Vistra alone signed 2,609 megawatts with big tech buyers.
Is it a good business? · 6 · What does its record show?
Losses turned to solid profit
Sales, $B
1.0
1.2
1.3
1.5
2.0
3.8
Dec 2021Dec 2022Dec 2023Dec 2024Dec 2025last 12 mo
Sales$3.83Brevenue in the past year
Profit−22% → 15%operating profit per $1
Per $112% → 32%kept per $1 after making it
The gain rests on what each machine keeps; in 2022 that fell and losses deepened.
Is it a good business? · 7 · How much room is there to grow?
Demand is measured; supply may be the limit
22% to 33%of commercial power to servers by 2050
About 40 tonsof world scandium output in 2024
About 220 tonsa short seller says Bloom's goal needs
Bloom has not said how much scandium, a rare metal in its cells, it uses, so the two cannot be compared.
Is it a good business? · 8 · Who is in charge?
The founder took it to profit
Since 2002KR Sridhar has run Bloom
About 3%of the shares held by insiders
April 2026Simon Edwards finance chief
The founder spent two decades building the technology before the company turned a lasting profit.
The story right now · 9 · Why do data centres want Bloom?
AI data centres cannot wait for the grid
138%Bloom's sales growth last quarter
317%the shares' rise
New AI data centres need power faster than utilities can connect them, and a grid link can take years. Power made on site skips the queue.
The story right now · 10 · Why is cash lagging the profit?
Sales are counted before the cash arrives
A machine is deliveredSales are counted.
Billing comes laterCash waits.
Two buyers dominate69% of the quarter.
One is a related partyNot fully arm's length.
Sales counted is not cash collected. When few buyers take most sales, the gap matters more.
The story right now · 11 · What changed lately?
Sales rose 138% and the forecast went up
$1,235.8Msales in the quarter
$225.4Moperating profit, from $7.8M
$4.4B to $4.6Bthe raised yearly forecast
−$27Mcash from operations
Mixed quarter
Billing lagged sales, and one of the two big customers is a related party.
The story right now · 12 · What is everyone arguing about?
A lasting turn, or two buyers' orders?
Many investors think
The cleanest way to own on-site AI power
We think
The profit is real, resting on one unnamed buyer
Against us: the shares are up 317% and seventeen analysts rate it a buy.
The price and our view · 13 · What is a share worth?
The price is nearly twice our value
$74.86what we think a share is worth · the last close is of that
$74.86our value
Attractive below $59.89
Unattractive above $89.83
$31.27gloomy: growth halves and less kept per $1
$74.86our estimate: experts' path, then slowing
$148sunny: growth beats forecasts by a quarter
Value per share in three futures. None is a prediction of the share price.
We estimate what a share is worth from the cash the business should earn over many years. Our rule: buy only when the price is at least a fifth below our estimate, as a cushion for being wrong.
The price and our view · 14 · What if it all works?
As a bet on on-site power, it is fragile
Priced for successthe kind of bet this is
0.8×if it works
4%kept if it fails
Multiples of the company's value when the bet was last struck, 30 Sep 2026.
Worth less than today even if it works, with about a tenth left if sales fall back.
The price and our view · 15 · How could we be wrong?
Two buyers, slow cash and a fraud claim
Two big customersBiggest
69% of the quarter's sales.
Cash lags the salesShowing now
Operating cash went negative.
A fraud allegationWatch
A class action is open.
Where scandium comes fromUnknown
No supplier is named.
Future orders are counted on Bloom's own definition, and its biggest named site is late.
The price and our view · 16 · How will we know?
Early 2027 shows who is buying
Cash from operations turning positiveGood sign
New buyers named beyond the twoGood sign
The biggest site slipping furtherBad sign
FEB 20274
Fourth-quarter 2026 results, 4 February. Companies report every three months; each report is a test of the story.
Our view also changes on price alone: if the shares fall to within a fifth of our estimate, we move to Neutral.
The price and our view · 17 · So, what do we think?
A real profit turn resting on two buyers
$74.86our value
Attractive below $59.89
Unattractive above $89.83
UnattractiveMedium conviction
Bloom's fuel cells give data centres power on site, and its latest quarters turned to profit. But two customers took about two thirds of sales, and the price sits far above even our sunny future.
↑ ↓ to move · C for contents
What it does
More about the company
Who it is
An AI data centre can wait years for a grid connection. Bloom sells it a way around the wait: a fuel cell system on the customer's own site that turns natural gas, biogas or hydrogen into electricity without burning it and with no moving parts.
The buyer's other choices are gas engines, small gas turbines, or solar panels with batteries. About 1,100 sites in nine countries run on Bloom systems.
How it makes money
Bloom often sells the system not to the user but to a financier who owns it and sells the power, because most users would rather pay only for the electricity they use. Service adds the rest. In the June quarter, products were $935.4M of the $1.07B, and Bloom kept 33.4 cents of each sales dollar after the cost of making it.
