Powell Industries
Builds the custom electrical rooms, switchgear and breakers that sit between a power source and a big user of power: a data centre, a gas terminal, a chemical plant.
The scores
Growth2/4 ↑
TTM growth of 7.0% is the cohort's slowest and management has capped it, though quarterly growth accelerated from 4.0% to 8.9% through fiscal 2026.
Business quality3/4 →
Returns are more than double the cohort on scarce rated capacity, but the work is competitively bid, non-recurring, and a rival bought comparable capacity outright.
Financial position4/4 →
No drawn debt, cash and short-term investments of $634 million against equity of $756 million, a dividend paid and capacity expansion funded from cash.
Execution record3/4 ↑
Operating margin rose from 0.2% to 19.7% over five years and capacity landed on plan, though the last two quarters missed analyst models.
Per share capture4/4 →
Share count flat quarter over quarter, no debt to service and negligible intangibles, so operating profit reaches per-share earnings with almost no leakage.
Potential3/4 ↑
Data-centre load growth is real and behind-the-meter raises content per project, but no published conversion turns megawatts into switchgear dollars.
Risk profile3/4 →
No customer above 10% and no debt, against lumpy non-recurring project revenue, a 49% petrochemical decline and craft labour flagged as a fiscal 2027 constraint.
Valuation2/4 →
Second cheapest of four on the frame metric, but 26.2 times operating profit still assumes a backlog converts that management says will convert slowly.
Evidence quality2/4 ↓
A load-bearing quote on data-center boundaries did not reproduce against an independent transcription of the same call; corrected this review, so the cap applies.
The Verdict
We rate Powell neutral. It builds the custom electrical rooms and switchgear that sit between a power source and a big user such as a data centre, and it is a solid business that earns more than twice its rivals' return on capital. It just booked its biggest order ever, yet the shares fell by a third, because the backlog is turning into sales slowly. The price allows for that slow conversion, so it is full, not cheap, and becomes a bargain only if deliveries speed up. The view hinges on whether Powell can build fast enough. We think the fall removed the excess without creating a bargain.
The long caseData centres, gas export plants and the grid need power faster than switchgear can be built; if Powell can deliver faster, scarce delivery slots become its edge.
The evidence under this - 8 claims
The chief executive said on the same day's call that this order scale is not repeatable: 'Projects upwards of $400 million are not common in the pipeline other than potential future phases of this particular program.' A book-to-bill of 3.0 driven by one order above $400 million should therefore not be annualised, and the company's own release cautions that backlog 'may not be indicative of future operating results as orders may be cancelled or modified by our customers and associated backlog may not be recognized as revenue on the timeline we expect or at all'.
Gross margin of 30.6% was essentially flat against 30.7% a year earlier, so the 8% gross profit growth is volume rather than margin, and the headline mix shift is not costless: the 49% decline in Petrochemical is a real contraction in what was historically Powell's core market, offset by a commercial market that is newer, less proven for this company and - on the chief executive's own account - less price sensitive only while capacity stays scarce.
This does not make the thesis's headline wrong - the order was genuinely booked in fiscal Q3 and the $934 million order total is genuinely a record - but it does separate two events the August page ran together. The market learned of the largest order in Powell's history on 4 May and marked the shares up to a $328 high by early July; the 34% decline began after the 3 August call, whose new information was the 54% convertibility figure and the growth cap, not the order. What repriced the stock was conversion, not demand.
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
Management capping its own growth expectation is unusual and is evidence of candour rather than weakness, and the same executive framed the 9% print as one he was 'very happy with'. But it is also the single most important forward statement in this file: a company whose order book tripled telling the market not to expect double-digit revenue growth is describing a capacity ceiling, not a demand one, and that ceiling is what the multiple has to be judged against.
- nVent's 46.2% revenue growth - the figure that lifts the cohort median to 15.5% and makes Powell look slowest - is substantially acquired rather than organic: nVent bought Avail Infrastructure Solutions' switchgear and enclosures businesses for approximately $1 billion in cash during 2025. Comparing Powell's organic 7% against a median inflated by an acquisition overstates the growth gap, though it does not close it, since Eaton and Hubbell grew 15.5% and 10.6% without a deal of that scale.
- Powell's frame-metric discount to the cohort widened between 24 August and 6 September - from 27.5 against 31.4 to 26.2 against 32.0 - but roughly half of the subject's move is the vendor share-count correction recorded at C22 rather than anything the market did. The August reading was overstated by about 4.6%, so the thesis was harsher on Powell's price than the filed share count supported.
- 2026-10-27editedcohort multiples are as of the 4 September pack and were not re-struck after the fall
This own-history comparison is close to meaningless as a valuation signal, and saying so is the point: the prior medians span fiscal 2021 and 2022, when Powell's operating margin was 0.2% and 1.4% (C12). A company earning a 19.7% operating margin and a 22.6% return on invested capital is not the company those multiples were struck on, so the expansion from 1.01 to 5.70 times sales measures a change in the business at least as much as a change in what buyers will pay. The cohort comparison in C13, against live peers today, is the more informative reading.
- 2026-09-26editedCounter-evidence quoted price to sales of 5.95, the August pack's figure before the vendor's share-count correction (C22); the claim itself and the 6 September pack give 5.70.
