Solid oxide fuel cells for on-site power generation. The "pick and shovel" play for AI data center power. Leopold's #1 position. | Analysis date: 2026-03-12
Leopold Aschenbrenner -- the most AGI-bullish public investor -- made Bloom Energy his #1 position at $1.7B (22.5% of his portfolio). He tripled his shares from Q3 to Q4 2025 (10.1M shares + 408K calls). Bloom makes solid oxide fuel cells that convert natural gas to electricity without combustion. The thesis: AI data centers need power faster than the grid can provide, and Bloom can deploy on-site power in 90 days vs years for grid connections. The question: at $44B market cap and 22x revenue, has the market already priced in everything?
Bloom manufactures solid oxide fuel cell (SOFC) systems -- modular power generators that convert natural gas (or hydrogen, or biogas) directly into electricity through an electrochemical process. No combustion. No turbines. No moving parts. Fuel flows in, electricity comes out.
| Segment | FY2025 Revenue | % of Total | Type | Gross Margin |
|---|---|---|---|---|
| Product (Equipment) | $1,531M | 75.7% | One-time | 35% |
| Installation | $204M | 10.1% | One-time | -1% |
| Service (O&M contracts) | $228M | 11.3% | Recurring | 10% |
| Electricity (PPAs) | $60M | 3.0% | Recurring | 46% |
| Total | $2,024M | 100% | 86% one-time | 29% |
Key observation: Revenue is 86% one-time equipment sales, only 14% recurring. This means revenue visibility depends on new orders. The $5B Brookfield partnership and AEP 1GW framework provide some backlog, but this is not a SaaS-like recurring model.
We structure the analysis as a decision tree. Each question either eliminates the stock, moves to the next question, or changes the required entry price.
| Cash | $2,454M | Unrestricted |
| Total Debt | $2,618M | Mostly convertible |
| Net Cash Position | -$164M | Roughly break-even |
| Undrawn Revolver | $600M | Available if needed |
| Operating Cash Flow (FY2025) | $114M | Positive and growing |
The debt structure is favorable: $2.5B of the debt is 0% convertible notes due 2030. Zero coupon means no cash interest payments on the largest tranche. The remaining ~$100M is 3% green convertibles due 2028-2029. Total annual interest expense: ~$38M.
Bankruptcy risk is LOW. Cash roughly equals debt, operating cash flow is positive ($114M), and the company has a $600M undrawn revolver. Even if revenue growth stalls, the company can service its debt for years. However, the convertible notes create significant dilution risk -- 280M shares outstanding already grew from 229M a year ago.
Result: Proceed to next question.
Specifically: the AI compute buildout creates power demand that the grid cannot satisfy, forcing hyperscalers and colocation operators to bring their own power. Bloom captures a meaningful share of this behind-the-meter generation market.
Evidence this is happening:
Result: The thesis is real and has validation. But the question is price.
The correct way to frame 10x: start with the bull case terminal value, divide by 10, and compare to today's price.
| Bull Case Scenario | Revenue | Net Margin | Net Income | P/E | Bull Mkt Cap | 10x Entry Mkt Cap | 10x Entry Price* |
|---|---|---|---|---|---|---|---|
| Conservative bull | $8B | 25% | $2.0B | 30x | $60B | $6.0B | ~$21 |
| Mid bull | $10B | 28% | $2.8B | 30x | $84B | $8.4B | ~$30 |
| High bull | $12B | 30% | $3.6B | 30x | $108B | $10.8B | ~$39 |
*10x entry price = Bull Case Mkt Cap / 10 / 280M diluted shares.
