Purpose-built AI data center operator. AGI Score: 10/10 (highest possible). Leopold holds $278M (~6.3% of AUM). 286 MW blockchain hosting live + 600 MW HPC under construction. | Analysis date: 2026-03-12
Applied Digital scored a perfect 10/10 in our AGI impact framework -- one of only 5 companies out of 3,648 to achieve this. Leopold Aschenbrenner holds 11.3M shares worth $278M (6.3% of his AUM), having accumulated aggressively from $5.62/share in Q1 2025 to $24.52/share in Q4 2025. APLD is not a Bitcoin miner pivoting to AI -- it is a purpose-built AI data center operator that happens to also run blockchain hosting. It has $15.5 billion in contracted minimum lease payments from CoreWeave and a "US-based investment grade hyperscaler" (likely Oracle). The question: is this a real infrastructure business, or is it a levered bet on one customer (CoreWeave) that could blow up?
52-week range: $3.44 - $41.35. All-time low: $0.88 (early 2024). Leopold's avg cost: ~$14.51 (+89% gain).
Applied Digital operates in two segments, with a third being sold off:
GPU cloud service operating in CO, MN, UT. Generated $84M in FY2025 revenue but classified as held-for-sale. Being spun off into "ChronoScale" via merger with EKSO Bionics. This is a distraction from the core thesis -- APLD is exiting this business to focus purely on data center infrastructure.
This is the critical distinction. APLD does not own GPUs. It builds the building, brings the power, installs cooling, and leases the energized space to tenants (CoreWeave, a hyperscaler) who bring their own compute equipment. The revenue model is:
| Campus | Building | MW | Customer | Status | Expected Online |
|---|---|---|---|---|---|
| Jamestown, ND | Data Center | 106 | Crypto miner | Operating | Live |
| Ellendale, ND | Data Center | 180 | Crypto miner | Operating | Live |
| Subtotal: Blockchain Hosting | 286 | ||||
| Polaris Forge 1 | Building 2 (ELN-02) | 100 | CoreWeave | Energized Q2 FY26 | Late 2025 |
| Polaris Forge 1 | Building 3 (ELN-03) | 150 | CoreWeave | Under Construction | 2026 |
| Polaris Forge 1 | Building 4 | 150 | CoreWeave | Under Construction | 2026-2027 |
| Subtotal: Polaris Forge 1 | 400 | ||||
| Polaris Forge 2 | Buildings 1-2 | 200 | US hyperscaler (likely Oracle) | Under Construction | 2026-2027 |
| Subtotal: Polaris Forge 2 | 200 | ||||
| TOTAL CAPACITY (all campuses) | 886 | ||||
Blockchain (live) vs HPC (under construction)
Total: 886 MW across 3 locations. 100% of HPC capacity is under long-term contract.
On March 4, 2026, APLD guaranteed a Design-Build Agreement between Base Electron (a subsidiary/affiliate) and Babcock & Wilcox for a 1.2 GW power generation facility. This is a natural gas power plant that could supply power to APLD's own data centers. In exchange, APLD received 10% equity in Base Electron. If built, 1.2 GW of captive power would be transformative -- it removes the single biggest bottleneck (grid power availability) and could support 1,000+ MW of additional data center capacity. But this is early-stage and speculative.
| Metric | FY2023 May 2023 | FY2024 May 2024 | FY2025 May 2025 | H1 FY2026 Nov 2025 |
|---|---|---|---|---|
| Revenue | $55M | $137M | $144M | $191M |
| Revenue Growth | -- | +147% | +6% | +169% (vs H1 FY25) |
| Cost of Revenue | $44M | $107M | $101M | $156M |
| Gross Profit | $11M | $30M | $43M | $35M |
| Gross Margin | 20% | 22% | 30% | 18% |
| SG&A | $54M | $45M | $83M | $86M |
| Operating Loss | -$43M | -$33M | -$17M | -$53M |
| Net Loss (cont. ops) | -$44M | -$74M | -$158M | -$59M |
| Adjusted EBITDA | $1M | $22M | $20M | -- |
| SBC | $32M | $17M | $23M | -- |
FY2025 revenue ($144M) was entirely from the blockchain hosting business. HPC revenue has barely started. The first HPC building (100 MW at Polaris Forge 1) was energized in Q2 FY2026 (fall 2025). H1 FY2026 revenue of $191M already exceeds full-year FY2025. When all 600 MW of HPC capacity is online (expected 2027), annual contracted lease payments will be approximately $856M/year based on the contracted schedule. This is a 6x revenue increase from today's run rate.