- 2026-10-27editedown-history multiples are as of the 4 September pack and were not re-struck
The company
Anything that draws a lot of power, from a data centre to a gas export terminal, needs equipment between the supply and the load to distribute, switch and protect the electricity. Powell, founded in Houston in 1947, designs and builds that equipment, often as a whole prefabricated electrical room.
The alternative is buying switchgear, breakers and controls from several vendors and paying a contractor to assemble them on site, which is slower. In data centres, Powell wins on delivery speed rather than price.
The evidence under this - 3 claims
Competing on delivery speed is an advantage only while capacity is scarce, and the same executive expects the market's rate to attenuate. The rating diversion is a real if small cost of entering the data-centre vertical: the products Powell sold into these new jobs required certifications it had to go and obtain, which is evidence the commercial vertical is genuinely new ground rather than an extension of the existing catalogue.
How it makes money
Every job is custom and won by competitive bid, so Powell earns by the project: an order is booked, built over one to three years, and becomes sales as it ships. Big projects rarely repeat with the same customer, and no customer was more than 10% of sales in either of the last two years.
The evidence under this - 3 claims
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
Where the money comes from, stream by stream
| Stream | Share of sales (nine months to June 2026) | How it earns it |
|---|---|---|
| Oil and gas (excluding petrochemical) | 36.9% | Switchgear, power control rooms and motor control for upstream, midstream and LNG projects, bid and built to order. |
| Electric utility | 27.7% | Substation switchgear, protection and control equipment, including Remsdaq's remote terminal units, sold to utilities. |
| Commercial and other industrial | 19.9% | The same engineered equipment for data centres and other industrial sites; the fastest-growing line. |
| Petrochemical | 8.0% | Electrical systems for chemical plants; down from 14.0% of revenue a year earlier. |
| Light rail traction power and all others | 7.5% | Traction power for transit systems, plus universities, government and other buyers. |
Who runs it
Brett Cope has run Powell since 2016 and chairs its board; Michael Metcalf has run its finances since 2018. Under them it went from barely breaking even to keeping about a fifth of each sales dollar as operating profit. Insiders own only about 2%; the largest holder, Thomas Powell, is not on the board. Bonuses and share awards ride on a profit measure and on safety, with goals set below results already reached, and nothing pays on the share price.
The evidence under this - 5 claims
Insiders have been net sellers since: 79 open-market sales and no purchases over the year to September 2026, including the chief executive's 10b5-1 sale of 4,500 shares in August 2026 (C18), so the group's stake is if anything smaller now.
The proxy states that 75% of the chief executive's target pay is at risk, and both plans cap the payout at 200% of target, so the design is variable and bounded even if its goals were low (C32).
The fiscal 2023-2025 goal was set in September 2022, when operating margin was 1.4% (C12), so a low bar was a fair reading of the business at the time it was set.
Why it wins, and who could beat it
Why customers come back
The scarce thing in this industry is not design but factory capacity and the skilled workers to run it. So buyers commit early to secure a delivery slot: last quarter Powell booked orders worth 3.0 times its sales, and its backlog takes one to three years to deliver.
The same squeeze shows at rivals. nVent says its own large orders can take more than a year to build.
The evidence under this - 3 claims
The chief executive said on the same day's call that this order scale is not repeatable: 'Projects upwards of $400 million are not common in the pipeline other than potential future phases of this particular program.' A book-to-bill of 3.0 driven by one order above $400 million should therefore not be annualised, and the company's own release cautions that backlog 'may not be indicative of future operating results as orders may be cancelled or modified by our customers and associated backlog may not be recognized as revenue on the timeline we expect or at all'.
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
The threat
Its rivals are much larger, and Powell's own filing concedes some have far greater resources and lower costs. nVent, Eaton and Hubbell are the listed ones, and nVent's $1.0B purchase of a switchgear maker targets Powell's two growth markets. If rivals add capacity faster than demand grows, the delivery-slot advantage goes.
The evidence under this - 3 claims
- nVent's 46.2% revenue growth - the figure that lifts the cohort median to 15.5% and makes Powell look slowest - is substantially acquired rather than organic: nVent bought Avail Infrastructure Solutions' switchgear and enclosures businesses for approximately $1 billion in cash during 2025. Comparing Powell's organic 7% against a median inflated by an acquisition overstates the growth gap, though it does not close it, since Eaton and Hubbell grew 15.5% and 10.6% without a deal of that scale.
- Powell's frame-metric discount to the cohort widened between 24 August and 6 September - from 27.5 against 31.4 to 26.2 against 32.0 - but roughly half of the subject's move is the vendor share-count correction recorded at C22 rather than anything the market did. The August reading was overstated by about 4.6%, so the thesis was harsher on Powell's price than the filed share count supported.