Revenue growth assumptions:
The stock already 10x'd in the last year ($15 to $157). Bull case market cap: $72-108B. 10x entry = $7.2-10.8B ($21-39/share). At $157 ($44B market cap), you are 4-7x above the 10x entry zone. The stock touched $15 at its 52-week low -- at that price it was genuinely a 10x+ opportunity. But the market figured out the thesis and priced it in.
| Bull Case Assumptions | Value | Basis |
|---|---|---|
| Revenue by 2031 | $8-12B | Analyst consensus implies ~$5B by 2027; compound to $8-12B by 2031 |
| Net Margin at scale | 25-30% | Gross margin trending to 35-40%, OpEx leverage at scale |
| Net Income | $2.4-3.6B | $8-12B rev x 30% margin |
| Terminal P/E | 25-30x | High-growth industrial, decelerating but still growing |
| Implied Market Cap | $72-108B | $2.4-3.6B x 30x |
| Upside from $44B | 1.6-2.5x | Not 10x. Decent return but not life-changing. |
This is still a strong outcome -- doubling your money in 5 years is ~15% annualized. But it is far from 10x.
| Target Market Cap by 2031 | Required Entry for 10x | Entry Price/Share (~280M shares) |
|---|---|---|
| $72B (low bull) | $7.2B | ~$25/share |
| $90B (mid bull) | $9.0B | ~$32/share |
| $108B (high bull) | $10.8B | ~$39/share |
That is $4.2B market cap. At $15, this was genuinely a 10x opportunity (even a 15-20x if the bull case plays out). But that was before the AI power thesis took hold. The stock has since 10x'd. The market figured out the thesis and priced it in.
To get our 10x, we need BE at $25-35/share ($7-10B market cap). That is an 80% decline from current levels. It could happen in a broad market crash or if execution stumbles badly. But we cannot count on it.
| Floor Component | Value | Notes |
|---|---|---|
| Stockholders' Equity (Book) | $793M | Includes accumulated deficit of $3.99B |
| Tangible Book Value | ~$793M | No goodwill, no material intangibles |
| Cash | $2,454M | |
| Less: Total Debt | -$2,618M | Mostly 0% convertible due 2030 |
| Net Cash Position | -$164M | |
| PP&E (Net) | $399M | Manufacturing facilities (Fremont, Newark) |
| Inventory + AR + Contract Assets | ~$1.1B | Working capital |
| P/B at market | 57x | Extremely elevated |
Asset-based floor: ~$800M (tangible book value). But this understates the business value because:
Business-value floor: $2-4B. In a distressed scenario (revenue growth stalls, thesis breaks), the manufacturing capacity, contracts, and recurring service revenue support a valuation of $2-4B. This is a ~90% decline from current $44B.
At $44B, the downside to the floor is $40-42B (90%+). This is extremely asymmetric -- on the WRONG side. The realistic bull case gives 1.6-2.5x upside, but the downside to fair value (if growth disappoints) is 50-80%. This is the opposite of what we look for. We want limited downside and large upside. Here we have large downside and limited upside.
| Metric | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 | FY2026E | FY2027E |
|---|---|---|---|---|---|---|---|
| Revenue | $972M | $1,199M | $1,333M | $1,474M | $2,024M | $3,290M | $5,210M |
| Revenue Growth | 22% | 23% | 11% | 11% | 37% | 63% | 58% |
| Gross Profit | $198M | $148M | $198M | $405M | $587M | — | — |
| Gross Margin | 20% | 12% | 15% | 27% | 29% | — | — |
| Operating Income | -$115M | -$261M | -$209M | $23M | $73M | — | — |
| Net Income | -$193M | -$315M | -$308M | -$27M | -$87M | — | — |
| EPS (Diluted) | -$0.95 | -$1.62 | -$1.42 | -$0.13 | -$0.37 | $1.42 | $3.01 |
| Cash Flow | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 |
|---|---|---|---|---|---|
| Operating Cash Flow | -$61M | -$192M | -$373M | $92M | $114M |
| CapEx | -$50M | -$117M | -$84M | -$59M | -$57M |
| Free Cash Flow | -$111M | -$309M | -$457M | $33M | $57M |
Cash flow just turned positive in FY2024-2025. At $57M FCF on a $44B market cap, the FCF yield is 0.13%. Compared to KRC's 15% FCF yield, this shows how much future growth is priced in.
