GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
← Back to feed

SpaceX Signs $6.3B Reflection AI Deal — Colossus 2 Becomes the Frontier's Shared Compute Rack

SpaceX has signed a third major AI compute lease. Reflection AI will pay $150 million per month starting July 1, 2026, through 2029 for immediate access to Nvidia GB300 chips and supporting hardware at Colossus 2 in Memphis, Tennessee. The deal is worth up to $6.3 billion over the full term. Either party can exit with 90 days’ notice after the first three months.

The deal brings SpaceX’s committed compute revenue to $2.32 billion per month: Anthropic at $1.25B (signed May 2026), Google at $920M (signed June 5), and Reflection at $150M beginning July.

The Tenant Roster

Colossus 2 now has three paying frontier AI labs where xAI used to be the only user. The irony is structural: xAI’s models have underperformed, but the hardware buildout was overbuilt — and overbuilt data centers become revenue when you rent them out. SpaceX’s position is now closer to a hyperscaler than to an AI lab. It owns the GB300 racks. The models are someone else’s problem.

Reflection AI is a different kind of tenant than Anthropic or Google. It was founded in 2024 by two former Google DeepMind researchers. Nvidia invested $800 million in the company. The mandate is open-weight models — publicly released parameters, no licensing restrictions. That positioning has gotten sharper since the U.S. government’s export ban on Anthropic’s Fable and Mythos models in June. Reflection cited the ban directly in its announcement, framing open source as the alternative for “nations and enterprises recognizing the risks and costs associated with exclusively depending on closed models.”

The timing is not accidental. Governments and enterprises that built workflows on Anthropic’s frontier models hit a wall when those models became inaccessible outside the U.S. Reflection is pitching itself as the solution. It now has compute capacity to back the pitch.

The Compute Math

TenantMonthlyThrough
Anthropic$1,250MJuly 2029
Google$920MJuly 2029
Reflection AI$150MJuly 2029
Total$2,320M—

At current rates, SpaceX collects roughly $27.8 billion in committed AI compute revenue through July 2029 across these three deals alone, before any capacity expansion or new customers. The Colossus campus was originally capitalized as xAI infrastructure and later rebranded under SpaceX’s xAI division after the merger.

Open-Weight Context

Reflection’s $6.3 billion deal is the largest open-source AI infrastructure commitment announced to date, per the company’s own statement. No model has shipped under the Reflection brand yet. The compute deal is the public commitment that the lab is training at scale and expects to compete at the frontier.

The Fable/Mythos ban has accelerated a structural shift. Enterprise buyers who cannot or will not depend on export-controlled closed models now have financial incentive to support open-weight alternatives. Reflection — with Nvidia on the cap table and SpaceX GB300s in the data center — is positioned to be that alternative by the time its models are ready.