Quiz: AI Data Center Project Delivery & Procurement (Part 2)

10 questions from Part 2 of the guide. Score instantly; every explanation links to the chapter that derives the answer.

The single milestone an AI build program is managed toward is…
certificate of occupancytime-to-first-train (or first-token)MEP substantial completionutility energization only
Which track sets the program's critical path?
the building (12–18 months shell-to-MEP)the power track — the ~128-week transformer and the interconnection, not the building or the GPUsGPU deliverynetwork cabling
Long-lead procurement in AI builds means ordering critical equipment against…
a 100% complete designa design that is still ~60% complete — the register accepts change risk to buy schedulevendor stock, off the shelfthe previous project's spec
The fundamental contract-stack fork is…
fixed price vs cost plus, onlyone integrated EPC wrap carrying schedule risk vs multi-prime with the owner holding the interfacesunion vs non-uniondomestic vs imported equipment
Why carve the project into a bankruptcy-remote SPV?
to avoid building codesto enable non-recourse project financing — lenders underwrite the asset's contracts, not the parent's balance sheetto skip interconnection queuesSPVs get tax-free power
Owner-furnished equipment (OFE) is the default for GPUs and transformers because…
contractors refuse to buy themallocation and lead times make them strategic supply, not commodity procurement a GC can be trusted to scheduleit avoids sales taxwarranties require it
What does the guide say about simulation before steel is cut?
it's optional polishthermal/CFD, power-system/EMT, and network simulation to design-validating fidelity is now standard for gigawatt programsonly cooling gets simulatedsimulation replaces commissioning
Delay-in-startup (DSU) insurance matters because…
it's legally mandatorylenders require the layer, and the constrained 2026 market (~$10B premium pool) can be the financing bottleneckit covers GPU price dropsit replaces performance bonds
Best-in-class fleet failure rate after burn-in — the number acceptance testing hands to operations — is roughly…
1 failure per 512 GPUs per week1 failure per 10 GPUs per weekzero failures1 failure per GPU per year
Level-5 integrated systems testing for a liquid-cooled AI hall now takes about…
a weekend1–2 weeks10–14 weeksa year

Check your understanding: Project Delivery, Procurement & Risk