Bloom Energy

IThe business

Bloom Energy builds fuel cells that make electricity from natural gas without burning it. Its customers are data centers, factories, utilities and hospitals that want power on site, now, instead of waiting for the grid. Dr. KR Sridhar founded the company in 2001. It lists on the New York Stock Exchange, employs about 2,200 people and has about 1,100 sites running in nine countries.⁵ Its October release notes that it has been added to the S&P 500.²

Almost everything it sells is one product, the Energy Server. Inside are thousands of small ceramic fuel cells that run at very high temperature and turn gas and air into direct-current power. There is no flame and no spinning machine.⁵ The company says systems installed after 2020 have run at about 99.9% availability, that in some projects power can flow about 90 days in, and that it responds to a sudden jump in load at least twice as fast as a turbine or engine.⁵ Those are the company’s claims, and I cannot test them. Two parts are checkable in its accounts: what it sells, and what it earns.

In the third quarter, product, meaning the Energy Server itself, was $1,073.6M of $1,235.8M in revenue, 87%.³ Installation was $84.9M, service $67.6M and electricity from machines Bloom still owns $9.8M. The product line earned a 37.9% gross margin. Installation lost money, at −4.8%, and service earned 17.8%.² The company is pulling out of installation as it grows. Large sites now use a certified outside contractor and Bloom acts as a consultant.³

Where the growth came from

The jump is recent. Revenue was $1.47B in 2024, $2.02B in 2025 and $3.05B in the first nine months of 2026 alone.⁵³ Management started the year guiding $3.1B to $3.3B for 2026. It has raised the figure three times: to $3.4B to $3.8B in April, $3.9B to $4.2B in July, and $4.4B to $4.6B on 22 October. Non-GAAP earnings per share guidance went from $1.33 to $1.48 to $3.10 to $3.40 in the same year.⁸⁶⁷² The midpoint is now 41% above where the year started. That is a business learning how fast it can ship. It is also one whose own forecasts have been too low at every update.

Three kinds of customer drive this. The first is data-center builders. Oracle announced a partnership in October 2025 and received a warrant for 3,531,073 shares at $113.28.³ Brookfield Asset Management agreed in August 2025 to a financing framework of up to $5.0B over five years for Bloom projects, held in an AI infrastructure fund.⁵ American Electric Power signed in 2024 for up to 1 GW, starting with 100 MW.⁵ The second is utilities and independent power producers. The third is the original base: businesses buying one or a few units, plus SK ecoplant in Korea, where about 682 MW is installed.⁵ In the third quarter 97% of revenue was American, against 72% in the first nine months of 2025.³

Capacity and what is not disclosed

The 10-K says the Fremont factory runs at a rate of 1 GW a year and is being doubled to 2 GW by the end of 2026.⁵ A gigawatt is a million kilowatts, about what a large nuclear reactor makes. Capital spending is rising to match: $231.3M in nine months against $33.8M a year earlier.³ Management told investors in July that it could support 25 GW of fuel cells a year of material supply, and would keep expanding capacity.¹⁴ That is a statement about supply, not about orders or about revenue per gigawatt, and the company does not publish revenue per gigawatt shipped. I do not use the 25 GW figure in my cases.

What the company does disclose about the future is thin and comes in two measures that are very far apart. At year end 2025 it showed a backlog of about $6.0B for products and $14.0B for service, a total near $20B.¹⁵ The accounting figure, called unsatisfied performance obligations, was $394.4M at the same date and $661.2M at 30 September 2026, with another $56.2M in service contracts.³ The gap is a difference of definition. Bloom’s backlog counts contracts it considers committed, including expected tax-credit value, and service contracts of 5 to 20 years that customers can end each year.⁵ The accounting figure counts only what is firm. I do not call either number wrong. I think the second is the one to weigh when I ask what will certainly be delivered, and the first when I ask what the company hopes to deliver.

IIThe moat

Two things protect a business: something rivals cannot copy, and customers who find it costly to leave. Bloom has a good case for the first. It has a weak case for the second.

