VFF - The signal in the noise
VFF - The signal in the noise
NewsTrending

SpaceX, Reflection AI ink $150M monthly compute deal through 2029

TechCrunch AI
Read original
Share
SpaceX, Reflection AI ink $150M monthly compute deal through 2029

Reflection AI, an open source AI lab, has signed a three-year compute agreement with SpaceX worth $150 million per month starting July 1, 2026. The deal grants Reflection AI immediate access to Nvidia's latest GB300 AI chips and supporting hardware at SpaceX's Colossus 2 data center near Memphis, Tennessee through 2029. The arrangement represents a significant infrastructure commitment for an open source AI research organization.

TL;DR

  • Reflection AI commits to $150 million monthly payments for compute access through 2029
  • Deal provides access to Nvidia GB300 chips at SpaceX's Colossus 2 data center in Memphis
  • Agreement begins July 1, 2026 and runs through 2029
  • Reflects growing demand for high-end AI compute resources among research organizations

Why It Matters

This deal signals that open source AI labs now have access to capital and resources comparable to commercial AI companies. The scale of the commitment, $150 million monthly, demonstrates the computational intensity required for frontier AI research and the consolidation of compute resources among a small number of providers like SpaceX.

Business Impact

For SpaceX, the deal represents a new revenue stream from its data center infrastructure beyond traditional aerospace operations. For Reflection AI, securing dedicated access to latest-generation chips at scale is critical for competitive AI research, but the monthly cost structure creates significant ongoing financial obligations.

Key Implications

  • Open source AI research organizations now require enterprise-scale compute budgets and multi-year financial commitments
  • SpaceX's Colossus 2 facility is positioned as a major compute provider competing with established cloud infrastructure players
  • Nvidia's GB300 chips are in high demand, with allocation going to both commercial and research-focused organizations

What to Watch

Monitor whether other open source AI labs pursue similar long-term compute deals and at what price points. Track Reflection AI's research output and whether the dedicated compute access translates to competitive model releases. Watch for additional announcements about SpaceX's data center utilization and whether it becomes a significant business line.

Share
AI HardwareInfrastructureOpen SourceFunding & Startups

Subscribe to the newsletter

The latest stories and analysis, delivered to your inbox.

Free. No spam. Unsubscribe any time.

Related stories

Context, Not Compute, Is Becoming The Bottleneck In AI Inference
AI HardwareNews

Context, Not Compute, Is Becoming The Bottleneck In AI Inference

As AI inference workloads shift from discrete queries to persistent, multi-step agentic systems, the bottleneck has moved from GPU compute to context management. Context volumes are growing faster than GPU efficiency improvements due to expanding context windows, chained model calls in agentic systems, and enterprise requirements for persistent inference state across sessions. A new dedicated storage tier, optimized for key-value cache and retrieval data, is emerging between GPU memory and bulk storage to address this gap.

· VentureBeat AI
Los Alamos Deploys NVIDIA Vera CPUs for Agentic AI Science
AI HardwareNews

Los Alamos Deploys NVIDIA Vera CPUs for Agentic AI Science

Los Alamos National Laboratory is deploying three new supercomputers, Mission, Vision, and Veritas, built with HPE and NVIDIA hardware including the NVIDIA Vera CPU to accelerate scientific discovery and agentic AI research. Early testing shows the Vera CPU delivers 7x higher performance on URSA (Universal Research and Scientific Agent) workloads and over 3x performance on Monte Carlo simulations compared to the previous Crossroads x86 supercomputer. The systems, expected operational in 2027, will support classified national security work, fundamental science research, and testing of AI agents that can autonomously form hypotheses, run simulations, and refine experiments.

by Chris Porter· NVIDIA Blog (AI)
NVIDIA Accelerates Scientific Computing with Real-Time AI Tools
AI HardwareNews

NVIDIA Accelerates Scientific Computing with Real-Time AI Tools

NVIDIA introduced new AI software tools at ISC Hamburg designed to accelerate scientific research across chemistry, materials discovery, and astronomy. The tools, including DAQIRI, ALCHEMI NIM microservices, and cuPhoton reference code, deliver GPU-accelerated pipelines that reduce processing times from hours or days to real-time. Early results show cuPhoton achieved 14,900x speedup in loading FITS astronomical data and 8,400x faster signal processing on NVIDIA GB200 NVL72 systems.

by Chris Porter· NVIDIA Blog (AI)
JUPITER Shows Exascale Computing's Real-World Impact
AI HardwareTrendingNews

JUPITER Shows Exascale Computing's Real-World Impact

JUPITER, Europe's first exascale supercomputer at Germany's Forschungszentrum Jülich, is running four major science projects that demonstrate the practical capabilities of exascale computing. These projects span brain mapping at cellular resolution, global climate simulation at 1-kilometer resolution, AI for wireless networks, and quantum computing simulation. The work shows that problems previously intractable are now solvable with exascale hardware and software.

by Chris Porter· NVIDIA Blog (AI)