Nvidia/Dell/Broadcom/Oracle/Google/Arm/Corning/SUSE/Marvell/MSFT/SK ramps & bottlenecks + China GPU boom + Roche factory + semicon testing + Terafab/Firmus/Nvidia SchedMD
Key Questions
What compute expansion has Anthropic announced with Google and Broadcom?
Anthropic has signed agreements with Google and Broadcom for 3.5GW of TPU capacity starting from 2027, building on 1GW planned for 2026. This supports their skyrocketing demand, with annual revenue run rate surpassing $30 billion.
What is Anthropic's current revenue run rate and customer growth?
Anthropic's run-rate revenue has exceeded $30 billion, up from $9 billion at the end of 2025, with over 1000 enterprise customers spending $1M each, doubling recently. They project $100 billion revenue by 2027.
What funding did Firmus secure and what are they building?
Firmus raised $505 million at a $5.5 billion valuation, backed by Nvidia and Coatue, to expand 1.6GW AI factories in APAC. This funds AI data center infrastructure amid the boom.
What does Nvidia's acquisition of SchedMD mean for supercomputers?
Nvidia acquired SchedMD, which controls Slurm used in 60% of supercomputers, giving Nvidia influence over open-source AI scheduling roadmaps. This raises questions for mixed-vendor GPU clusters.
How is Uber utilizing AWS chips for AI?
Uber is adopting AWS Trainium3 chips to accelerate AI model training and computing tasks. This is part of broader adoption of Amazon's custom AI chips.
What AI factory has Roche deployed with Nvidia?
Roche and Nvidia have launched the pharmaceutical industry's largest AI factory using Blackwell GPUs. It supports advanced AI infrastructure for drug discovery and research.
What challenges persist in AI infrastructure like power and HBM?
Power and HBM crunches continue to bottleneck AI factories, complicating scaling despite ramps by Nvidia, Dell, Broadcom, and others. Semiconductor testing for AI accelerators adds further complexity.
What is Intel's involvement in the Terafab project?
Intel has joined Elon Musk's Terafab AI chip project with SpaceX and Tesla to power humanoid robots and data centers. This aims to build new U.S. semiconductor capacity.
Anthropic expands to 3.5GW TPU/Broadcom/Google from 2027 atop 1GW '26, confirming $30B ARR and doubled 1000+ $1M enterprise custs toward $100B '27 rev; Firmus $505M $5.5B val Nvidia/Coatue 1.6GW APAC factories; Nvidia SchedMD acq controls 60% supercomputers; Uber AWS Trainium3; Roche Blackwell; power/HBM crunches persist.