What is hard to copy

Bloom’s fuel cell stack is its own design, and the 10-K describes two decades of work on materials, printing of the cells and the system around them.⁵ Other fuel cell types exist, and the 10-K runs through three of them, saying each has a weak point: platinum cost and poisoning in PEM cells, short life in molten carbonate, and heat-matching trouble in phosphoric acid.⁵ It names no other maker of large solid-oxide power plants. Time is a protection too. The company points to a track record across 1,100 sites.⁵ I give that real weight. A buyer taking a risk on a new power source at a data center will want a supplier with a long record in the field.

What is not protected

Customers do not stay by contract once they have bought. A purchase is a one-time sale of equipment, 99% of revenue comes from direct purchase, and service contracts can be ended by the customer each year.³⁵ Service, at 17.8% gross margin and 5% of third-quarter revenue, is too small to anchor anything today.² The stack wears out and is replaced, so the long-term income from a fleet depends on the company beating its own warranty estimates. It says plainly that if its estimates of useful life are wrong, results could be hurt.⁵ It has guaranteed output on its contracts to a cap of $847.2M, with $466.1M still possible to pay out, and it paid $21.3M against those guarantees in nine months.³

The short-seller report

On 8 July 2026 a report by Hunterbrook Media, an outlet that said it might profit from a fall in the stock, challenged Bloom’s supply of a raw material, scandium oxide, and its accounting.¹⁴ Bloom called the claims false and misleading. It said it has enough scandium for current demand and backlog and is not dependent on China for it.¹⁴ The stock fell from $269.57 on 7 July to $254.29 the next day, and finished July at $205.81, down 32% on the month.¹ A class action followed on behalf of buyers since 27 February 2025, alleging misleading statements about the supply chain. A lead plaintiff was named on 15 October, and Bloom says the case is without merit.³ I cannot judge the raw-material question. It matters because the whole expansion rests on stack materials the company says it can get. If the company is right, the report cost shareholders a good deal of value for nothing. If it is wrong, the factory ceiling in my cases is too high. I leave the base case where the company’s own disclosures put it and name this as an open item.

Judgement

A real technology advantage, in a market short of power, sold with a thin protection after the sale. I would pay for the first. I do not pay for the second.

IIIThe competitive advantage

The competition for a data center’s power is mostly not other fuel cells. It is gas turbines, gas engines, the grid, and solar with batteries.⁵ Bloom’s case is that on time to power and on air permits it wins, and that it can follow the swings in an AI data hall better than a turbine can.⁵

Two facts from the accounts say the advantage is real today. First, the gross margin on product rose from 35.0% to 37.9% over a year in which product revenue nearly tripled.² A seller without an edge usually cuts price to win volume. Second, the customers now include the largest buyers of data-center power. Oracle is in Bloom’s accounts as the customer’s customer, and it received a warrant on Bloom stock as part of its arrangement.³ I cannot say why it wanted one. The warrant tells the other side of the story. Bloom paid the customer’s customer in shares: 1,905,433 shares plus 248,798 extra, a fair value of $323.9M, booked as a reduction of revenue as units are delivered.³ A seller with pricing power does not usually need to do that.

Where Bloom wins

Speed. Customers want power sooner than the grid or the large turbine makers can give it, and the 10-K says those makers are supply constrained.⁵ Air quality. Bloom’s units make almost no nitrogen oxide, and the 10-Q says that even so the air-permit process for larger installations is taking longer than expected.³ Load following. A data hall’s power draw can change in milliseconds, and the company claims its fuel cell paired with supercapacitors handles this better than engines.⁵

Where it is exposed

Cost. Bloom publishes no price or cost per kilowatt, so I cannot show the customer’s economics. Tax credit. The 30% investment credit that most customers use for fuel cells starts for projects beginning construction after 2025 and phases out after 2033.⁵ Part of the demand is bought by a tax benefit. Finance. In the 10-Q, the company says it arranges financing for its customers and that data-center financing has become harder.³ Local opposition. New York State’s one-year pause on new large data-center permits, signed in July 2026, is one of several moratoriums, and the company says that even fuel cell projects now face more hearings and delays.³

