Notes

Notes on reserved
GPU capacity.

Long-form notes for infrastructure and finance teams on reserving GPU capacity: contract terms, depreciation, delivery and operations.

  1. Economics13 min readBuy vs rent a GPU cluster: 5-year TCOA decision guide for a five-year GPU cluster: owning in colocation, a dedicated multi-year reservation, or short-term rental, compared per productive GPU-hour, with the conditions under which owning comes out ahead.
  2. Economics14 min readGPU lease agreement: a checklist before signingA clause-by-clause checklist for multi-year GPU lease, reservation and offtake agreements: term and rate scope, delivery and acceptance, SLA and node replacement, prepayment protection, exit, refresh, placement, data, expansion and provider failure, with worked numbers and a table to negotiate from.
  3. Deployment12 min readGPU cluster SLA: uptime vs training goodputWhat uptime means for a GPU cluster and why goodput matters more: published failure rates from large training runs, what fails inside a node, interruptions by cluster size, checkpoint and spare-node math, and how to write an SLA for training and for inference.
  4. Economics8 min readGB300 vs Vera Rubin: buy now or wait?For teams signing a multi-year GPU commitment in 2026 or 2027: what Vera Rubin gains over GB300 on public numbers, what waiting costs, and what happens to value when the next generation lands mid-term.
  5. Economics9 min readGPU depreciation and useful life in 2026The 2026 depreciation schedules for GPU servers at the hyperscalers and two listed GPU clouds, how they changed since 2020, the Burry critique and the evidence against it, and what sets a GPU's economic life.
  6. Deployment10 min readHow long to build an AI data center in 2026How long it takes to stand up an AI cluster yourself in 2026: two to four years for a greenfield data center, roughly 6 to 18 months in leased colo space, and what fills each month.
Get started

Start with one rack.

Tell us the size and timing you have in mind. We usually reply within a few hours and come back with a written proposal.

sales@lightyard.ai