- 2026-10-27editedcohort multiples are as of the 4 September pack and were not re-struck after the fall
The rivals: NVT, ETN, HUBB, and 1 more that is private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | How far along | Priced against it |
|---|---|---|---|
| NVT | nVent Electric, which in 2025 bought the enclosures, switchgear and bus systems businesses of Avail Infrastructure Solutions for approximately $1 billion in cashPriced against it: Yes | On sale | Yes |
| ETN · HUBB | Eaton and Hubbell, the two other US-listed multinationals selling power distribution equipment into the grid, data centres and industrialsPriced against it: Yes | On sale | Yes |
| Siemens Energy, Schneider Electric and ABB, the European multinationals in the same class Powell's Item 1 namesPriced against it: Listed abroad | On sale | Listed abroad |
How big it could get
How big this could get
The demand is large and hard to size. Government projections have data-centre servers rising from about 7% of the electricity American commercial buildings use in 2025 to 22% to 33% by 2050, but nobody converts that into equipment dollars.
The limit is Powell's own capacity. It made $1.10B of sales in fiscal 2025, and only about $1.3B of its $2.4B backlog can be delivered in the next year.
The evidence under this - 3 claims
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
Who has signed
Three large orders last quarter, none with a named customer: a data-centre power system worth more than $400M, a $75M fertiliser plant and a $60M gas export terminal. The data-centre job is only the first phase of a larger programme, and the later phases have not been awarded.
The evidence under this - 2 claims
The chief executive said on the same day's call that this order scale is not repeatable: 'Projects upwards of $400 million are not common in the pipeline other than potential future phases of this particular program.' A book-to-bill of 3.0 driven by one order above $400 million should therefore not be annualised, and the company's own release cautions that backlog 'may not be indicative of future operating results as orders may be cancelled or modified by our customers and associated backlog may not be recognized as revenue on the timeline we expect or at all'.
The future phases are unawarded and the chief executive qualified them twice - 'assuming we're successful' and 'nothing in the near term'. A two-and-a-half-year burn on a single order also means most of its revenue lands beyond this thesis's horizon, so the largest order in the company's history contributes to the backlog figure long before it contributes to earnings.
Where the growth is coming from, business by business
Where the growth is coming from
| Business | Share of revenue | Growth: last quarter | Last half-year | Last year |
|---|---|---|---|---|
| Oil and Gas Service | 34% | +0.8% | +1.5% | - |
| Electricity | 28% | +18.3% | +25.2% | - |
| Commercial and Other Industrial | 24% | +54.1% | +24.6% | - |
| Petrochemical | 6% | -49.2% | -40.7% | - |
| Other, Customers | 5% | +22.9% | +28.7% | - |
| Light Rail Traction Power Customer | 3% | -6.9% | -1.2% | - |
| Whole company | 100% | +8.9% | +6.7% | - |
The filed split through 2026-06-30. These rates are not carried forward: the reading below says why.
How we read it: The blend. Electricity (+25%) and data-centre-led Commercial (+25%) are one switchgear product, offset by Petrochemical's 49% fall; lumpy projects and 54% backlog conversion cap growth, per the CFO. 2 holds.
The results
The numbers
Sales grew 7.0% over the twelve months to Jun 2026, about the same pace as the 9.1% in the year to Sep 2025.
Operating margin is 20%, up from 8.9% in the year to Sep 2023.
Running costs take 10¢ of each sales dollar, down from 16¢ in the year to Sep 2021.
Each dollar put into the business earns 26% a year after tax, up from 0.2% in the year to Sep 2021.
It turned 21% of sales into free cash, $244M over the last twelve months.
Yearly sales growth has ranged from −9.2% to +45% over the last 5 years, with sales falling in 1 of them.
Free cash flow was 1.3 times net income.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-04. Fiscal years end in September.
| Sep 2021 | Sep 2022 | Sep 2023 | Sep 2024 | Sep 2025 | 12 mo to Jun 2026 | |
|---|---|---|---|---|---|---|
| Size and growth | ||||||
| Sales | $471M | $533M | $699M | $1.01B | $1.10B | $1.16B |
| Sales growth | −9.2% | +13% | +31% | +45% | +9.1% | +7.0% |
| Profit | ||||||
| Gross margin | 16% | 16% | 21% | 27% | 29% | 30% |
| Operating margin | 0.2% | 1.4% | 8.9% | 18% | 20% | 20% |
| Net margin | 0.1% | 2.6% | 7.8% | 15% | 16% | 16% |
| What growth costs, per sales dollar | ||||||
| Research and development | 1.4% | 1.3% | 0.9% | 0.9% | 1.0% | 1.3% |
| Selling, general and admin | 14% | 13% | 11% | 8.4% | 8.6% | 9.0% |
| Pay in shares | 0.5% | 0.8% | 0.7% | 0.5% | 0.4% | 0.5% |
| Capital spending | 0.6% | 0.5% | 1.1% | 1.2% | 1.2% | 1.1% |
| Cash | ||||||
| Cash from operations | −$30.5M | −$3.58M | $183M | $109M | $168M | $256M |
| Free cash flow | −$33.4M | −$6.03M | $175M | $96.7M | $155M | $244M |
| Balance sheet | ||||||
| Cash and short-term investments | $134M | $117M | $279M | $358M | $476M | $634M |
| Debt | $400K | $0 | - | - | - | - |
| Net cash | $134M | $117M | - | - | - | - |
| Returns and shares | ||||||
| Return on capital | 0.2% | 2.4% | 15% | 33% | 30% | 26% |
| Shares (diluted) | - | - | - | - | - | 36.6M |
What growth costs
For years, growing made Powell more profitable: sales doubled while office, selling and research costs grew far less. That effect has stalled; since 2024 those costs have grown a little faster than sales, so recent margin gains came from the factory floor. It spends little on buildings, leasing a plant rather than building one.