IVManagement

Management here means one founder and a team that changed a lot in the last year. The founder is Dr. KR Sridhar, who started the company in 2001 and is its chairman and chief executive.¹⁶² He and the board have led it through the long losses, the 2018 IPO, the pivot to data centers, and a 2025 in which the stock rose 291%.⁹

How they are paid, and what that pays them to do

Dr. Sridhar’s pay is mostly stock tied to growth. In 2024 the board gave him performance units valued at $42.4M at grant, making his 2024 pay $45.0M against $3.5M in 2025.⁹ Those units earn on product revenue growth and non-GAAP product gross margin, year by year to 2027. The proxy says that for 2025 the financial targets for one-third of the award were met at 300% of target.⁹ A second award came on 15 June 2026: performance units for 319,082 shares at target, up to 300% of that, which vest on the best four consecutive quarters of total revenue from July 2026 to December 2029, adjusted up or down by 2029 product gross margin. He must keep the net shares until the end of 2031.¹⁶¹¹

The incentive is plain. Revenue is what pays, and the margin test is one lever on top of it. That is a fine design for a company that must scale fast. It also pays for growth that eats cash: nothing in the measures I can see is tied to free cash flow, return on capital or the share count. The 2025 cash bonus for executives was half non-GAAP operating income and half total revenue.⁹ So the people who run the company are paid on two lines that the working-capital build in the cash flow statement does not touch.

How much of their own money is in

Dr. Sridhar’s stake of 5.66M shares is worth about $1,150M at $203.17, and 1.70M of those shares are options.⁹ Directors and officers together own 8.34M shares, 3.0%.⁹ That is a real stake. The record since last November is lopsided: 11 insiders made 56 open-market sales worth $165.2M, and I found no open-market purchase.¹⁰ Selling after a nine-fold rise is not a signal on its own. The Form 4 data does not show whether the sales were planned in advance, and I do not read motive into them. Mary Bush, a director, sold $11.8M, and John Chambers, another director, sold $23.2M.¹⁰

What they did with capital

Capital, until 2025, was something Bloom consumed. From 2016 to 2025 the company lost a cumulative $1.36B at the operating line and burned $0.82B of cash from operations, paid for by equity and by convertible notes.⁴ The accumulated deficit was $3.99B at the end of 2025.² The weighted share count went from 115M in 2019 to 240M in 2025, and the cover of the October 2026 10-Q shows 295.7M.⁴³ In November 2025 it sold $2.5B of 0% convertible notes due 2030, which holders can convert at $194.97 a share, and with the stock at $203.17 they sit just above the line.³ It has drawn nothing on its revolving credit facility, and $510M is available after $90M of letters of credit.³

The record on the way up has been better. Capital spending is rising to build the factory, the company redeemed or saw converted most of its 3% green notes.³ The Oracle warrant and the Brookfield funds are creative, not cheap. The first cost $323.9M in shares, and in the second, Bloom keeps 6.2% of two funds that buy its own machines.³ I cannot call that bad capital allocation. I can say it has been tested for one good year, and the cycle has not turned.

VFinancials

The table has five years and the last twelve months. All of it is from filings; the last column adds nine months of 2026 to the end of 2025 and removes the same nine months of 2025.

Look at the rows for stock pay. It was $145.0M in 2025 and $173.5M in nine months of 2026, and the Q3 charge of $60.1M is 4.9% of revenue.³² Non-GAAP profit leaves this out. I do not. A share paid to an employee costs me the same as a share sold, so I count stock pay as a real cost, and I do that in my model.

The lines of the business

Nearly all the profit comes from the product line. Installation is a loss, service is a small, low-margin income, and electricity is a run-off business from earlier owned machines. The 10-Q says revenue grew because acceptances of new units and deliveries on existing contracts rose, and it ties the growth to demand for time to power.³

Who pays

In the third quarter two customers made up 43% and 26% of revenue.³ The 10-Q says “customer” can mean a project-finance affiliate instead of the end user.³ The 26% customer is a related party. Revenue from related parties was $321.6M in the quarter and $697.7M in nine months, from the Brookfield joint ventures in which Bloom holds 6.2%, and Bloom eliminates part of its profit on those sales.³ Receivables were just as concentrated: three customers were 31%, 24% and 15% of the balance.³ The 10-Q does not name the 43% customer. Bloom has said no more, and I will not guess.

