Applied Digital collects its $36 billion AI backlog only as fast as it builds

Revenue rose $261.0 million from the prior-year quarter ^[23]. That is slightly less than the $262.6 million that high-performance computing hosting brought in on its own ^[10]. Hosting, the segment that runs AI workloads for large customers, was about 77% of the quarter's revenue ^[26].

The contracts are priced on usable power, or more precisely on critical IT load: the electricity available to run servers, not the total a site pulls ^[7]. In Crypto Briefing's account, Polaris Forge 1's live capacity equals about 18% of the leased load ^[25]. The results as reported do not give a company-wide live total. Deploying more than 600 MW over the next 12 months, as planned ^[4], would add the equivalent of about 43% of the leased load ^[19].

The Seeking Alpha author puts Applied Digital's debt at $6.4 billion ^[6]. Annualize the quarter's adjusted EBITDA and the debt is about 25 times it ^[21]. That multiple shrinks only if new megawatts start billing faster than the company adds borrowing to build them. If the schedule slips, the debt stays in place and the earnings come later.

Applied Digital is building through the loss. Its headline project is a 1.2 GW natural gas facility in North Dakota ^[15], and management tied the 2030 target to long-term power purchase agreements and hyperscaler demand ^[5]. The company is choosing construction over a smaller reported loss ^[12].

If the build lands on schedule, the loss should narrow as leased megawatts start billing. If construction or power slips, interest keeps accruing while the backlog waits. The Seeking Alpha author argues for a third path, in which expansion leases close at the 15% higher rates and longer durations management is targeting, helped by scarce power and ready-to-build sites ^[17]. I think the $36 billion ^[1] is good evidence of demand for the capacity already leased and thin evidence for the 2030 target, because the first is signed and the second is still a plan. The higher-rate leases are the strongest case against that view: if hyperscalers pay a premium for powered sites, the harder problem is financing the remaining capacity, not finding tenants for it.

My view that the risk sits in construction and financing would be wrong if the North Dakota site hits its year-end capacity target ^[14] and the GAAP loss still does not narrow over the following quarters. In that case the problem would be what each megawatt earns, about $25.5 million of base-term revenue apiece ^[18], set against what it costs to build and carry.