Growth and profit
What growth costs
Growth is making Bloom more profitable: research, selling and office costs grow much more slowly than sales, so each new sale costs less to win. The catch is that this rests on the margin on each server holding up; when that slipped in 2022, losses deepened.
Its record
The numbers
Is it growing?Fast
Sales grew 111% over the twelve months to Sep 2026, faster than the 37% in the year to Dec 2025.
Is it profitable?Healthy
Operating margin is 15%, up from −16% in the year to Dec 2023.
Does growth pay off?Pays off
Running costs take 18¢ of each sales dollar, down from 32¢ in the year to Dec 2021.
Does it earn good returns?Strong
Shareholders' money earned 40% a year, up from −11× in the year to Dec 2021.
Is it making cash?Solid
It turned 11% of sales into free cash, $437M over the last twelve months.
Is the balance sheet safe?Net cash
It holds $2.49B of cash against $2.48B of debt, $7.25M more cash than it owes.
Are shares being diluted?Diluting
The share count rose 87% in 4.8 years.
Why it wins
The evidence
Bloom designs and prints its own cells and builds the stacks, so the chemistry, the machine and the factory are one platform it owns rather than a kit of bought parts.
From outside, that shows up as more than a decade of working installations at about 1,100 sites: the reference list a buyer of critical power checks before signing.
Who could beat it
The rivals in full
The larger threat is a buyer choosing not to generate on site at all. Constellation and Vistra have signed 20-year contracts to supply Microsoft and Meta from existing power plants, 2,609 megawatts in Vistra's case alone.
Each rival
Caterpillar and Cummins, the listed makers of natural-gas reciprocating gensets that Bloom's own Item 1 names as its first competitive classCATCMI
The whole on-site generation purchase: a buyer installs engine gensets instead of fuel cells, accepting higher emissions and maintenance downtime for equipment that is cheaper per kilowatt and available from an established service network.
Constellation Energy and Vistra, selling long-term co-located and grid-delivered power purchase agreements off existing nuclear and gas fleetsCEGVST
The decision to generate on site at all - a buyer signs a decades-long contract for firm carbon-free power and never becomes a power-plant owner, permit-holder or fuel buyer.
GE Vernova and Mitsubishi Power, in the small-gas-turbine class Bloom's Item 1 names secondGEV7011.T
On-site firm generation, once turbine order books free up - Bloom's own filing concedes these makers are currently supply constrained, which is the window it is selling into rather than a permanent advantage.
How big it could get
The market and who has signed
The market
Demand is measured: US government projections have data-centre servers rising from about 7% of commercial-building electricity in 2025 to 22% to 33% by 2050.
Supply may be the limit. The cells use scandium, and the world produced about 40 tons of it in 2024, mostly in China. A short seller estimates Bloom's five-gigawatt goal would need about 220 tons a year. Bloom has not said how much it uses, so demand and supply never meet in one number.
Who has signed
The firm buyers the filings show are customers Bloom does not name. One bought about 73% of first-half 2026 sales; in the September quarter two bought 43% and 26%, the second a related party. A year earlier, no single customer was more than 30% of a quarter's sales.
Who runs it
Leaders, pay and the cash
Who runs it
KR Sridhar founded Bloom and has run it since 2002, taking it from years of losses to its first operating profit in 2024. A new finance chief, Simon Edwards, joined in April 2026. Insiders own about 3%, two-thirds of it Sridhar's. His bonus rides on sales and adjusted operating profit, his shares on product sales growth, product margin and winning large AI data-centre deals; none of it on cash or on profit per share.
What they did with the money
Bloom is not short of money, with $2.67B of cash at June. But the share count has risen about 25% since March 2025, so each share owns a smaller slice of the turn.
The last report
The latest report
Bloom's sales rose 138% to $1.24B and its forecast went up again
What it did to the thesis Mixed quarter
It confirmed the profit turn and the forecast. It did not answer the doubts behind the Unattractive view: two buyers still hold most of the sales, no filing says where the scandium comes from, and the class action has a lead plaintiff.
The report
In the quarter ended September 30, 2026 sales were $1,235.8M, up 138%. Bloom kept 33.8 cents of each sales dollar after making the product, operating profit was $225.4M against $7.8M, and the yearly forecast rose to $4.4B to $4.6B.
The cash did not follow the quarter: operating cash flow was minus $27M, because billing lagged the sales Bloom booked. Two customers, one a related party, bought 69% of the quarter.
What's happening now
The background
The report is in. The shares closed at $228.08 on 28 October, 21% below the 25 September close our earlier price claims used, so those claims are now re-struck at that close.
The next test is the full-year report in early 2027. The lines to read are who bought, whether a filing says where the scandium comes from, and whether cash from operations turns positive again.
Where we disagree
Both sides in full
The market
Bloom is the cleanest way to own on-site power for AI data centres, and the July short report was noise the results answered.