Cash is not earnings

This is the most important table on the page. Net income for nine months was $521.5M. Cash from operations was $272.9M. Capital spending took $231.3M, leaving $41.6M. Stock pay of $157.6M, a real cost, makes owner cash minus $116.0M.³ Within that, customers paid $610.6M more in deposits than they used, and without it operating cash flow would have been negative $337.7M.³

The reasons are in the balance sheet. Contract assets, which are work done and revenue booked but not yet billed, rose from $241M to $859M in nine months.³ In the third quarter alone they rose $430M.² Inventory rose from $643M to $928M.² Receivables rose from $372M to $635M. Other long-term assets rose $275M in the cash flow statement, and the 10-Q does not explain it.³ On the other side, suppliers financed $285M, and accrued items $248M.³ Management says the contract-asset build is the timing of milestone billings and customer acceptance cycles.³ That is a fair explanation and may reverse. It is also the nature of selling large power plants to financed projects: you ship first and wait. The revenue is real. The cash comes later, and more is needed for each extra dollar of growth.

My model starts from that fact. It charges stock pay in full, takes capital spending as a percent of revenue, and builds working capital at 16% to 22% of revenue in the base. Even in a year where profit is high, the model shows modest owner cash: in 2027 my base has operating profit before stock pay of $1.65B and owner cash of $0.45B.

The balance sheet

At 30 September 2026 Bloom had $2,486.7M of cash, almost all in money-market funds, and recourse debt of $2,522.8M of face value: $2,500M of 0% convertible notes due November 2030 and $22.8M of 3% notes due 2029.³ Financing obligations, from sale-leaseback deals that failed accounting, were $185.0M.² Net of that, the company has about $0.2B of net debt on paper, $2.5B of which is a convertible that turns into about 12.8M shares if the stock stays above $194.97. The company can redeem the notes from November 2028 if the shares trade above 130% of that price, $253.46.³ It also has take-or-pay and minimum-purchase commitments with suppliers of $5.8B over more than 12 months: $2.6B in 2027, $1.3B in 2028 and $1.8B in 2029.³ Those commitments are the plan for growth, and they are fixed in advance of the orders that must pay for them.

The long memory

Bloom reported an operating loss in every year from 2016 to 2023, and the cash flow from operations was negative in seven of ten years.⁴ Its best year before 2024 had an operating loss of $81M. The market now prices a different company from the one that filed those numbers: a decade of failing to turn a fuel cell business into earnings, then two years of data-center demand. I give the last two years their full weight. I do not give the first ten none.

VIValuation

First, how this goes badly wrong

Before any number, I ask what would most likely hurt a buyer here. There are three ways, and they are not the same thing.

The money for the AI buildout dries up, and the customers are few. Two customers were 69% of third-quarter revenue.³ The 10-Q itself says financing for data centers has become harder, that hyperscalers are borrowing more, that neoclouds and co-location tenants are viewed as high credit risks, and that air permits and moratoriums are slowing projects.³ Bloom generally cannot close a sale to a financier until the permit is in hand.³ If two or three buyers pause, a company built for a 2 GW factory would have the capacity and the orders would be missing. My bear case has revenue peaking at $6.2B in 2028 and falling to $4.4B by 2031.

Growth eats the cash, and the commitments are fixed. The company has $5.8B of purchase commitments with suppliers, $2.6B of it in 2027, and it is building working capital faster than it earns cash.³ A slowdown would leave inventory on hand and take-or-pay bills. The 0% notes of $2.5B come due in November 2030 and convert into shares only if the stock stays above $194.97.³ In a bad outcome the company could need to raise equity at a low price, which my per-share values do not capture.