| Item | FY2024 | FY2025 | Notes |
|---|---|---|---|
| ASSETS | |||
| Cash & Equivalents | $803M | $2,454M | 3x increase from convertible raise |
| Accounts Receivable | — | $372M | |
| Inventory + Contract Assets | — | ~$700M | Working capital for manufacturing |
| PP&E (Net) | $403M | $399M | Manufacturing facilities |
| Total Assets | $2,657M | $4,397M | |
| LIABILITIES | |||
| Long-Term Debt | $1,014M | $2,614M | $2.5B 0% convertible added Nov 2025 |
| Current Liabilities | $637M | $624M | |
| Total Liabilities | $2,072M | $3,604M | |
| EQUITY | |||
| Stockholders' Equity | $585M | $793M | Accumulated deficit: $3.99B |
| Instrument | Principal | Coupon | Maturity | Risk |
|---|---|---|---|---|
| 0% Convertible Senior Notes | $2,500M | 0% | Nov 2030 | Major dilution if converted |
| 3% Green Convertible (remaining) | ~$100M | 3.0% | Jun 2028 | Mostly exchanged |
| 3% Green Convertible (extended) | ~$75M | 3.0% | Jun 2029 | Mostly exchanged |
| Revolving Credit Facility | $600M | SOFR+ | Dec 2030 | Undrawn, available |
The 0% convertible is clever financing: Bloom borrowed $2.5B at zero interest, due 2030. If the stock stays high, the notes convert to equity (dilution). If the stock crashes, Bloom must repay $2.5B in cash in 2030 (refinancing risk). The cash-vs-debt balance makes this manageable but not without risk.
| Customer | FY2025 Revenue Share | FY2024 Share | Trend |
|---|---|---|---|
| Customer 1 (key hyperscaler) | 43% | 23% | Doubling concentration |
| Customer 2 | 13% | 16% | |
| Customer 3 | 12% | 14% | |
| Top 3 Total | 68% | 53% | Increasing |
This is alarming. 43% of revenue from one customer is extreme concentration. The filing references "share-based consideration" and a "key hyperscaler" -- likely related to the Brookfield AI Fund partnership. If this customer pulls back, cancels, or delays, Bloom's revenue growth collapses. The concentration is INCREASING, not decreasing.
| Shares | 10,100,000 |
| Call Options | 408,000 contracts |
| Position Value (at cost) | ~$1.7B (22.5% of portfolio) |
| Portfolio Rank | #1 (largest position) |
| Growth | Tripled shares Q3 to Q4 2025 |
Leopold is ALL-IN on the "power is the bottleneck for AI" thesis. BE is his #1 position. He also holds EQT (natural gas), CORZ (data center infrastructure), and CRWV (GPU cloud). These all fit the same thesis: the physical infrastructure for AI is the constraint, not the software.
But note: Leopold likely accumulated much of his position at lower prices (the stock was $15-40 for most of 2024-early 2025). His cost basis is probably well below $157. He has already made 3-8x on this position. What was a 10x opportunity for him may only be a 2x opportunity for us at current prices.
| TAM Component | Value | Source / Assumption |
|---|---|---|
| Current US data center power | ~17 GW | IEA |
| Projected US DC power by 2030 | 35-45+ GW | Goldman Sachs (160% growth) |
| % going behind-the-meter by 2030 | 27% | Jan 2026 industry survey |
| On-site generation demand (US) | ~10.8 GW | 27% x 40 GW |
| At Bloom pricing ($5,000-8,000/kW) | $54-86B | US hardware TAM alone |
| Bloom current installed base | ~1-2 GW | All applications |
| Bloom current capacity | 1 GW/yr | Expanding to 2 GW/yr by end 2026 |
The TAM is massive -- $54-86B in US data center hardware alone, before considering international markets and recurring service revenue. Bloom's current $2B revenue is capturing less than 4% of the projected US TAM. There is enormous room to grow.