| Asset | Nov 2025 | May 2025 | Notes |
|---|---|---|---|
| Cash & Equivalents | $1,913M | $42M | Massive increase from debt/equity raises |
| Restricted Cash | $382M | $72M | Debt service reserves for construction |
| PP&E (net) | $2,195M | $1,295M | Data centers under construction |
| Construction in Progress | ~$1,500M | $1,123M | HPC campus build-out |
| Other Assets | $508M | $181M | Including deferred financing costs |
| Total Assets | $5,229M | $1,870M | 2.8x increase in 6 months |
| Liabilities | |||
| Accounts Payable | $93M | $248M | Paying down construction AP |
| Accrued Liabilities | $207M | $30M | |
| Deferred Revenue | $29M | $0 | Customer prepayments starting |
| Long-term Debt | $2,274M | $678M | Senior Secured Notes + other |
| Total Liabilities | $3,447M | $1,236M | |
| Equity | |||
| Common Stock Equity | $1,332M | $498M | Massive share issuance |
| Preferred / NCI (Temp Equity) | $450M | $136M | Macquarie preferred units |
| Shares Outstanding | 279M | 225M | +24% dilution in 6 months |
| Metric | FY2023 | FY2024 | FY2025 | H1 FY2026 |
|---|---|---|---|---|
| Operating Cash Flow | $59M | $14M | -$115M | -- |
| Capital Expenditure | -$131M | -$142M | -$682M | ~-$1,000M |
| Free Cash Flow | -$73M | -$128M | -$797M | ~-$1,000M |
| Debt Issued | $82M | $145M | $1,112M | ~$2,500M |
| Equity Issued | $0 | $131M | $390M | ~$700M |
APLD is a capital-raising machine. In the last 18 months, it has raised roughly $4-5 billion through a combination of senior secured notes ($2.15B for Forge 2 + $1.3B for Forge 1), equity issuance (ATM offerings, PIPEs), preferred stock, and the Macquarie JV ($450M). Total debt stands at ~$2.6B. Shares outstanding have grown from 96M (FY2023) to 280M today -- nearly 3x dilution. On top of that, there are ~28M warrants outstanding and ~15M unvested RSUs. Fully diluted share count is approximately 322M shares, implying a fully diluted market cap of ~$8.8B.
This is the single most important number in the entire analysis. As of November 30, 2025, APLD has $15.55 billion in minimum contracted lease payments:
| Fiscal Year | Minimum Payments | Notes |
|---|---|---|
| FY2026 (remaining) | $72M | Forge 1 Building 2 coming online |
| FY2027 | $480M | Forge 1 ramp + Forge 2 initial |
| FY2028 | $856M | Full run-rate approaching |
| FY2029 | $883M | Full capacity + escalators |
| FY2030 | $908M | Escalators continuing |
| Thereafter | $12,352M | ~13 years remaining at ~$900M+/yr |
| Total | $15,551M |
APLD has extreme customer concentration. Every dollar of HPC revenue comes from (or will come from) just two customers:
| Metric | APLD | CORZ |
|---|---|---|
| Primary HPC Customer | CoreWeave (400 MW) | CoreWeave (590 MW) |
| Customer Diversification | 2 customers (CoreWeave + hyperscaler) | 1 customer (100% CoreWeave) |
| Lease Duration | 15 years | 12 years |
| Credit Quality of 2nd Customer | Investment grade (Oracle?) | N/A -- no second customer |
| Customer Warrants Issued | ~21M shares to CoreWeave | Significant dilution to CoreWeave |
On October 3, 2025, APLD closed a joint venture with Macquarie Asset Management (MAM) -- one of the world's largest infrastructure investors (~$600B AUM). Key terms:
| Assets | Amount | Liabilities | Amount |
|---|---|---|---|
| Cash | $1,640M | Accounts Payable | $76M |
| Restricted Cash | $178M | Accrued Liabilities | $193M |
| PP&E | $1,832M | Long-term Debt | $2,273M |
| Other | $742M | Other | $68M |
| Total Assets | $4,392M | Total Liabilities | $2,610M |
| Net Assets | $1,782M | ||
Why Macquarie matters: Infrastructure investors like Macquarie underwrite 15-20 year asset lives. They don't invest $450M on a whim. Their participation validates the physical asset, the power contracts, and the customer leases. It also provides APLD with a sophisticated partner for future capital raises and expansions.