The product or its supply falls short of the claims. Output and life guarantees run to a cap of $847.2M.³ The raw-material question in the short-seller report is unresolved and the class action is open.³ If a key input is scarce, the factory cannot reach the size the plan needs. I cannot test any of that.

What an owner can take out

Reported profit is a poor guide here. In the first nine months of 2026 Bloom reported $521.5M of net income and had owner cash of minus $116.0M after capital spending and stock pay.³ My model counts only cash: non-GAAP operating profit, less stock pay, less tax after using up $2.3B of federal losses, plus depreciation, less capital spending, less the build in working capital.⁵ In my base the company earns $3.0B of operating profit before stock pay in 2030 and I count $1.4B of owner cash from it, 47%. Even in 2036, when growth has slowed, the conversion is 53%.

What the price implies

At $203.17 the shares are worth about $63.8B counting 314M diluted shares: 295.7M outstanding, 4.6M net from options, the 12.8M the 0% notes would become because the price is above the $194.97 conversion price, and 1.1M from the 2029 notes. Cash less other obligations leaves an enterprise value of about $61.6B. Stock awards not yet vested are not counted, because their cost is already in my stock pay line and charging it twice would be wrong.

On analysts’ non-GAAP estimates that is 41 times 2027 earnings of $5.00 and 27 times 2028 earnings of $7.42, with only two estimates for 2028 and the 2026 estimate of $2.61 below the company’s $3.10 to $3.40 guide, so these have not all caught up.¹⁷ My base 2027 matches the consensus: it makes about the same earnings. The disagreement is about after 2028.

To justify $203.17 on my cash conversion, Bloom would need about 5 times my base revenue by 2030, around $61B a year. I cannot point to a hardware company that grew from $4.5B to that size in four years. The price can be read another way: the market may use a lower rate and a longer horizon than I do. At 10.0% my bull is $219, just above the price. Either way, a buyer at this price is paying for my bull case or better.

What I can’t judge

Four inputs sit outside my reach. The first is how long orders last: the only backlog figures are the company’s own and a GAAP figure that is much smaller. The second is whether the scandium and other materials are there for 2 GW and then more. The third is stack life, which sets service cost and warranty for a fleet only a few years old at this size. The fourth is who the 43% customer is and how firm its plans are. I show a wide range instead of a precise number.

Nothing in the grid reaches $203.17. The top right cell, three times my base revenue with margins 4 points higher, is $161. Read this table as a statement about how large the gap is, not as a forecast of any cell.

The discount rate

I use 12.5%. The basis is the 10-year Treasury yield of 4.94% on 30 October, plus a beta times a 4.20% equity premium.¹⁸¹⁹ I measured Bloom’s weekly beta against the S&P 500 at 2.22 over five years, 2.63 over three, and 2.90 over two.¹²⁰ A raw 2.22 would give 14.3%. A common adjustment pulls beta toward 1, to 1.82, and gives 12.6%, which I round to 12.5%. This is high next to the 10% I use on the large companies on this site. It reflects one year of profits and two customers that are most of the revenue.

Three cases

I weight the cases 30%, 50% and 20%, for a value of about $54. Those weights are my guesses. They are not used in any value above. My base reaches $12.2B of revenue in 2030, 2.7 times the 2026 guide, and has operating profit before stock pay of $3.0B that year. That is already a bet that the order flow is not a spike. My bull has the company at $54B in 2036, larger than GE Vernova is today, and margins that I think stretch the product’s economics.

Other uses of the same dollar

Treasuries pay 4.94% for ten years.¹⁸ On my base, Bloom returns −6.3% a year from today’s price on my cash flows. Even my bull case returns 8.6%, below the 12.5% I require, because the price is above it. The nearest comparable on this site is GE Vernova, which sells the turbines Bloom competes with: it returns 5.9% a year on my base at its page’s 10.0% rate, and its price is 34% above its base value. Vertiv, which supplies cooling and power gear to the same data halls, returns 3.8% at a 12.0% rate. Neither is a bargain on my numbers. Both are cheaper per unit of value than Bloom.