But TAM is not destiny. Bloom needs to actually capture this market against potential competitors, execution challenges, and the possibility that the TAM itself is overstated (what if grid buildout accelerates, or AI power demand is lower than projected).
| Capacity Metric | Current | By End 2026 | Potential Max |
|---|---|---|---|
| Annual production capacity | 1 GW | 2 GW | 5 GW |
| At ~$5-8K/kW, revenue capacity | $5-8B | $10-16B | $25-40B |
| Expansion cost per GW | $100-150M, 6-9 months | ||
| Manufacturing sites | Fremont, CA (cells) + Newark, DE (assembly) | ||
| Employees | 2,214 | ||
The capital-light expansion model is attractive. Each 1GW increment costs only $100-150M and takes 6-9 months. Compare this to a semiconductor fab ($20B+) or a nuclear plant ($10B+). Bloom can scale manufacturing relatively quickly if demand materializes.
| Year | Shares Outstanding (Basic) | Change |
|---|---|---|
| FY2021 | 173M | — |
| FY2022 | 186M | +7.2% |
| FY2023 | 213M | +14.5% |
| FY2024 | 227M | +6.7% |
| FY2025 | 240M | +5.7% |
| End of FY2025 | 280M | +16.7% from avg |
Shares have grown 62% in 4 years (173M to 280M). This is significant dilution, driven by convertible note conversions and stock-based compensation ($145M in FY2025 alone). If all outstanding convertibles convert, dilution could be even larger. This erodes per-share economics even as aggregate revenue grows.
Bloom is at 1GW annual capacity, expanding to 2GW by end 2026. Fremont can hold 5GW. Each 1GW ≈ $2B revenue. So:
| Current capacity (1GW) | ~$2B revenue ceiling |
| End 2026 (2GW) | ~$4B |
| Max Fremont (5GW) | ~$10B |
| Each expansion: 1GW | $100-150M capex, 6-9 months |
Supply is NOT a hard constraint — they can build more factories. But each expansion takes time and capital. The $8-12B bull case requires 4-6GW capacity, meaning at least one expansion beyond Fremont or a second facility. Achievable but not trivial.
Bloom's thesis depends on being the best option for data center power. But what are ALL the options?
| Power Source | Deploy Time | Cost/kWh | 24/7? | Scalable? | Status |
|---|---|---|---|---|---|
| Grid expansion | 4-5 years | Cheapest | Yes | Yes | Queue bottleneck NOW, but utilities are building fast |
| Bloom SOFC | ~90 days | Higher | Yes | Modular | Working, scaling. Uses natural gas. |
| Natural gas turbines | 1-2 years | Moderate | Yes | Yes | Proven, but OEMs supply constrained |
| Solar + storage | 6-12 months | Moderate | No (intermittent) | Land-intensive | 125x more land than Bloom per MW |
| SMRs (nuclear) | 5-10+ years | Unknown | Yes | Unknown | Not commercially ready until 2030+ |
| Diesel generators | Weeks | Very high | Yes | No | Backup only, not baseload |
| Geothermal | 2-3 years | Low | Yes | Location-limited | Only works in specific geologies |
The critical question: Bloom's advantage is "time to power" — 90 days vs years. But this advantage ERODES as the grid catches up. If utilities build enough capacity by 2029-2030, do data centers still need on-site fuel cells? Or does Bloom become unnecessary once the grid queue clears?
Counter-argument: Even if the grid catches up, on-site power may be permanently preferred for reliability (99.999% uptime), control (no utility dependency), and cost predictability (fixed fuel cost vs volatile grid prices). Bloom may transition from "grid can't keep up" to "on-site is just better."
This is sector-level knowledge that applies to evaluating ALL power companies — utilities (PNW, OGE, D, AEP), nuclear (CEG, OKLO), natural gas (EQT), and infrastructure plays (CORZ, APLD).
This is the deepest question and the one the template-based analysis completely missed.
Leopold is betting that AGI creates insatiable power demand that outpaces supply. But AGI itself could solve the supply problem:
Net assessment: On a 3-5 year horizon, Bloom is probably safe — physical infrastructure takes time regardless of how smart your AI is. You can't build a power plant in 90 days just because you have AGI. On a 10-year horizon, the risk increases. AGI-enabled breakthroughs in batteries, fusion, or grid optimization could undercut the thesis.
This is a fundamental tension in ALL AGI infrastructure plays: AGI creates demand for infrastructure but also creates the tools to eventually make that infrastructure unnecessary. The investment window is probably 3-7 years — long enough to be right, short enough to avoid being disrupted by your own thesis.