The evidence under this - 3 claims
The fiscal 2025 10-K attributes part of that year's SG&A rise to higher compensation and to costs of the Remsdaq acquisition (C34), so some of the reversal is a one-off rather than a trend.
- The leased facility does not produce anything until late in fiscal Q2 or early Q3 2027 - March to June 2027 - which is after the two prints this thesis's next test and horizon turn on. So the capacity that would lift backlog convertibility above 54% is not available during the period in which convertibility is expected to recover, and the recovery, if it comes, has to come from the existing footprint. Against that, the chief executive attributes the current throughput to running the job across at least five existing plants rather than to any single new one.
- The $70 million to $100 million greenfield figure is the alternative Powell declined, so it quantifies the capital discipline the thesis credits - but it also quantifies what a competitor willing to spend would need in order to add the capacity that constitutes Powell's moat, which is roughly a tenth of what nVent paid to buy comparable capacity outright.
The last report (3 Aug)
Quarter ended June 2026, filed figures.
The third quarter of fiscal 2026, to June 30, set records for orders and backlog. Sales grew 9%, the third quarter in a row of faster growth, led by data centres and other commercial work, up 54%, while petrochemical sales fell 49%.
Two things spoiled it. The share of backlog deliverable within a year dropped from the low sixties to 54%, and the finance chief said double-digit growth was probably not coming. And gross margin has stopped rising against the same quarters a year earlier.
What it did to the thesisIt proved demand is strong but showed the limit is Powell's own capacity: the record orders will turn into sales slowly, and margins stopped rising. That is why the view is Neutral rather than Attractive. The fall in the shares removed the excess without making them cheap.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $1.42 against $1.47 expected; $1.42 as filed, with no adjustments.
The evidence under this - 6 claims
Gross margin of 30.6% was essentially flat against 30.7% a year earlier, so the 8% gross profit growth is volume rather than margin, and the headline mix shift is not costless: the 49% decline in Petrochemical is a real contraction in what was historically Powell's core market, offset by a commercial market that is newer, less proven for this company and - on the chief executive's own account - less price sensitive only while capacity stays scarce.
The chief executive said on the same day's call that this order scale is not repeatable: 'Projects upwards of $400 million are not common in the pipeline other than potential future phases of this particular program.' A book-to-bill of 3.0 driven by one order above $400 million should therefore not be annualised, and the company's own release cautions that backlog 'may not be indicative of future operating results as orders may be cancelled or modified by our customers and associated backlog may not be recognized as revenue on the timeline we expect or at all'.
The acceleration is against a fiscal 2025 whose own quarters strengthened through the year, so part of it is a comparison base that was weakest at the start; and it is acceleration within a band the chief financial officer has explicitly capped below double digits, so it is a better shape rather than a different destination. Trailing-twelve-month growth of 7.0% remains the slowest of the four-name cohort against a 15.5% median.
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
Management capping its own growth expectation is unusual and is evidence of candour rather than weakness, and the same executive framed the 9% print as one he was 'very happy with'. But it is also the single most important forward statement in this file: a company whose order book tripled telling the market not to expect double-digit revenue growth is describing a capacity ceiling, not a demand one, and that ceiling is what the multiple has to be judged against.
Two quarters of 13 and 30 basis points is a small movement and well inside the noise of a project business whose margin depends on which jobs closed in which quarter - it is a flattening rather than a deterioration, and the level itself, 30.6%, is the highest third-quarter margin Powell has printed. It matters only because the thesis's A02 assumption is that margin holds near 30% as closeout benefits fade, and the like-for-like series is the first evidence that the fade is now showing up rather than still pending.
What management did with the money
It has spent lightly on capacity: $12.4M to expand a Houston yard by 62%, and a leased plant due in 2027 rather than a new build it put at $70M to $100M. It has no borrowings and $634M of cash.
The evidence under this - 2 claims
- The leased facility does not produce anything until late in fiscal Q2 or early Q3 2027 - March to June 2027 - which is after the two prints this thesis's next test and horizon turn on. So the capacity that would lift backlog convertibility above 54% is not available during the period in which convertibility is expected to recover, and the recovery, if it comes, has to come from the existing footprint. Against that, the chief executive attributes the current throughput to running the job across at least five existing plants rather than to any single new one.
- The $70 million to $100 million greenfield figure is the alternative Powell declined, so it quantifies the capital discipline the thesis credits - but it also quantifies what a competitor willing to spend would need in order to add the capacity that constitutes Powell's moat, which is roughly a tenth of what nVent paid to buy comparable capacity outright.
The price, and where we disagree
In short. Full. The price already allows for orders Powell has won turning into sales slowly; it becomes a bargain only if they start arriving faster.
The shares have fallen 44.77% from their $328.00 high. The data-centre order was announced in May and the shares peaked with it already known, so the fall after the June-quarter results was a verdict on how fast the backlog converts, not on the order.
