Breaking

SpaceX Signs Up to $6.3B Compute Deal With Reflection AI

June 22, 2026 at 12:56 EDT

Infra & Chips
Open Source
Funding & M&A

SpaceX signed a deal on June 22, 2026 to supply compute to open-source AI lab Reflection AI, according to reports. Reflection will gain immediate access to Nvidia GB300 chips at SpaceX's Colossus 2 data center and pay $150 million per month from July 1, 2026 through the end of 2029, amounting to roughly $6.3 billion at full term.

June 22 · SpaceX × Reflection AI

A $6.3 Billion Compute Bet — Before the Model Ships

SpaceX will rent its most advanced Nvidia GB300 chips at "Colossus 2" in Memphis to open-source lab Reflection AI for $150M per month from July 2026 through 2029 — even though Reflection's flagship frontier model has yet to be released.

$150M

per month, starting July 1, 2026

~$6.3B

total contract value through 2029

$8B

Reflection AI valuation (Oct 2025)

flagship models shipped so far

SpaceX's "neocloud" tenants — monthly spend

Surplus Colossus capacity leased to AI labs. Reflection's deal is a fraction of Google's.

$920M

Google
~110,000 GPUs

$150M

Reflection AI
GB300, count n/a

n/a

Anthropic
300MW · 220k+ GPUs

Each block ≈ $150M/month · Anthropic's monthly figure undisclosed (holds all of Colossus 1)

The strategy

"American open intelligence" — a Western open-source standard-bearer against China's DeepSeek.

Nvidia-backed Open-weight MoE models Selected for Pentagon classified networks 90-day exit clause

▲ Seen as a tailwind

A boost for SpaceX's compute business and the U.S. open-source push — an option for governments and enterprises to avoid closed-model dependence.

▼ Seen as overheated

Committing $6.3B in compute before shipping a flagship model is read by some as a symbol of runaway AI infrastructure spending.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.