| Metric | APLD | CORZ | Edge |
|---|---|---|---|
| AGI Score | 10/10 | 9/10 | APLD |
| Market Cap | $7.7B | $5.1B | -- |
| Revenue (most recent FY) | $144M | $319M | CORZ |
| HPC Revenue (current) | Starting (Q2 FY26) | $65M (FY2025) | CORZ |
| Total MW Capacity | 886 MW | 1,400 MW (gross) | CORZ |
| HPC MW Under Contract | 600 MW | 590 MW | ~Tie |
| Primary HPC Customer | CoreWeave + hyperscaler | CoreWeave only | APLD |
| Contracted Revenue | $15.6B (15yr) | $8.7B (12yr) | APLD |
| Total Debt | $2.6B | $1.2B | CORZ |
| Stockholders' Equity | $1.3B | -$963M | APLD |
| Shares Outstanding | 280M | 315M | -- |
| EV/Contracted MW | $15.3M/MW | $10.3M/MW | CORZ (cheaper) |
| Leopold Position | $278M (passive) | $419M (13D activist) | -- |
| Purpose-Built for AI? | Yes | No (converting mining) | APLD |
| Origin | Purpose-built AI DC | Bitcoin miner pivot | APLD |
| Employees | 205 | ~500 | -- |
CORZ is a Bitcoin miner converting existing mining infrastructure to AI hosting -- it has legacy assets, legacy operations, and negative equity. APLD is a purpose-built AI data center developer that happens to run some blockchain hosting on the side. APLD's facilities are designed from the ground up for high-density GPU workloads (40-100+ kW/rack), while CORZ is retrofitting facilities originally built for 5-10 kW/rack mining. APLD trades at a premium ($15.3M/MW vs CORZ's $10.3M/MW) precisely because of this. The question is whether the premium is justified -- and the $15.6B contract book vs CORZ's $8.7B suggests it might be.
| Metric | Value |
|---|---|
| Total Shares | 11,339,060 |
| Current Value | $278M |
| Average Cost Basis | ~$14.51/share |
| Total Cost | ~$164M |
| Unrealized Gain | +$146M (+89%) |
| % of Leopold AUM | 6.3% |
| % of APLD Outstanding | ~4.1% |
| Filing Type | Passive (no 13D) |
Leopold's behavior here is telling. He bought at $5.62, watched it run to $23, trimmed slightly, then added $129M more at $24.52. That Q4 2025 addition was one of his largest single-quarter purchases across his entire portfolio. He clearly has high conviction that APLD is undervalued even at current levels. However, unlike CORZ, he has not filed a 13D -- this is a passive investment, not an activist one.
| Valuation Basis | EV/MW | Implied EV (600 MW HPC) | Implied Share Price |
|---|---|---|---|
| Bitcoin miner (crypto multiple) | $3-5M | $1.8-3.0B | $0-5 |
| Transitioning (blended) | $5-8M | $3.0-4.8B | $2-9 |
| AI/HPC infrastructure | $10-15M | $6.0-9.0B | $12-22 |
| Premium AI infra (purpose-built) | $15-20M | $9.0-12.0B | $23-34 |
| Best-in-class (Equinix comps) | $20-30M | $12.0-18.0B | $34-55 |
Note: Price calculations use 600 MW HPC + ~$2.6B net debt, ~280M shares. Blockchain hosting adds ~$1B EV at $3-5M/MW for 286 MW.
| Scenario | Probability | Midpoint Price | Weighted |
|---|---|---|---|
| Bull (everything works) | 60% | $28 | $16.80 |
| Base (delays, lower margins) | 30% | $10 | $3.00 |
| Bear (serious problems) | 10% | $3 | $0.30 |
| Expected Value | $20.10 |
At $27.48, the market is pricing APLD at or above the midpoint of our bull case. The probability-weighted expected value is ~$20, which means the stock is roughly 37% overvalued on an expected-value basis. You're paying for success. The question is whether the upside tail extends further -- if APLD builds beyond 600 MW (the 1.2 GW power facility suggests this is the plan), the bull case goes to $50+.