Nearly all of Bloom's $2.6B debt is convertible. This matters for valuation:
| Note | Principal | Conv. Price | Max New Shares | Status |
|---|---|---|---|---|
| 0% Notes (Nov 2030) | $2,500M | $194.97 | ~12.8M | OUT of money (stock $157 < $195) |
| 3.0% Green (Jun 2028) | ~$100M | $20.84 | ~4.8M | Deep IN money, mostly converted |
| 3.0% Green (Jun 2029) | ~$75M | $18.85 | ~4.0M | Deep IN money, mostly converted |
Key risk: The $2.5B 0% notes convert at $195. If the stock stays below $195, Bloom must REPAY $2.5B in cash by Nov 2030. With $2.45B cash today, they'd need to generate ~$50M+ in cumulative FCF over 5 years or raise more capital. Current FCF is $57M/year — barely enough. A bad year could force a dilutive equity raise.
If the stock goes ABOVE $195: Notes convert, ~12.8M new shares created (~4.6% dilution on 280M base). Manageable. The debt disappears from the balance sheet.
For our 10x math: At the bull case ($72-108B), fully diluted shares would be ~301M. Our 10x entry price should use diluted shares: $72B ÷ 301M ÷ 10 = ~$24/share. Consistent with our $25-35 range.
Instead of guessing "$8-12B revenue," let's build it up:
| Assumption | Value | Source/Reasoning |
|---|---|---|
| US data center power demand by 2031 | ~90 GW | IEA/Goldman estimates, ~3x current 30GW |
| % going on-site (not grid) | 20-30% | Industry surveys; grid can't keep up for all of it |
| On-site addressable market | 18-27 GW | 90GW × 20-30% |
| Bloom's market share of on-site | 15-25% | Dominant in SOFC but turbines/engines also compete |
| GW Bloom installs cumulative by 2031 | 3-7 GW | 18-27 GW × 15-25% |
| Revenue per GW installed | ~$2B | Current pricing, ~$2/watt |
| Annual revenue by 2031 (new installs + service) | $6-14B | 3-7 GW/year × $2B + growing service revenue |
| Gross margin at scale | 35-40% | Currently 29%, improving with scale |
| Net margin at scale | 15-25% | Operating leverage from fixed cost base |
| Net income at scale | $1-3.5B | $6-14B × 15-25% |
| P/E at scale (growth + moat) | 25-35x | High-growth infrastructure with moat |
| Bull case market cap | $25-120B | Wide range reflects high uncertainty |
| 10x entry market cap | $2.5-12B | $9-43/share at 280M diluted shares |
Honest assessment: The range is very wide ($25-120B) because the key variables (on-site %, Bloom's share, margins) are all uncertain. The midpoint is ~$60-70B, giving a 10x entry of ~$6-7B (~$22/share). This is consistent with our $25-35 target but with better reasoning behind it.
The most sensitive variable is on-site %. If on-site goes to 30%+ (because grid truly can't keep up), bull case is $100B+. If grid catches up and on-site stays at 10%, bull case is only $25B and our 10x entry is $2.5B (~$9/share).
| 10x Target Entry | $25-35/share ($7-10B mkt cap) |
| 5x Target Entry | $50-65/share ($14-18B mkt cap) |
| "Decent Return" Entry | $80-100/share ($22-28B mkt cap) |
| Current Price | ~$157/share ($44B mkt cap) |
Bottom line: This is a genuine "wonderful company at a not-wonderful price" situation. The business is excellent, the thesis is validated by the most AGI-bullish investor in the market (Leopold), and the secular tailwind is massive. But at $44B, the market has already priced in enormous growth. The stock already delivered its 10x. For us to get 10x, we need the stock to drop 80%+ to the $25-35 range. That could happen in a broad market crash, a recession, or if a quarter or two of execution stumbles break the momentum narrative. Until then, this goes on the watchlist.
Leopold is clearly right about the thesis. The question is whether he's right about the price. He likely bought much cheaper. Following him in at $157 is not the same trade as following him in at $30-50.
Data sources: SEC EDGAR XBRL (CIK 1664703), yfinance, 10-K filing (FY2025 filed 2026-02-09), Situational Awareness LP 13F (Q4 2025), bloomenergy.com, stockanalysis.com, Goldman Sachs research, IEA. Analysis date: 2026-03-12.