The margin of safety

My margin-of-safety price is $6. I set it so that my bear value of $5 is a quarter below it, which leaves a buyer room to be wrong about the cases. A price that low means no margin exists anywhere near $203.17. The distance from $203.17 to $6 is not a forecast. It says how much of the price is a hope.

Two different questions

How good is the business, and how good is the price? On the business: it has a real product, very fast growth and an improving margin, with a concentrated customer base and a short record. On the price: it is above all three of my cases. A strong business at a price that needs a better outcome than the strong business is likely to deliver can be a poor purchase. That is what this looks like to me. It is my reading of these filings, not a rating.

Where this could be in 5 and 10 years

For the base to be right at five years, Bloom has to reach about $12B of revenue by 2030 at a 25% margin before stock pay. That is a business 2.7 times the size of the 2026 guide. At ten years, it has to keep those margins as growth fades to about 3%, with competition from turbines and engines. The numbers that drive this are growth, margin, the cash tied up in working capital, and the 12.5% rate. The most likely way the base is wrong is that revenue is simply lower: I cannot see orders for 2028 and 2029, and two customers make up most of the revenue I can see.

The bear and bull differ in one thing, how long the buildout lasts. In the bear it stops in 2028, and the value is close to today’s book value of about $6.50 a share. In the bull it continues to 2036 and Bloom is a $54B business. The base value in year 10 is $74 per share, with $42 of owner cash on top before then. These are estimates of value, not forecasts of the share price, and the price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three facts cut in the company’s favour and are not in the headlines. First, tax. Bloom had $2.3B of federal operating losses at the end of 2025, and it booked an income tax charge of $0.8M on $249.8M of third-quarter profit.⁵² A good share of the next few years of profit can reach the company without a federal tax bill. Second, price. Product gross margin rose from 35.0% to 37.9% in a year in which product revenue nearly tripled.² Third, customer money. Customer deposits rose from $78M to $661M in nine months, so buyers are putting cash down.³

The bull case

In plain words: power is the limit on AI, Bloom can ship it in months, and the company is only starting. Management says the move to firm, on-site power has spread from the hyperscalers to every business that needs reliable power to grow, and CFO Simon Edwards cited demand when he raised guidance for the third time this year.² Brookfield’s framework of up to $5.0B over five years exists to finance projects for Bloom’s customers.⁵ The Street agrees on direction: the mean price target on Nasdaq is $295.16, with a range of $182 to $351, with 10 positive ratings, 10 neutral and none negative.²¹ Consensus earnings are $5.00 for 2027 and $7.42 for 2028.¹⁷

The holder list shows who else is in. Situational Awareness LP, the fund of Leopold Aschenbrenner, held 6,272,808 Bloom shares at 30 June 2026, worth $1,898,778,982, along with a call option on 145,800 shares, per its 13F-HR filed on 14 August 2026.²² That is a dated fact from a filing that does not show shorts or non-US holdings, and it says nothing about what the fund holds now or will do. Ameriprise, BlackRock and Vanguard each owned more than 5% at the proxy date.⁹

Against reality

Today’s numbers support the near part. Revenue, margin and earnings have beaten the plan four times in a row, and nine months of profit were $521.5M.²³ If Bloom delivers about $7B of revenue in 2027 and $12B in 2030, at a 25% margin before stock pay, then that is my base, and it supports $44 a share. That is not close to $203.17. The distance is the whole argument: the price needs about 5 times my 2030 revenue at the same margin, or about $61B.

It helps to multiply the steps out. Say orders keep growing through 2028, 70%. The company shows enough scandium and other materials for several times today’s output, 60%. Stack life holds so that warranty costs stay low, 70%. Customers broaden so that no one buyer is over 30%, 60%. And the margin holds above 25% at ten times today’s volume against turbines and engines, 50%. Those are my guesses and not independent, since strong demand helps most of them. Multiplied, they give 9% E (estimate). Allowing for the overlap, perhaps 15%, which is below the 20% I give the bull. Each step looks reasonable. All five together are a long way from certain.