The next check is the fiscal fourth-quarter results on November 17.
The evidence under this - 2 claims
- 2026-10-27editedre-struck at the 26 October close of $157.07 after a 13.3% fall
This does not make the thesis's headline wrong - the order was genuinely booked in fiscal Q3 and the $934 million order total is genuinely a record - but it does separate two events the August page ran together. The market learned of the largest order in Powell's history on 4 May and marked the shares up to a $328 high by early July; the 34% decline began after the 3 August call, whose new information was the 54% convertibility figure and the growth cap, not the order. What repriced the stock was conversion, not demand.
How we read the multiple
Debt included, buyers pay about 26 times Powell's yearly operating profit, the profit from its business before interest and tax. Its three listed rivals' middle is 32. Powell also earns the most on the money invested in it: 22.6% against their 9.6%.
The catch is growth. At 7.0% a year it is the slowest of the four, and the finance chief has said it probably will not reach double digits at the current pace.
Against its own history. Far dearer than its own past, but that past includes years it barely made money, so the old normal is a poor guide.
The evidence under this - 4 claims
- nVent's 46.2% revenue growth - the figure that lifts the cohort median to 15.5% and makes Powell look slowest - is substantially acquired rather than organic: nVent bought Avail Infrastructure Solutions' switchgear and enclosures businesses for approximately $1 billion in cash during 2025. Comparing Powell's organic 7% against a median inflated by an acquisition overstates the growth gap, though it does not close it, since Eaton and Hubbell grew 15.5% and 10.6% without a deal of that scale.
- Powell's frame-metric discount to the cohort widened between 24 August and 6 September - from 27.5 against 31.4 to 26.2 against 32.0 - but roughly half of the subject's move is the vendor share-count correction recorded at C22 rather than anything the market did. The August reading was overstated by about 4.6%, so the thesis was harsher on Powell's price than the filed share count supported.
- 2026-10-27editedcohort multiples are as of the 4 September pack and were not re-struck after the fall
This own-history comparison is close to meaningless as a valuation signal, and saying so is the point: the prior medians span fiscal 2021 and 2022, when Powell's operating margin was 0.2% and 1.4% (C12). A company earning a 19.7% operating margin and a 22.6% return on invested capital is not the company those multiples were struck on, so the expansion from 1.01 to 5.70 times sales measures a change in the business at least as much as a change in what buyers will pay. The cohort comparison in C13, against live peers today, is the more informative reading.
- 2026-09-26editedCounter-evidence quoted price to sales of 5.95, the August pack's figure before the vendor's share-count correction (C22); the claim itself and the 6 September pack give 5.70.
- 2026-10-27editedown-history multiples are as of the 4 September pack and were not re-struck
Management capping its own growth expectation is unusual and is evidence of candour rather than weakness, and the same executive framed the 9% print as one he was 'very happy with'. But it is also the single most important forward statement in this file: a company whose order book tripled telling the market not to expect double-digit revenue growth is describing a capacity ceiling, not a demand one, and that ceiling is what the multiple has to be judged against.
Priced against its rivals
The competitors its own annual report names, each confirmed against the rival's filing. Measured on company value for each $1 of operating profit, cheapest first:
| Ticker | Company | company value for each $1 of operating profit |
|---|---|---|
| HUBB | Hubbell Incorporated | 23.3× |
| POWL | Powell Industries | 26.2× |
| NVT | nVent Electric | 32.0× |
| ETN | Eaton | 34.0× |
The middle of the other 3 is 32.0. Powell Industries, Inc. is 2nd cheapest of 4.
Rivals' figures come from a data provider, so read the spread of the group rather than any single pair.
Mid-pack against its listed rivals despite the best returns of the group. The discount is its own, and it is for slower growth.
Private or foreign rivals that cannot be priced: Siemens Energy, Schneider Electric, ABB and Mitsubishi Electric, Smaller regional competitors.
Where we disagree with the market
A slowing industrial that ran too far: sales grew 9% while orders tripled, and only 3 of 10 analysts say buy.
The fall removed the excess without creating a bargain. The growth ceiling management describes is real, and set by its own capacity.
The market is reading Powell against its own history, when it barely made money. Against live rivals it is mid-pack on price and earns double their returns, which is why we are not bearish.
We would be too cautious if the deliverable share of backlog recovers toward the low sixties, growth outpaces rivals again, and the later data-centre phases come to Powell.
What analysts say, and the evidence
The evidence under this - 6 claims
- The two sales closest to the fiscal Q3 print were both pre-arranged, which removes the timing inference: the chief executive's 4,500 shares at $207.01 on 2026-08-13 were sold under a Rule 10b5-1 plan he adopted on 2025-11-26, and director Richard Williams's 2,250 shares at $203.05 on 2026-08-14 under a plan adopted 2026-04-06 - both months before the convertibility disclosure of 2026-08-03. The chief executive retained 512,733 shares, so the sale was about 0.9% of his holding. Zero open-market purchases across a year remains a real absence; selling ten days after a bad call does not.
- Short interest at 8.66% of float is elevated in absolute terms but has fallen from 2,817,817 shares in late May to 2,499,783 in mid-August, so the short base was being covered into the decline rather than built. The figure is settled twice a month and is read as of 2026-08-14, not as of today.
