SpaceX inks $150M monthly compute deal with Reflection AI

Read original on TechCrunch AI

#gpu-compute#data-centercolossus-2-data-centerspacexreflection ainvidiagb300colossus 2

Major $5.4B compute deal reveals the future of AI infrastructure scaling and Nvidia's next-gen hardware deployment.

30-Second TL;DR

What Changed

Reflection AI secures massive compute capacity at SpaceX's Colossus 2 facility.

Why It Matters

This deal underscores the extreme demand for high-end GPU clusters and highlights SpaceX's pivot into a major AI infrastructure provider. It signals a shift where non-traditional tech giants are becoming critical compute hubs.

What To Do Next

Monitor the availability of Nvidia GB300-based cloud instances if you are planning large-scale training runs for 2026.

Who should care:Founders & Product Leaders

Key Points

•Reflection AI secures massive compute capacity at SpaceX's Colossus 2 facility.
•The contract spans from July 2026 through 2029 with a $150 million monthly fee.
•The deal provides immediate access to Nvidia's next-generation GB300 AI chips.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

•The Colossus 2 facility, located in Memphis, Tennessee, represents a strategic expansion of Elon Musk's xAI infrastructure, now being leveraged for third-party commercial compute leasing.
•Reflection AI is reportedly utilizing this compute cluster to train a new class of 'reasoning-first' open-source models designed to compete with proprietary frontier models.
•The GB300 chip architecture utilizes advanced HBM4 memory stacks, providing a significant bandwidth increase over the previous generation Blackwell B200 series.
•This deal marks a shift in SpaceX's business model, diversifying revenue streams by monetizing idle or excess data center capacity during non-peak training cycles.
•Industry analysts suggest the $150 million monthly valuation reflects a premium pricing model due to the immediate, large-scale availability of GB300 hardware in a single contiguous cluster.

Competitor Analysis

AWS (Trainium/Inferentia)

Compute Offering: Custom Silicon
Pricing Model: Pay-as-you-go
Key Advantage: Deep integration with AWS ecosystem

CoreWeave

Compute Offering: H100/B200 Clusters
Pricing Model: Reserved/Spot
Key Advantage: Specialized GPU cloud infrastructure

Microsoft Azure

Compute Offering: ND-series (Nvidia)
Pricing Model: Enterprise Agreement
Key Advantage: Massive scale and global availability

Reflection AI/SpaceX

Compute Offering: GB300 Cluster
Pricing Model: Fixed Monthly Lease
Key Advantage: Unprecedented single-cluster density

Provider	Compute Offering	Pricing Model	Key Advantage
AWS (Trainium/Inferentia)	Custom Silicon	Pay-as-you-go	Deep integration with AWS ecosystem
CoreWeave	H100/B200 Clusters	Reserved/Spot	Specialized GPU cloud infrastructure
Microsoft Azure	ND-series (Nvidia)	Enterprise Agreement	Massive scale and global availability
Reflection AI/SpaceX	GB300 Cluster	Fixed Monthly Lease	Unprecedented single-cluster density

Technical Deep Dive

The GB300 architecture integrates the Grace CPU and Blackwell GPU on a single board with a high-speed NVLink interconnect.
The Colossus 2 facility utilizes a liquid-cooling infrastructure capable of supporting rack densities exceeding 100kW to accommodate the thermal output of GB300 chips.
The cluster employs a custom InfiniBand fabric topology designed to minimize latency for distributed training across tens of thousands of GPUs.
Reflection AI's implementation focuses on FP8 and FP4 precision training workflows to maximize throughput on the Blackwell architecture.

Future ImplicationsAI analysis grounded in cited sources

SpaceX will launch a dedicated cloud compute subsidiary by Q4 2026.

The scale of the Reflection AI deal suggests that SpaceX is transitioning from internal infrastructure management to becoming a commercial provider of high-performance computing.

Reflection AI will release a model surpassing 2 trillion parameters by early 2027.

Access to the massive GB300 cluster at Colossus 2 provides the necessary compute throughput to train models at this scale within the contract's timeframe.

Timeline

2024-09

xAI brings the original Colossus cluster online in Memphis.

2025-03

SpaceX begins infrastructure upgrades to support next-generation GPU clusters.

2026-02

Colossus 2 facility reaches initial operational capacity.

2026-05

SpaceX and Reflection AI finalize terms for the compute partnership.

2024-09
xAI brings the original Colossus cluster online in Memphis.
2025-03
SpaceX begins infrastructure upgrades to support next-generation GPU clusters.
2026-02
Colossus 2 facility reaches initial operational capacity.
2026-05
SpaceX and Reflection AI finalize terms for the compute partnership.