$POWL: Two Years of Orders, Priced Like One

The 60 second version

• Powell Industries builds custom medium voltage switchgear, prefabricated electrical rooms and control systems, and it sells them one engineered project at a time.
• The shift: data centre developers can no longer wait for grid power, so they are building generation on site and buying the electrical gear that ties it together.
• On our curve the product is in Flight. The constraint easing is Powell’s own engineering and fabrication capacity, with new yards and leases arriving through fiscal 2027.
• The evidence: $934M of orders in the June quarter, including a data centre order above $400M, took backlog to a record $2.4B, about 2.1 years of revenue.
• Next test: fiscal Q4 results, expected in November 2026. We want revenue of at least $335M and backlog holding at $2.4B or more.
• The thesis breaks on Q4 revenue under $310M, backlog under $2.2B, any cut to the big data centre order, or gross margin under 27%.

In the three months to June 30, Powell Industries took in $934M of new orders and shipped $312M of revenue. When a manufacturer books three dollars for every dollar it delivers, it has a production problem. We think that is the best problem a hardware company can have.

The market sees it differently. The shares have fallen from a 52 week high of $328 to $161.36, about half, and that slide includes a 10.9% drop on October 22 that we could not tie to any company news. What the price reflects is 9% revenue growth, a small earnings miss and a general retreat from anything touching AI spending.

Our thesis in one sentence: Powell’s revenue is limited by how fast it can engineer and build, the capacity to do more arrives during fiscal 2027, and at roughly $5.3B of enterprise value the price asks only for growth that signed orders already cover.

The claim is a narrow one. Orders Powell has already accepted have to get built, and the company has the cash to build the floor space they need. Everything below tests that.

The shift

For most of the last decade, putting up a data centre came down to land, fibre and a grid connection. That last item has turned into the binding constraint. Utilities cannot add transmission and substations at the speed AI developers want to add load, so the developers with the most money have started generating their own power behind the meter and using the grid as backup or not at all.

That changes who gets paid. A building fed by the grid needs switchgear at the fence line. A building with its own gas turbines or engines needs an entire electrical system to be engineered: generation tied into medium voltage distribution, protection, controls and monitoring, all working as a self contained island. The industry calls this a power island. Powell’s order above $400M was Phase 1 of exactly that design, a multiphase behind the meter project.

We place data centre power procurement in a phase where buyers have stopped asking whether to spend and are asking how fast suppliers can deliver. The constraint eased on the buyer’s side during 2025 and 2026, once developers accepted that paying for on site generation was cheaper than waiting years for a grid connection. The bottleneck then moved one step down the chain to the companies that build the equipment.

A short aside, because it helps. $AAON, which sells cooling equipment into many of the same buildings, also reported a record quarter in August. The cooling suppliers and the power suppliers are running into the same wall, which is factory output. Watching both helps us separate real demand from hype, since a developer can announce a campus far more easily than it can fund two separate equipment orders.

The company

Powell is a Houston manufacturer that designs and builds custom equipment to distribute and control electrical power. Its main products are medium voltage switchgear, prefabricated electrical rooms that ship as finished buildings, automation and control systems, and field services. Every job is engineered to order. Revenue is recognised as the equipment is fabricated and delivered, so the income statement can grow only as fast as the shop floor.

The company grew up serving oil and gas, petrochemical and LNG owners and their contractors on the Gulf Coast, along with electric utilities. That heritage matters more than it appears to. Offshore platforms are power islands by definition: no grid, everything self contained, everything built to survive. Powell spent decades building the very configuration that behind the meter data centres now want.

The detail that stood out to us in the filings is how little the business spends to grow. Over nine months Powell earned $139.4M of net income and spent only $10.4M on capital expenditure. Cash and short term investments rose from $476M to $634M. There is no debt. The quarterly dividend costs about $3.3M, which is close to a rounding error against that cash pile.

The automation side carries a less obvious detail. Through Remsdaq, its SCADA remote terminal unit business, Powell now makes in house some components it used to buy. That raises the content Powell keeps on each job. The company also split its shares three for one on April 2, 2026, and the diluted count has barely moved, at 36.60M against 36.53M a year earlier.