What a 10x would take. Ten times $203.17 is $2,032 a share, about $638B of market value. At 25 times earnings that needs about $26B of net income. At a net margin of 22%, which is above anything Bloom has reported, it needs about $116B of revenue, some 26 times the 2026 guide and more than twice my bull case for 2036. To get there by 2036 the stock would have to compound at 26% a year for ten years. Bloom has already done something like it: the shares are up 2.3 times since the end of 2025, from $86.89.¹ But that started from a company worth about $25B that had just reported a loss, and the next tenfold begins from $64B. A maker of fuel cell plants would do it with factories, long lead times and customers who finance projects. It is close to impossible, and I would not build a case on it.

Where I come down

The tax shield and the rising margin moved me: without them my base would be lower. The headline backlog and the guidance raises did not, because I cannot see what is firm and because the raises have not yet shown up as cash. Evidence that free cash flow after stock pay turns solidly positive while revenue still grows would raise my values most, and a second and third large buyer that each take more than a fifth of revenue would raise them next.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Customer concentration. Two buyers were 69% of third-quarter revenue and one is a related party.³ Losing or delaying one changes a year.
• Financing and permits. Customers need project finance, air permits and, in many places, public approval. The company says all three are getting harder.³
• Tax credit. Much of the customer economics depends on the 30% investment credit, which phases out after 2033.⁵ A rule change would reach demand quickly.
• Materials and quality. The raw-material question is open, and the output guarantees carry $466.1M of possible payments.³¹⁴
• Accounting complexity. Revenue to the Brookfield funds is partly eliminated, a $306.5M asset for the Oracle warrant is amortized against revenue, and 26% of receivables sit with a related party.³ I have read these notes and find them disclosed, but they take work to follow.
• Dilution and debt. Shares rose from 280.0M to 295.5M in nine months.² The $2.5B 0% notes become shares above $194.97 and a cash claim in 2030 below it.³
• Litigation. A securities class action covers statements since 27 February 2025. The outcome is unknown.³
• Competition. Turbine makers, engine makers and other fuel cell firms are all adding capacity into the same shortage.⁵ Prices that look good now invite them.
• The stock itself. It fell 41% from its June high to now, and 32% in July alone.¹ Even if I am right on value, the path can be violent both ways.

Geopolitical risk, and what it means for the value

Bloom sells to America: 97% of third-quarter revenue and 93% of the nine months.³ The geopolitics sits upstream, in what goes into the machine, and in the law that helps pay for it. For each case below I rerun my base. The odds are my guesses, with the basis stated.

Tariffs. The 10-Q lists tariffs of 10% to 12.5% on about 60 economies since late July, a separate regime on steel, aluminum and copper, and a February Supreme Court ruling that some earlier tariffs were unlawful.³ The 10-K expects the 50% metals tariffs to raise costs, yet says tariffs did not move 2025 gross margin materially.⁵ Bloom also booked $37.2M of refunds in nine months.³ The current tariffs are in my margins. The risk is a further step.

Chinese inputs and Taiwan. China put scandium under export licences on 4 April 2025, and that rule stands. A wider October 2025 package is suspended to 10 November 2026.²³²⁴ Bloom says its scandium oxide supply covers its backlog, does not depend on China, and can support 25 GW a year.¹⁴ The 10-K says its second and third tier suppliers use Chinese rare earth compounds, expects no effect on 2026 output, and says its parts come principally from Taiwan.⁵ The short seller disputed the scandium claim and I cannot settle it.

The credit, Korea, gas and the grid. The July 2025 law set a 30% credit for fuel cells that begin construction after 2025 and before 2033, and bars owners controlled by China, Russia, Iran or North Korea.⁵ Korea is the second market, about 682 MW, sold through SK ecoplant, which holds 2.5% of the shares and a joint venture that assembles units there.⁵ It matters little to value. Bloom runs on pipeline gas, so gas supply and interconnection rules matter to customers; the 10-K notes FERC is reviewing rules for on-site generation.⁵

The price of $203.17 is above my bull case of $152, so it pays for none of the downside rows, and even the best row, $50, is a quarter of it. Only the current tariffs are in my numbers. The scandium and Taiwan rows are what I cannot see into.