Bull case terminal value: If APLD executes on its full pipeline and captures AI demand tailwinds, a bull case of $50-76B market cap is conceivable (requires $3-3.8B EBITDA at 20x, implying 4-7 GW of operational HPC capacity). 10x entry = $5-7.6B market cap, or ~$1.80-2.75/share. Current price of $27.48 is 10-15x above the 10x entry zone even under the most aggressive bull case. At a more conservative bull case of $30B, 10x entry = $3B (~$1.08/share) -- essentially back to pre-pivot levels. The stock has already done its 10x move from $2-3 to $27+. From here, the realistic upside is 2-3x if the bull case plays out, not 10x.
| Entry Level | Price | Rationale |
|---|---|---|
| Back-up-the-truck | $8-12 | Below expected value, margin of safety on assets. Only if CoreWeave fear creates panic. |
| Attractive entry | $12-18 | Decent discount to expected value. Leopold bought at $14.51 avg. |
| Fair value | $18-25 | Roughly fair for the risk profile. No margin of safety. |
| Overvalued | $25+ | Current price. Priced for perfection. |
YES, equity could theoretically be wiped out. If CoreWeave defaults on leases AND the hyperscaler walks AND construction costs overrun, the $2.6B debt could exceed asset values. The buildings would have some liquidation value but purpose-built AI data centers in North Dakota are not easily repurposed.
But true zero is very unlikely. The physical assets (land, buildings, power infrastructure, transformers) have real value. Even in liquidation, PP&E of $2.2B and $1.9B cash provides a floor. At zero equity, the stock would be ~$0 -- but this requires essentially all contracted revenue to disappear AND assets to be worth less than debt. This is a ~5% probability event.
| Level | Our Assessment |
|---|---|
| Can we compute a floor? | Yes, but with low confidence |
| Estimated asset floor | $5-8/share |
| Rationale | $1.9B cash + $2.2B PP&E - $2.6B debt - $0.45B preferred = ~$1.05B net. Divided by 280M shares = ~$3.75. Add premium for contracted revenue: ~$5-8. |
| Confidence | Low |
| Reason for low confidence | Heavy leverage, no sustained profitability, customer concentration, construction risk. Too many unknowns to feel good about a floor estimate. |
| Dimension | Rating | Notes |
|---|---|---|
| AGI Tailwind | 10/10 | Perfect score. AGI is the customer. Can't be disrupted by AI. |
| Revenue Visibility | 9/10 | $15.6B contracted over 15 years. Almost unprecedented for a company this size. |
| Execution Risk | 3/10 | Building 600 MW of AI data centers with 205 employees is extremely ambitious. C-suite turnover adds risk. |
| Balance Sheet | 5/10 | $1.9B cash but $2.6B debt. Net negative. Levered to construction execution. |
| Customer Risk | 4/10 | Two customers, one of which (CoreWeave) is itself a credit risk. |
| Valuation | 5/10 | $27.48 prices in near-full execution. Limited margin of safety at current price. |
| Management | 5/10 | New team, multiple departures, but Macquarie partnership provides governance backstop. |
| Dilution Risk | 3/10 | 3x dilution already. 28M warrants outstanding. More capital raises likely. |
| Leopold Signal | 8/10 | $278M position, buying at $24.52. Strong conviction signal but passive (not activist like CORZ). |
| Competitive Position | 8/10 | Purpose-built for AI, not converted mining. Proprietary waterless cooling. North Dakota power + land advantage. |
Applied Digital is a high-conviction, high-risk bet on AI infrastructure demand. The $15.6B contract book is extraordinary and provides revenue visibility that almost no company of this size possesses. But the risk profile is significant: $2.6B debt, CoreWeave concentration, construction execution, and 3x dilution already inflicted on shareholders.
At $27.48, the stock is fairly to slightly overvalued on a probability-weighted basis. Leopold clearly disagrees -- he added $129M at $24.52 -- but he may be underwriting a 5-10 year thesis that assumes APLD builds well beyond 600 MW.
The right approach for us:
Floor Price Confidence: LOW. Too much leverage, too much construction risk, too much customer concentration to feel confident about any floor. This is an asymmetric bet, not a value investment. If it works, it could be a 3-5x from here. If it doesn't, equity could go to near-zero.