Where it sits on our curve

We put the product the thesis depends on, custom medium voltage packages for data centres and utility generation, in Flight. Demonstration is long past and the buyers are paying. The open question is execution: can Powell convert a backlog that has nearly doubled into revenue at the margin it earns today?

The buyer’s threshold is speed. Management said on the August call that delivery time is the main thing data centre customers pay for, ahead of price. We have no public lead time figure, so we judge the threshold through our own numbers. Quarterly revenue has to rise from $312M to an average of roughly $370M across fiscal 2027 while gross margin stays in a 29% to 31% band. The gap is meaningful but closable, and the capacity to close it is already funded and partly built.

On the evidence sequence, Powell has reached expansion. These are accepted purchase orders. Revenue from Commercial and Other Industrial, the segment that holds data centres, rose 54% to $77M in the June quarter, and the segment now makes up 40% of backlog. The large order is Phase 1 of a design meant to have later phases, which is what expansion looks like before it reaches the revenue line.

Is it repeatable or a surge? Partly both, and we will say so plainly. One customer’s Phase 1 accounts for a large slice of the backlog jump. But orders ran at $439M and $490M in the two quarters before it, well above the roughly $300M quarterly average of fiscal 2025, and utility revenue grew 18%. The base was rising before the big order arrived.

The numbers and what the street expects

Start with the quarter the market punished. Fiscal Q3 revenue was $311.7M, up 9% on the year and 5% on the prior quarter. Analysts had expected $316.7M. Earnings per share came in at $1.42 against estimates of $1.47 to $1.49. These figures come from Powell’s August 3 results release and the consensus tracked by MarketBeat.

Look underneath and the mix is changing fast. Data centre revenue grew 54% and utility revenue 18%, while petrochemical revenue fell 49%. Nine month revenue of $859.5M grew only 6.6% because the old core is shrinking while the new one grows. We read the 9% headline as a blend of two very different businesses.

Orders tell the forward story. Powell booked $271M in its fiscal Q4 of 2025, then $439M, $490M and $934M in the first three quarters of fiscal 2026. That is a book to bill of 3.0 for the June quarter and 2.2 for the year to date. Backlog went from $1.4B at September 2025 to $1.8B at March and then $2.4B at June, up 69% on the year.

Against trailing revenue of $1.16B, that backlog equals about 2.1 years of sales. By end market it splits 40% Commercial and Other Industrial, 24% utility and 30% petrochemical plus oil and gas. Three mega orders landed in the quarter: the data centre job above $400M, a fertilizer order of $75M and an LNG order of roughly $60M.

Margins have held. Gross margin was 30.6% in the June quarter against 30.7% a year earlier and 29.4% for fiscal 2025. Be careful with that figure, though. Project closeouts added about 100 basis points year to date, so the underlying run rate is nearer 29.5%. Management also flagged moderate inflation in copper, aluminum, steel and engineered components.

Now the forward picture, from StockAnalysis and MarketBeat as of October 29. Fiscal Q4 consensus is EPS of $1.52 on revenue of $335M, up 12% on the $298.0M of a year earlier. For the full year to September 2026, analysts expect revenue of about $1.19B to $1.20B and EPS of $5.36 to $5.38, against $4.95 last year. Revisions over the last ninety days have been roughly flat, with modest cuts after the August miss.

Fiscal 2027 is where the spread matters. Revenue consensus has ranged between $1.43B and $1.48B in recent snapshots, growth of 19% to 24.5%, with EPS of $6.87 to $6.99. Management gives no numeric guidance. It expects gross margin to stay in line with the trailing twelve months and called fiscal 2027 “another year of strong financial performance.” So the street is already modelling a step up of about a fifth that the company has not put a number on.

That gap is the most important thing in this section. Consensus is ahead of management’s words but behind the backlog. Which side it falls on depends almost entirely on when the new capacity starts producing.

Why it wins

Follow the money through a single job. The customer pays a contract price for an engineered project: a switchgear lineup, a prefabricated electrical room, a power island package. Each extra project brings copper, aluminum, steel, bought components, engineering hours, fabrication labour and floor space. None of those costs falls with scale the way software costs do, so Powell’s advantage has to come from what it charges and how fast it delivers.