The chief executive said on the same day's call that this order scale is not repeatable: 'Projects upwards of $400 million are not common in the pipeline other than potential future phases of this particular program.' A book-to-bill of 3.0 driven by one order above $400 million should therefore not be annualised, and the company's own release cautions that backlog 'may not be indicative of future operating results as orders may be cancelled or modified by our customers and associated backlog may not be recognized as revenue on the timeline we expect or at all'.
- nVent's 46.2% revenue growth - the figure that lifts the cohort median to 15.5% and makes Powell look slowest - is substantially acquired rather than organic: nVent bought Avail Infrastructure Solutions' switchgear and enclosures businesses for approximately $1 billion in cash during 2025. Comparing Powell's organic 7% against a median inflated by an acquisition overstates the growth gap, though it does not close it, since Eaton and Hubbell grew 15.5% and 10.6% without a deal of that scale.
- Powell's frame-metric discount to the cohort widened between 24 August and 6 September - from 27.5 against 31.4 to 26.2 against 32.0 - but roughly half of the subject's move is the vendor share-count correction recorded at C22 rather than anything the market did. The August reading was overstated by about 4.6%, so the thesis was harsher on Powell's price than the filed share count supported.
- 2026-10-27editedcohort multiples are as of the 4 September pack and were not re-struck after the fall
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
The future phases are unawarded and the chief executive qualified them twice - 'assuming we're successful' and 'nothing in the near term'. A two-and-a-half-year burn on a single order also means most of its revenue lands beyond this thesis's horizon, so the largest order in the company's history contributes to the backlog figure long before it contributes to earnings.
What could go wrong, and how we would know
The risks
- Not enough skilled workers. The chief executive expects craft labour to be a challenge in 2027 and 2028, which would delay deliveries.
- Margins have stopped rising. Gross margin was lower than a year earlier in each of the last two quarters, as one-off gains from closing projects fade.
- Orders can be cancelled. Customers may walk away from the backlog, usually paying costs and a margin rather than the full order.
- Insiders selling. Insiders made 79 open-market sales and no purchases in the past year.
The evidence under this - 5 claims
- The leased facility does not produce anything until late in fiscal Q2 or early Q3 2027 - March to June 2027 - which is after the two prints this thesis's next test and horizon turn on. So the capacity that would lift backlog convertibility above 54% is not available during the period in which convertibility is expected to recover, and the recovery, if it comes, has to come from the existing footprint. Against that, the chief executive attributes the current throughput to running the job across at least five existing plants rather than to any single new one.
- The $70 million to $100 million greenfield figure is the alternative Powell declined, so it quantifies the capital discipline the thesis credits - but it also quantifies what a competitor willing to spend would need in order to add the capacity that constitutes Powell's moat, which is roughly a tenth of what nVent paid to buy comparable capacity outright.
Two quarters of 13 and 30 basis points is a small movement and well inside the noise of a project business whose margin depends on which jobs closed in which quarter - it is a flattening rather than a deterioration, and the level itself, 30.6%, is the highest third-quarter margin Powell has printed. It matters only because the thesis's A02 assumption is that margin holds near 30% as closeout benefits fade, and the like-for-like series is the first evidence that the fade is now showing up rather than still pending.
Roughly 100 basis points of a 29.6% year-to-date gross margin is a non-recurring benefit from closing out completed projects favourably, so the underlying run-rate margin is nearer 28.6%. The contribution is falling year over year, which means the tailwind is already fading rather than about to; against that, the chief executive argues margins have not peaked because automation and the service strategy are accretive, and those are recurring where closeouts are not.
The cost-plus-reasonable-margin reimbursement materially softens this: a cancelled Powell project is not a total loss, and the company reports bookings net of cancellations rather than gross, so the historical order series already absorbs whatever cancellation has occurred. But it is why the three mega orders are recorded here as conditional rather than firm commitments. Nothing in the record quarter is a filed offtake agreement with a named counterparty; they are purchase orders in a backlog whose own governing document says customers may terminate at their discretion.
- The two sales closest to the fiscal Q3 print were both pre-arranged, which removes the timing inference: the chief executive's 4,500 shares at $207.01 on 2026-08-13 were sold under a Rule 10b5-1 plan he adopted on 2025-11-26, and director Richard Williams's 2,250 shares at $203.05 on 2026-08-14 under a plan adopted 2026-04-06 - both months before the convertibility disclosure of 2026-08-03. The chief executive retained 512,733 shares, so the sale was about 0.9% of his holding. Zero open-market purchases across a year remains a real absence; selling ten days after a bad call does not.
- Short interest at 8.66% of float is elevated in absolute terms but has fallen from 2,817,817 shares in late May to 2,499,783 in mid-August, so the short base was being covered into the decline rather than built. The figure is settled twice a month and is read as of 2026-08-14, not as of today.
What we still do not know
- Whether the later phases of the data-centre programme come to Powell. None has been awarded.
- Whether the leased plant, due in 2027, lifts how much of the backlog converts each year.
