Is access to reliable power becoming a more binding constraint on AI datacenter growth than access to compute hardware?
Atomic facts extracted · comparative matrix not yet formed
The first governed research track compares power/grid constraints with compute-hardware constraints. The evidence register contains recoverable primary-source facts on both sides, but it does not yet support a dominant-constraint thesis.
Open canonical Research record →Power / grid
IEA evidence records rising data-centre electricity demand and conditional grid-delay risk; PJM records regional large-load growth, resource-adequacy pressure and policy responses.
Compute
NVIDIA reports remaining supply constraints while scaling commitments; TSMC reports major capacity investment and does not foresee bottlenecks to its own capacity-expansion plans.
Multiple constraints
Microsoft, Alphabet and NVIDIA disclosures show that compute, power, land, capital and broader datacentre capacity can interact rather than resolve into a single universal bottleneck.
Evidence boundary
No dominant constraint, live Signal, investment conclusion or numerical confidence is authorized. Forecasts remain forecasts, regional evidence remains regional, and company disclosures remain actor-specific.