For now Powell keeps the benefit. Gross margin has held near 30% while volume climbs, management described pricing as stable, and it sees room on price in the commercial market because speed matters more than cost there. A developer with a billion dollar building waiting on power will not haggle over a few points on the switchgear. Customers are not yet pushing those gains back to Powell through lower prices.

What happens if AI models get far better and far cheaper? Cheaper inference means more inference, more racks and more megawatts. Powell sells the hardware that moves those megawatts, and a better model replaces none of it. If anything, efficiency gains widen the demand base by putting AI within reach of more users.

The moat is narrower than a patent and more practical. It rests on decades of building self contained power systems for offshore platforms, Gulf Coast yards sized for large modules, an engineering team that has delivered mega projects, and in house automation through Remsdaq. Catalogue products can be matched by any big electrical maker. A custom power island delivered on schedule is harder to copy, and we think that is where Powell’s edge actually lives.

Where the moat is weakest is capacity itself. If Powell cannot deliver fast enough, the buyer’s reason to choose it disappears. That is why the factory schedule matters as much as the order book.

What could go wrong

The strongest alternative reading of the evidence runs like this. Backlog jumped because one hyperscale customer placed one huge Phase 1 order. Remove it and Powell is a company growing 9% while its petrochemical and oil and gas business shrinks. If AI capital spending pulls back, the later phases could be deferred, and the first phase could be cut or slowed. On that reading the October sell off is the market getting ahead of a cyclical peak.

We take that seriously. A single customer can move this backlog by a sixth, and we do not know who the customer is. Projects of this size usually come with contract terms that let the buyer stretch the schedule, and a stretched schedule hurts a company that recognises revenue as it builds. The 49% fall in petrochemical revenue is real too. It shows how quickly one of Powell’s end markets can turn.

The second risk is margin. Mega projects carry fixed price risk, copper and steel are getting dearer, and a factory staffed with new hires in new buildings rarely runs at full productivity in its first year. If gross margin falls to 27% just as revenue is supposed to climb, our math stops working. The 100 basis points from closeouts will not repeat on demand.

Execution is the third. The 300,000 square foot lease is due to be ready late in fiscal Q2 or early in Q3 of 2027, roughly February to April. Building slips are common. Each quarter of delay pushes revenue out, and consensus already assumes the growth arrives inside fiscal 2027.

Financing is the one risk we can set aside. Powell funds itself, holds $634M of cash with no debt and earns far more than it spends. Even the $70M to $100M greenfield plant under evaluation could be paid for from cash on hand. The next test and the cash that follows it are both within reach with no new shares and no borrowing, and that removes a whole category of downside.

The investment thesis

Powell’s revenue is capped by factory output rather than demand, the cap lifts during fiscal 2027, and today’s price needs only a little less growth than consensus already expects.

The chain runs as follows. The constraint easing is Powell’s own throughput, through the Jacintoport yard due by the end of fiscal 2026, a 50,000 square foot Houston lease, an Ohio expansion and the 300,000 square foot lease. Adoption has already responded, with $934M of orders in one quarter and data centres at 40% of backlog. Powell is keeping the value, with gross margin near 30% and pricing stable. And financing leaves the gains with shareholders, since the diluted count rose 0.2% in a year and the company has no debt.

Now reverse the price. At $161.36 the market value is about $5.9B and enterprise value about $5.3B, net of cash. At 21 times earnings, that enterprise value needs about $250M of net income in fiscal 2027, or EPS near $6.80. At a 17% net margin, slightly above the 16.4% of fiscal 2025, that means roughly $1.45B of revenue. Cash adds about $17 a share on top.

Our downside case has fiscal 2027 revenue of $1.30B, growth of only 9%, and a 15% net margin. EPS lands near $5.30, and at 20 times plus cash the shares would be worth about $124. That is the outcome if the big order slips and the new space opens late.

Our central case has revenue of $1.45B, up 21%, which is below the top of consensus, with a 17% margin. EPS of about $6.75 at 21 to 22 times, plus cash, gives roughly $159 to $165. That supports today’s price. Our upside case has revenue of $1.60B at an 18% margin, EPS near $7.90 at 25 times plus cash, about $214, and it needs a Phase 2 award and the new capacity on time.

