The strongest new infrastructure signal is that AI compute is becoming a financeable, productized capacity layer outside the classic hyperscaler boundary. Blackstone and Google announced a dedicated TPU cloud JV with an initial $5B equity commitment and 500MW targeted online in 2027, while CoreWeave closed a $3.1B publicly syndicated HPC-backed loan facility tied to committed customer deployments. In parallel, Dell is packaging rack-scale AI infrastructure...
5GWAWS/Anthropic capacity commitment ceiling
$100B+Anthropic ten-year AWS technology commitment
$27BNebius/Meta five-year capacity agreement ceiling
1.2GWCrusoe Abilene target total power capacity
Enterprise / EMEA / Lenovo Angle
Enterprise buyers will want validated stacks with control. Dell/Mistral and Dell/Palantir-style packaging tells customers: keep data, models, workflows, and IP under local control while scaling from deskside to data center...
What To Watch Next
- Whether Google/Blackstone names anchor customers, pricing model, TPU generation, regions...
- Whether CoreWeave's GPU/HPC financing spreads tighten further and whether lenders start demanding more transparency on customer...
- Whether Dell publishes detailed PowerRack availability, validated configurations, power/cooling envelopes...
- Claims that TPUs "beat GPUs" or GPUs "beat TPUs" without workload, software, supply, utilization, and lock-in context.