We believe
The profit turn is real, but it rests on one unnamed buyer, and the supply questions go to whether Bloom can grow.
The argument
Coverage reads the beat and the pile of future orders Bloom advertises. The filings show what those leave out: the concentration, an order tally Bloom defines itself that counts tax credits it expects to receive, and a denial on scandium that names no supplier.
Seventeen analysts rate it a buy, and the shares are up 317% in a year despite the July report. The market has decided the results answered the questions; the filings say they did not.
What it's worth
How we value it
Valued on the analysts' revenue path for three years, each later year closing 30% of the gap to 4%, on official operating profit with stock pay a cost and tax paid in full; the thesis's own evidence is the check on that path.
The three cases
Bear case$31.27
Growth runs at half the analysts' path, 30%, 25% and 15%, the margin settles a fifth lower at 14.6%, and new capital earns only the 8% rate: the downturn test.
Our value$74.86
Revenue grows 60%, 50% and 30% over three years, the analysts' path (C45), held to at most 60% a year, then fades toward 4%, closing 30% of the gap each year; the margin moves to 18.3%, what its September 2026 quarter earned before amortisation of past deals, not the analysts' 25.7%; a cyclical business's normal margin goes no higher than it has already filed; tax is 25%, the federal 21% plus state tax, the full rate once past losses are used up.
Bull case$148
Growth runs a quarter above the analysts' path, 75%, 62% and 38%, the margin reaches 20.1%, and new capital earns 11%.
The bet
Seen as a bet
Even if this works, the shares look expensive. Our best case is sales doubling again after this year's doubling, which needs Bloom to find about three more buyers as large as its one big unnamed customer. Even then, at a generous price for its sales, the company would be worth less than today.
If it fails, Bloom is still a real, profitable business with plenty of cash, but sales could fall back toward last year's level and the shares would be worth a fraction of today's price.
What could go wrong
Every risk
The risks
Two big customers
Two buyers took about 69% of the September quarter, one of them a related party, and nobody outside can see who they are.
Cash lags the sales
Operating cash flow was minus $27M in the quarter while billed-but-unpaid and not-yet-billed sales grew by about $880M over nine months.
Future orders nobody can check
Bloom's own tally counts tax credits it expects to receive; the audited figure for contracted revenue appears nowhere in its annual report.
A fraud allegation
A short seller says Bloom buys Chinese scandium through middlemen, and a class action followed. Bloom denied it without naming a supplier.
Its biggest named site is late
Oracle's New Mexico campus will run on Bloom cells, but its gas pipeline slipped to February 2027 and Oracle sent a force majeure notice.
Insiders selling, none buying
55 open-market sales and no purchases in a year, including four senior officers on 16 September.
What we still do not know
Who the customers behind 69% of September-quarter sales are, and why the Q3 10-Q's nine-month figures do not match the 73% in the Q2 one.
Where Bloom's scandium comes from. It says not China, and the 10-Q names no supplier.
What changes our view
Every route to a new view
Every route to a new view
→ Neutral
The share price falls to about $89.83, back within 20% of what we think the business is worth.
→ Attractive
The share price falls to about $59.89, 80% of what we think the business is worth.
Research reopens
A filing documents a scandium supply outside China, and the securities class action is dismissed.
The next test
It supports the case if
Two largest customers under half of revenue or named, the 10-K names a scandium supply chain outside China or the class action is dismissed, and operating cash flow is positive for the quarter.
It pressures the case if
Two customers are again about two-thirds of revenue, the 10-K adds nothing to the July denial, or operating cash flow is negative again while contract assets keep rising.
We would also be wrong if
Margins above 33% and cash coming in hold for four straight quarters; the September quarter's cash flow was negative, so the count restarts.
Insiders stop selling and buy shares with their own money.
The verdict
Scores and how we weighed it
The scores
Growth
4 of 4
Business quality
2 of 4
Financial position
4 of 4
Execution record
3 of 4
Per share capture
2 of 4
Potential
2 of 4
Risk profile
1 of 4
Valuation
1 of 4
Evidence quality
2 of 4
How we weighed it
ValueNegative
The price sits above our value and above the bull case.
BusinessMixed
A mixed business: a real profit turn, with weak evidence on customers.
GrowthPositive
Growth is fast and the yearly forecast held after the turn.
RisksNegative
Two buyers hold most sales, cash lags and future orders cannot be checked.
RivalsMixed
No filed rival compares directly; against its own past it is far dearer.
The groupMixed
Power names move together on AI power news.
Latest newsMixed
The last report confirmed the profit turn without answering the doubts.
In plain words
Our estimate of value
We forecast how much cash the business can earn over the next ten years and beyond, then ask what that stream is worth today to an investor who wants an 8% yearly return. We do it three times: a gloomy future, an expected one and a sunny one. We call a share Attractive only when its price is at least 20% below the expected estimate, and Unattractive when it is 20% above.