So here is the gap as we see it. The backlog already covers more than two years of revenue, so a 21% step up is a question of factory timing. The price assumes about that, and no more, while the stock trades as though the growth is in doubt. We think the market is wrong about where the risk sits: it is on the build schedule, not on demand.

We hold while the thesis holds and sell when it breaks. The next test is fiscal Q4, expected in November 2026. Strengthening would look like revenue of at least $335M, gross margin of 29.5% or more, orders of at least $350M without a new mega order, backlog of $2.4B or above, data centre revenue up 40% or more, the new facilities confirmed on schedule and fiscal 2027 growth framed at 20% or higher. Breaking would be any one of these: revenue below $310M, backlog below $2.2B, any cancellation or scope cut on the big data centre order, gross margin below 27%, or fiscal 2027 commentary that implies growth under 12%.

What we are watching

• Fiscal Q4 FY26 results, expected November 2026 with the date not yet announced: revenue of $335M or more and gross margin of 29.5% or more strengthen the case; revenue below $310M or margin below 27% breaks it.
• Backlog at September 30, 2026, reported with Q4: $2.4B or more with Q4 orders of at least $350M and no new mega order strengthens it; backlog under $2.2B, or any deferral or scope cut on the order above $400M, breaks it.
• Fiscal 2027 outlook on the Q4 call, November 2026: growth framed at 20% or more strengthens the case; language implying under 12% breaks it.
• Jacintoport yard status on the Q4 call, November 2026: confirmation that it finished by the end of fiscal 2026 strengthens the case; a slip into fiscal 2027 weakens it.
• Fiscal Q1 FY27 results, expected early February 2027: sequential revenue growth and backlog still at or above $2.4B strengthen the case; a second straight quarter of shrinking backlog with no new data centre orders weakens it.
• The 300,000 square foot facility at the February and May 2027 results: ready by fiscal Q3 strengthens the case; any slip past June 2027 breaks it.

The bottom line

Data centres have run out of patience with the grid, and the money is flowing to whoever can build their power systems fastest. Powell spent decades learning to build self contained power on offshore platforms and is now taking orders it cannot yet ship. That mismatch is why revenue grew only 9% while orders tripled.

At about $5.3B of enterprise value, the market asks Powell to turn roughly 60% of its current backlog into fiscal 2027 revenue at its current margin. The cash is there. The buildings are coming. We think the stock is priced for a slowdown the order book does not show.

What would change our mind is simple. If the order above $400M is cut or pushed out, or if backlog drops below $2.2B before the new capacity opens, then the demand we are counting on was one customer’s enthusiasm, and we would be wrong.

Not financial advice.

References

1. Powell Industries. Powell Industries Announces Third Quarter Fiscal 2026 Results. Exhibit 99.1, current report filed with the US Securities and Exchange Commission. Published August 3, 2026. Accessed October 29, 2026. https://www.sec.gov/Archives/edgar/data/0000080420/000008042026000103/ex991-powlq3xfy2026earning.htm
2. MarketBeat. Powell Industries (NASDAQ:POWL) Stock Price Down 10.9%. Here’s What Happened. MarketBeat. Published October 22, 2026. Accessed October 29, 2026. https://www.marketbeat.com/instant-alerts/price-powell-industries-nasdaq-powl-stock-price-down-109-heres-what-happened-2026-10-22/
3. AAON. AAON Reports Record Second Quarter 2026 Results. Exhibit 99.1, current report filed with the US Securities and Exchange Commission. Published August 10, 2026. Accessed October 29, 2026. https://www.sec.gov/Archives/edgar/data/0000824142/000082414226000052/liveaaonpressreleaseexs.htm
4. The Motley Fool. Powell Industries (POWL) Q3 2026 Earnings Call Transcript. The Motley Fool. Published August 10, 2026. Accessed October 29, 2026. https://www.fool.com/earnings/call-transcripts/2026/08/10/powell-industries-powl-q3-2026-earnings-call-transcript/
5. Powell Industries. Powell Industries Announces Fourth Quarter and Fiscal Year 2025 Results. