Applied Digital Corp. APLD

Key data

The business

Applied Digital builds power-dense data center campuses and rents them to companies that run AI. It is a landlord, not a cloud operator, and it chose to be: it sold off its own GPU-rental business and now leases finished buildings under long contracts. The flagship is Polaris Forge in Ellendale, North Dakota, where about 100 MW is operating and revenue-generating. Behind it sits a development pipeline of over 3 GW of utility power. The moat, such as it is, is holding land, executed utility power agreements and a repeatable build in the few places where hundreds of megawatts can actually be energized soon; power access, not technology, is the scarce thing. The tenant holds a warm hall of liquid-cooled racks. Applied Digital holds the dirt, the substation contract and the debt.

The numbers

The most recent reported quarter, Q4 of fiscal 2026 (ended May 31), lost $0.39 a share against a Street looking for a loss near $0.11, the third straight quarter the loss ran wider than expected as depreciation on the new campus landed. For the full year, revenue tripled while the operating loss ballooned, the signature of a builder spending ahead of the rent.

Revenue roughly doubled once the Ellendale halls energized, but the per-share loss jumped fivefold in the same stretch: this is a business whose costs arrive before its lease revenue does.

The loss per share narrowed across the last three years even as the operating loss more than tripled, a gap explained by a rising share count, not by the business turning. Cash from operations swung positive in FY2026 to $90 million, but capital spending ran near $3 billion (derived from the capex-to-revenue ratio of 4.97), so free cash flow is deeply negative and the balance sheet carries about $3.5 billion of net debt (enterprise value minus market cap) against $1.6 billion of cash, with interest coverage below zero.

The market values this on a promise, not a print: at $26.37 you pay about 19 times the $1.39 of earnings analysts pencil in for fiscal 2029, three years and one enormous construction cycle away, on a consensus built from only two or three estimates. What I believe that the tape does not is nothing clever: this is priced for the build finishing on time and the tenants paying under their long-term leases, and the single figure that settles it is megawatts energized converting into lease revenue on the total revenue line, quarter after quarter.

Management

No insider has bought a share in twelve months; twenty sales totaled $24.3 million. Founder-CEO Wes Cummins sold $6.0 million in January 2026, an officer $3.4 million in October, and a director $2.34 million in August 2026. The Form 4 plan status was not in the filings pulled this run, so I can't split scheduled sales from discretionary ones, which matters more than usual when the CEO is selling into a stock that has fallen by half. The company repurchased $31 million of stock in FY2025 and zero in FY2026, the right call for a business that needs every dollar for concrete and copper. Pay is tied to a company still deep in the red at the operating line, and management has missed its own guided losses three of the last four quarters as depreciation outran the plan.

How it fails or surprises you

Leverage and liquidity. Net debt near $3.5 billion, debt-to-equity of 2.9, negative interest coverage and free cash flow that is deeply negative on about $3 billion of annual capex. If AI capital spending cools or a construction milestone slips before lease revenue ramps, refinancing turns expensive fast. Watch net debt and interest expense against lease revenue commencing over the next two quarters.

Tenant concentration. The revenue that underwrites everything leans on a handful of leases with large customers the company describes as investment-grade hyperscalers but does not name. The filing's language is generic; specific tenant names, contracted capacity beyond the 100 MW operating, and lease terms are not disclosed. The accelerating quarterly loss, from $0.07 to $0.39, sits awkwardly next to a story sold on long-dated visibility. A material tenant renegotiation or non-renewal would break the thesis outright.

Pipeline conversion (right tail). Over 3 GW of utility power under development against roughly 100 MW operating today. Sign additional large tenant leases and energize another 250-plus MW on schedule, and revenue re-rates toward the $2.9 billion pencilled for 2029 while the multiple compresses. The market pays for the existing operating capacity, not the pipeline; the print is the next executed lease and MW energized.

Closing thoughts

Nothing in the next quarter proves the 2029 revenue Wall Street has drawn; what decides this is more basic, whether the buildings get financed to completion and the tenants keep paying. So this is a survival question wearing a growth costume. The left tail is real and fat: a $3.5 billion debt load and negative operating margins leave little cushion if either AI capex or a major tenant wobbles, and that is the shock that permanently impairs the equity. Against it sits a pipeline the price is not paying for, which is where the surprise lives if the build clicks. On judgment, not arithmetic, the downside is a financing or counterparty event and the upside is converting the development pipeline into signed leases and operating megawatts.

The bet is still that the AI boom keeps needing giant buildings full of power and cooling, that Applied Digital finishes building them and rents them to large cloud operators, and that those operators keep paying their long-term leases while the pipeline gets built. It breaks if a campus slips or a tenant strains before the rent covers the interest. The one pair to watch each quarter: megawatts energized at operating facilities against interest expense and net debt. If energized capacity stalls while net debt climbs, the thesis is wrong regardless of what the pipeline promises.