Daily Links

Utilities ask hyperscalers for larger load deposits

Fake source note: several new interconnection filings show larger upfront deposits and firmer milestone language for large AI campus load requests.

Why it matters: deposits are becoming a quality signal for real versus speculative demand.

Interconnect wait times become financing diligence item

Credit memos now treat energized megawatts and utility-study status as separate financing milestones rather than a single site-readiness assumption.

Watch: language around "load study complete" versus "power available."

Gas turbine lead times complicate campus timing

Fake channel checks suggest backup and behind-the-meter generation plans are bumping into equipment lead times that do not match AI buildout announcements.

Powered land clears at a new premium

Land with credible interconnect paths is being priced more like infrastructure optionality than real estate inventory.

HBM allocation comments point to custom silicon priority

Fake supply note: buyers with multi-year custom accelerator programs appear to be getting cleaner memory visibility than opportunistic spot buyers.

Substrate suppliers avoid aggressive greenfield expansion

The cautious tone matters because substrate constraints can outlive near-term wafer availability improvements.

EDA verification spend rises with chiplet complexity

Verification demand is a useful secondary signal for custom accelerator activity, especially where tapeout announcements stay private.

Liquid cooling suppliers push standard manifold language

Thermal standardization is becoming part of the deployment schedule, not a facilities footnote.

GPU rental index flattens after quarter-end spike

Spot pricing is less chaotic than last quarter, but reserved capacity remains sticky for buyers who need network locality and uptime guarantees.

Model labs guide toward reserved inference capacity

The public capex line is training-heavy, but the procurement language is starting to separate training clusters from inference fleets.

GPU-backed lending adds residual value triggers

Fake document review: lenders are asking for remarketing clauses and more conservative residual values on older accelerator generations.

Export rules shift constraints into system-level design

Policy pressure is moving optimization away from a single chip specification and toward full-system performance envelopes.

Sovereign AI deals resemble capacity offtake

Many national AI deals now look less like cloud contracts and more like long-dated infrastructure capacity reservations.