Why it matters
The strategic read
Adds a third major external customer to the SpaceXAI compute-capacity stack after Anthropic and Google. Reflection is an open-source frontier-model contender backed by NVIDIA, and the reported deal shows even model labs with strategic chip backing are renting dedicated GB300-era capacity from hyperscale AI infrastructure operators rather than waiting to build their own data centers. It also turns Colossus 2 into a multi-tenant AI factory platform with recurring revenue measured in billions.
Source-supported terms
What was disclosed
- 01
Axios, mirrored by Yahoo Finance, reported on June 22, 2026 that NVIDIA-backed open-source AI startup Reflection signed a major compute agreement with SpaceXAI.
- 02
The report says Reflection gets immediate access to chips and hardware from the SpaceX Colossus 2 data center.
- 03
After an initial ramp period, Reflection will pay SpaceXAI $150M per month starting July 1, 2026 through 2029.
- 04
Axios said the deal gives Reflection access to high-end reasoning GB300 chips and other hardware inside Colossus 2.
- 05
Either company can end the agreement with 90 days notice after the first three months.
- 06
Bloomberg, syndicated by Moneycontrol, separately reported that Reflection will tap hardware at SpaceX's Colossus 2 data center in Memphis, Tennessee.
- 07
The reported arrangement follows other large SpaceXAI compute agreements with Anthropic and Google that reserve capacity from the Colossus platform.
- 08
A Reflection spokesperson told Axios that recent events highlight the importance of open source to the AI ecosystem as more nations and enterprises weigh the risks and costs of depending exclusively on closed models.
Disclosure boundary
What is known—and what is not
Market reporting from Axios / Yahoo Finance and Bloomberg / Moneycontrol. Public details disclose the customer, infrastructure provider, Colossus 2 facility framing, $150M monthly payment level, July 1 full-rate start date after ramp, through-2029 term, GB300 / hardware access and 90-day termination right after the first three months. Public materials do not disclose the full agreement, exact maximum contract value, GPU count, reserved megawatts, SLA terms, delivery milestones, networking topology, power or cooling obligations, data-security terms, minimum utilization, remedies, payment step-downs, exclusivity, customer workload allocation or whether SpaceXAI has separately announced the deal.