The evidence under this - 3 claims
The future phases are unawarded and the chief executive qualified them twice - 'assuming we're successful' and 'nothing in the near term'. A two-and-a-half-year burn on a single order also means most of its revenue lands beyond this thesis's horizon, so the largest order in the company's history contributes to the backlog figure long before it contributes to earnings.
- The leased facility does not produce anything until late in fiscal Q2 or early Q3 2027 - March to June 2027 - which is after the two prints this thesis's next test and horizon turn on. So the capacity that would lift backlog convertibility above 54% is not available during the period in which convertibility is expected to recover, and the recovery, if it comes, has to come from the existing footprint. Against that, the chief executive attributes the current throughput to running the job across at least five existing plants rather than to any single new one.
- The $70 million to $100 million greenfield figure is the alternative Powell declined, so it quantifies the capital discipline the thesis credits - but it also quantifies what a competitor willing to spend would need in order to add the capacity that constitutes Powell's moat, which is roughly a tenth of what nVent paid to buy comparable capacity outright.
This is the central mechanical finding and it cuts against the headline: the record order quarter made the backlog longer-dated, not shorter. Near-term convertibility fell from the low sixties to 54% precisely because the orders won were large and slow. The offsetting reading is that a longer-dated backlog is also a more durable one - $2.4 billion against roughly $1.2 billion of annual revenue is more than two years of work already booked, which is a different and better problem than an empty order book.
What would change our view
- Neutral→AttractiveThe backlog converts on time
It needs all of:
- ○The share of backlog deliverable within a year holds at 54% or better, and gross margin stops falling against last year.checked at the next results, due by 17 Nov
- ○Quarterly orders stay well above the $362M of a year earlier, so the record is a level, not a one-off.checked at the next results, due by 17 Nov
- Neutral→UnattractiveLabour or a competitor takes the work
It needs any of:
- ○Shortages of skilled workers delay deliveries, as the chief executive says they may from fiscal 2027.watched as it happens
- ○The later phases of the data-centre programme go to a competitor.watched as it happens
We would also be wrong if
- Gross margin falls against the year before for three quarters in a row.
What is coming
- 17 November 2026 Q4 fiscal 2026 results and the fiscal 2026 10-K: Backlog convertibility against 54%, fourth-quarter gross margin against 31.39% a year earlier, orders against $362M, and whether the Jacintoport yard finished on time.
Every dated event ahead
- 17 November 2026 Q4 fiscal 2026 results and the fiscal 2026 10-K: Backlog convertibility against 54%, fourth-quarter gross margin against 31.39% a year earlier, orders against $362M, and whether the Jacintoport yard finished on time.
- January 2027 Proxy statement for the February 2027 annual meeting: The fiscal 2026 bonus against its EBITDA goal, whether fiscal 2027 goals are set above results already reached, and insider and Powell-holder stakes after the split.
- February 2027 Q1 fiscal 2027 results: Whether convertibility keeps recovering and overhead per sales dollar stops rising, before any new capacity is running.
- June 2027 Leased Houston facility begins manufacturing: Whether capacity arrives as promised, late in fiscal Q2 or early Q3 2027, the step that lets the $2.4B backlog convert faster.
Since we wrote this
- 27 OctPrice moveShares fell 13% with no company news; view heldThe price claim was re-struck at $157.07. The view rests on backlog conversion, which only the 17 November results can show.
- 5 OctThesis reviewThe long case now sits under the verdictAfter the five-name pilot, the owner asked for the long case on every thesis.
- 4 OctThesis reviewMargin-promise claim carried to the November print. View heldFiscal 2026 has closed but not been reported, so the claim can only be judged on the 17 November release. Nothing new exists to move the view before then.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the proxy and filings.
- 26 SepThesis reviewValuation text brought to 26.2 times profit; view heldFour passages still quoted 27.5 times operating profit against a peer median of 31.4, from before a share-count correction. The September figures, 26.2 against 32.0, leave Powell second cheapest of four.
- 6 SepThesis reviewOrder was known at the high; margins stalled. View heldThe data-centre order was public in May, so the fall priced slower conversion. A vendor share-count error had overstated the multiples, and gross margin has stopped rising. The evidence cut both ways.
- 24 AugThesis writtenNeutral, medium convictionWritten three weeks after record orders and a third off the share price, on a call that cut how fast the backlog becomes sales.
Also checked 1 time with nothing new to act on.
