The answer first: Applied Digital is a partial engine fit, and as of September 23, 2026 only one of five pillars passes. The thesis is real: AI compute infrastructure is a multi-decade runway, and the company reports about $36 billion of 15-year take-or-pay contracted revenue from investment-grade tenants across four AI Factory campuses. But the build behind the backlog is levered and cash-burning: about $5 billion of long-term debt against about $1.7 billion of equity, a debt-to-equity ratio near 2.9, and free cash flow of roughly negative $2.8 billion. The engine wants the business, not just the backlog. Reference price $27.30, market cap about $7.9 billion.
Pillar 1: Sunrise sector, PASS
AI compute infrastructure is a multi-decade demand wave with evidence, not adjectives. About 1.4 gigawatts of contracted critical IT load under 15-year take-or-pay leases worth about $36 billion in base-term revenue: CoreWeave holds 400 MW at Polaris Forge 1 (Ellendale, North Dakota) under a contract expanded to about $11 billion, with the first 100 MW Ready for Service in November 2025; a 200 MW Polaris Forge 2 (Harwood, North Dakota) lease worth about $5 billion went to an investment-grade hyperscaler, with the $3 billion campus breaking ground in September 2025; and a US investment-grade hyperscaler took 300 MW at Delta Forge 1 (Boyce, Louisiana) plus 300 MW at Polaris Forge 3, each worth about $7.5 billion, with initial operations expected in calendar 2027. Already at the ceiling; the downgrade triggers would be an AI compute demand collapse, a wave of lease cancellations, or chronic overbuild.
Pillar 2: Leadership, PARTIAL
Wes Cummins co-founded the company in 2021 with Jason Zhang as Applied Blockchain and renamed it Applied Digital in November 2022, exploring AI workloads for the company's power sites by mid-2022, ahead of the industry pivot. He is a technology investor by background: B. Riley from 2002 (publishing equity analyst covering tech hardware, Director of Research, then President), founder of 272 Capital (2020, sold to B. Riley 2021), and eight years at Nokomis Capital focused on TMT special situations. He holds roughly 4.2 million shares directly. The delivery record earns credit: he called the AI pivot early, signed about $36 billion of contracted revenue, and delivered construction on schedule. Against: a financier with no prior operating-company scale, a heavy dilution record (including a 49.4-million-share $160 million placement in September 2024), and a resolved-but-real June 2023 governance blemish, when the Audit Committee investigated a threatened, never formally asserted, harassment allegation arising from a personal relationship with then-CMO Regina Ingel, determined it was consensual and unfounded, and the board reaffirmed him as CEO. Upgrade path: two more years of clean delivery with no governance flags. Downgrade triggers: a governance flag, a major construction delay, or a guidance cut.
Pillar 3: Moat, PARTIAL
The $36 billion of contracted 15-year take-or-pay backlog with investment-grade counterparties is genuine commercial traction, and the build spec is competitive: closed-loop liquid cooling, a PUE of about 1.18, and 12-to-14-month construction timelines. But there is no structural barrier: powered land, buildings, and liquid cooling are replicable, and others are replicating them, with IREN signing a $9.7 billion five-year GPU cloud deal with Microsoft and TeraWulf and Cipher Mining pursuing the same hyperscaler lease model. Customer concentration is real, with CoreWeave alone accounting for about $11 billion of the backlog. And Nvidia ended its equity endorsement: it participated in the $160 million placement in September 2024, then sold its entire 7.72-million-share stake in Q4 2025 (disclosed February 2026), sending the stock down about 8.4% after hours. APLD remains an Nvidia Cloud Partner building on Nvidia GPUs, so the equity exit is a portfolio rebalance, not a commercial divorce. Upgrade: backlog growth and re-leasing at higher rates as the first campuses prove out. Downgrade: cancellations or renegotiations.
Pillar 4: Iron fortress, FAIL
About $4.97 billion of long-term debt against about $4.15 billion of cash and about $1.72 billion of common equity: debt-to-equity near 2.9 and debt-to-capital near 0.74. The cash is construction funding, not a fortress, and it arrived through financing, not through the business: $2.15 billion and $1.59 billion of 6.75% senior secured notes due 2031, a $430 million revolver, a $5 billion Macquarie perpetual preferred facility, and a $300 million Goldman bridge facility that has been repaid. The company cannot survive a bad two years without help. Upgrade: sustained free-cash-flow-positive quarters with net debt materially reduced, interest covered from operations.
Pillar 5: Free cash flow, FAIL
Free cash flow is roughly negative $2.8 billion against capital spending of roughly $2.9 billion: the campuses are being built with financed money, not with cash the business throws off. Operating cash flow of about $89.7 million in the latest quarter is progress, as are adjusted EBITDA of $107.2 million for fiscal 2026, adjusted net income of $36.1 million, and Net Operating Income of $90.4 million. The GAAP picture is darker: a $249.2 million fiscal 2026 net loss driven by heavy stock-based compensation and depreciation. The path is clear but unproven: fifteen-year take-or-pay leases convert to cash as buildings reach Ready for Service through 2027. Upgrade: two consecutive FCF-positive quarters.
What would change each grade
Everything about this package resolves through 2027, when the 1.4 GW of contracted capacity comes ready for service. Sunrise is already at the ceiling. Leadership upgrades on two more years of clean delivery with no governance flags; a governance flag, a major construction delay, or a guidance cut downgrades it. Moat upgrades on backlog growth and re-leasing at higher rates; cancellations or renegotiations downgrade it. The fortress and free-cash-flow fails only move when take-or-pay cash converts to two consecutive FCF-positive quarters with net debt materially reduced. If that conversion happens, the partials have their upgrade path too. If it does not, the leverage stays and the fails stay.
Bottom line: One of five pillars passes on evidence. The market prices it that way: at about $7.9 billion, APLD trades at roughly 0.2x its $36 billion contracted backlog, the cheapest multiple of contracted revenue in the AI data-center group, which is exactly what levered, cash-burning builds trade at. The sunrise sector is real, the backlog is real, and the 2027 conversion is the whole bet.
Download the full analysis (PDF)
Research and opinion, not investment advice. Do your own due diligence before investing.