Luxor's Mike San Miguel warns that dedicated AI clusters are a single point of failure. If your data center goes offline, your entire workload dies with it.
Hyperscalers like AWS/GCP have live migration tooling that can shift workloads across hardware during outages. With a dedicated cluster, you own hardware-level control but inherit infrastructure risk.
Real numbers: a $300M, 3-year contract for a 128-256 node cluster becomes a total loss if the DC fails. Worse, only Tier 3+ sites have direct peering to AWS/Google, so a non-Tier 3 outage can mean permanent data loss.
Supply constraints make this worse. Few sites exist, and provisioning the exact location you need takes 6-12 months. You're locked in with zero leverage if something breaks.
Hyperscalers like AWS/GCP have live migration tooling that can shift workloads across hardware during outages. With a dedicated cluster, you own hardware-level control but inherit infrastructure risk.
Real numbers: a $300M, 3-year contract for a 128-256 node cluster becomes a total loss if the DC fails. Worse, only Tier 3+ sites have direct peering to AWS/Google, so a non-Tier 3 outage can mean permanent data loss.
Supply constraints make this worse. Few sites exist, and provisioning the exact location you need takes 6-12 months. You're locked in with zero leverage if something breaks.