events highlight how important open source is to the AI ecosystem, with more nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models."

For practitioners: observers should track three indicators. First, whether Reflection publishes model checkpoints, training logs, or benchmarks that demonstrate utilization of the acquired GB300 capacity. Second, how pricing and contract structure at large colocated sites evolve as new entrants compete for scarce GB300-class inventory. Third, any regulatory, export-control, or procurement constraints that could affect access to large-scale Nvidia-equipped facilities for open-source labs. These signals will determine whether large compute contracts translate into faster model iteration or simply shift capacity scarcity elsewhere.

Editorial analysis: The deal, as reported, is a clear example of compute capacity being monetized as a strategic asset across the frontier-AI ecosystem. For teams building or evaluating large models, the transaction underscores that immediate access to next-generation accelerators increasingly depends on commercial arrangements with data-center operators rather than only ownership of on-prem hardware or standard cloud quotas.