CoreWeave's top-3 customers now 72% of revenue, down from Microsoft alone at 71% a year ago

AccBalanced · x · 2026-08-17

Analysis arguing neocloud customer concentration signals the capex cycle is still early:

• A year ago Microsoft was 71% of CoreWeave revenue; this quarter the top 3 customers were 36%, 26%, and 10%.
• Demand is broadening beyond frontier labs into software, industry (Caterpillar on Vera Rubin for physical AI), life sciences (Isomorphic Labs), financial services (Flow Traders, IMC), and public sector (Leidos).
• The Caterpillar deal is an underappreciated note from the earnings call: physical AI was only 1% of Q1 revenue backlog and wasn't disclosed this quarter, but with the world's largest construction company signed, growth is clearly coming.
• Current scale: 0.47GW.

Original post →

More from Infra

• Command example to run Qwen3.8-27B GGUF on DGX Spark — ariG23498 · 2026-08-17
• llama-server crashes with 'non-consecutive token position' on AMD/Vulkan — Gold-Drag9242 · 2026-08-17
• How to start running Qwen 3.8 locally with a 3090 GPU? — ozymandizz · 2026-08-17
• Qwen MLX Challenge Launches to Benchmark Local Model Inference Speed — corruptbytes · 2026-08-17
• MLX framework remains active as community shifts to custom inference engines — AccBalanced · 2026-08-17
• ByteDance & Tsinghua Release CUDA-Agent: RL-Based High-Performance Kernel Generation — AIFlow_ML · 2026-08-17

Infra channel →