Mike Donovan

Wes said the Macquarie partnership covers North America only, leaving open a similar European arrangement.

Wes said smaller 20-50 MW modular builds fit ChronoScale's capacity strategy rather than Applied's core campuses.

Wes said tenant fit-outs have become larger and more flexible, making the exact workload mix harder to discern.

Wes said the flexible AI-factory design handles GPUs/TPUs/CPUs but declined to give granular CPU-vs-GPU trends, as specs vary customer by customer.

Wes explained power generation has a fundamentally different risk and return profile than the data center business, so Base Electron was created as a separate company that raises its own capital, giving Applied shareholders upside without downside risk and letting it eventually trade publicly.

Wes affirmed every campus can scale to at least 1 GW (two to 2 GW+), giving a path to 5 GW+ of capacity by 2030-2032 without adding new campuses.

Wes described Corintis's chip-specific patterned cold-plate technology that keeps liquid volume constant as chip power density rises, helping future-proof infrastructure.

Wes said Applied Digital does not expect to contribute any additional funding to Polaris Forge One; it will be funded by project finance and Macquarie financing.

Wes said industry lead times have stretched, but Applied Digital secured its supply about two years ago and bought manufacturing capacity, so it hasn't seen much pricing inflation.

Kent Draper said pricing is very consistent globally, with possible scarcity premiums for sovereign AI locations.

Kent Draper confirmed other Spanish development sites hold the balance, with Badajoz the nearest-term, largest site.

Kent Draper said Spain's power and sites are secured with permitting well advanced at several sites; near-term power availability was the key attraction and IREN is already seeing direct customer requests for European capacity.

Dan Roberts said the mix will evolve organically toward a diversified base of hyperscalers, AI natives, enterprises and sovereign customers, distinguishing latency-sensitive inference from training workloads that can tolerate distance (e.g. Australia).

Land and full 1.6 GW of power are secured; remaining work is master planning and local permitting for 2028 ramp.

A phased build-out triangulated against customer demand, capital availability and construction pace.

Kent said customers overwhelmingly want a bare metal offering and layer their own orchestration; IREN can move up the software stack if demand emerges but hasn't seen it.

TAE technology is integrated into Vertebra for load-following at variable-demand sites and as a load bank for start-up data centers.

Roughly 90% of the ~70% owned-and-operated capacity is being evaluated for AI; expiring hosted capacity can be run at Matagorda while AI sites are built.