Sources 19
Sources - 19
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2025-11-19business and product description, competition classes, customer-concentration language, the Jacintoport capacity expansion, the backlog definition and the pre-split share countbacksC01C02C03C10C12C13C14C23C24C29C30C34C35
- S02primarycompany filing2026-08-03fiscal Q3 2026 revenue, margin and earnings; new orders, book-to-bill and backlog; the three mega orders; end-market revenue movements; cash and short-term investments; and the backlog cautionbacksC04C05C06C16C18C27C28
- S03primarycompany filing2026-08-04the three-for-one stock split detail, the post-split share count, the equity roll-forward including treasury shares, the undrawn revolver and letters of credit, and the intangible asset schedulebacksC03C16C22C28C34C35C36
- S04primarytranscript2026-08-04Powell Industries fiscal Q3 2026 earnings call transcript (question section)Alpha Vantage EARNINGS_CALL_TRANSCRIPT, POWL 2026Q3the backlog convertibility figures, the growth cap, the project-closeout margin contribution, the capacity and labour commentary, the data-centre order detail and the behind-the-meter content mechanism
Of the $2.4 billion of backlog, roughly just under $1.3 billion will be convertible over the next 12 months, so roughly 54%. As we spoke last quarter, that was in the low 60s, and with this big slug of orders - the $900-plus million order bookings that we recognized this quarter - that convertibility slipped down to about 54%. [...] our book-and-burn cadence is about $75 million a quarter
backsC06C07C08C09C10C11C20C26C27 - S05primarycompany filing2026nVent Electric plc Form 10-K for fiscal year 2025 nVent Electric plca competitor's own filing pricing the same switchgear capacity at $1 billion, confirming the product overlap, and corroborating the one-year-plus design and manufacturing cycle for infrastructure orders
In 2025, we completed the acquisition of the enclosures, switchgear and bus systems businesses of Avail Infrastructure Solutions (the "Electrical Products Group") for approximately $1.0 billion in cash.
backsC15 - S06secondaryvendor data2026-08-24Portfolio-management research pack (market tier and filed statement series)content/POWL/research-pack.jsonprice and technicals, volatility bucket and beta, analyst consensus and earnings-surprise history, insider Form 4 aggregation, measured news reaction, own-history valuation series, and the filed annual and quarterly revenue and margin series from the filer's own XBRLbacksC01C12C14C17C18C22
- S07secondaryvendor data2026-08-24Portfolio-management research pack (vendor_fundamentals tier: peer_valuation cohort)content/POWL/research-pack.jsoncohort valuation and operating metrics against nVent, Eaton and Hubbell, with all three comparable and none withheld for a filed-versus-vendor margin gapbacksC13
- S08primarycompany filing2026-08-03the declared quarterly cash dividend of $0.09 per share, payable September 16, 2026backsC16
- S09primaryindustry data2026EIA - Data center server energy use grows across the commercial building stock US Energy Information Administrationthe independent national demand base for the prize block, reused from this repo's Constellation Energy and Bloom Energy theses
Across all cases, servers alone accounted for an estimated 7% of commercial sector electricity consumption in 2025. Data center server electricity use grows to 22%-33% of commercial building electricity use by 2050 across our cases.
backsC19 - S10secondaryindustry data2026independent tracking of behind-the-meter generation actually online, reused from this repo's Vistra, Constellation Energy and Bloom Energy theses
approximately 2 GW of behind-the-meter generation capacity is online across four projects, with the vast majority belonging to xAI
backsC21 - S11secondaryvendor data2026-08-24Portfolio-management filed-competitor roster (data/reference/competitors.json, POWL entry)data/reference/competitors.jsonthe recorded cohort search: the class Powell names, the candidates chased, the confirmation verdict on each, and the unlisted European and regional rivals the cohort cannot reachread, but no claim rests on it
- S12primarycompany release2025-11-18backsC12C18C25
- S13primarycompany filing2026-05-04the 4 May announcement of the $400 million-plus data-centre order as a subsequent event, the chief financial officer's forward margin statement, and fiscal Q2 2026 revenue, gross margin, orders and backlog - the document the source audit named as unread at the August reviewbacksC26C27C28
- S14primarytranscript2026-08-04Powell Industries fiscal Q3 2026 earnings call transcript (FMP, full call including prepared remarks)FMP earning-call-transcript, POWL 2026Q3a second transcription of the same call, against which every quoted passage in this ledger was re-checked; the prepared remarks, which carry the leased-facility timing, the $70-100 million greenfield alternative and the 'previously announced' description of the data-centre order; and the finding that the August page's data-centre boundary quote does not appear in the call. Same origin as S04 - one call, two vendors, not two pieces of evidence
at the compute level, where we don't compete, the 1-megawatt rack is definitely on its way and it is built around DC distribution technology. And so as that moves up the power curve, the outside of the data center will remain largely AC Energy. There's a mix of ideas to do DC. I mean a really wide-ranging mix of DC stuff that will get inside the data center. And yes, I still see an opportunity for Powell more midterm on the DC side, but we definitely are looking at it.
read, but no claim rests on it - S15primarycompany filing2026-08-17the footnotes establishing that both sales closest to the fiscal Q3 print were executed under Rule 10b5-1 plans adopted 2025-11-26 and 2026-04-06, and the shares retained after eachbacksC18
- S16primarycompany filing2026-01-06directors and executive officers with tenure, beneficial ownership as of December 19, 2025, the fiscal 2025 bonus and performance-unit metrics, targets and payouts, and the summary compensation tablebacksC29C30C31C32C33
- S17primarycompany filing2026-07-06the special retention award of 36,000 restricted stock units to the chief executive, its back-loaded vesting and the retirement-at-60 provision it overridesbacksC33
- S18primarycompany filing2026-10-04revenue, cost of sales, R&D, SG&A, stock pay, capital spending and diluted shares for FY2021 to FY2025 and the four quarters to June 2026backsC32C34C35
- S19secondaryvendor data2026-10-27the 26 October close, 252-day range, moving averages, returns and the 20-day rangebacksC17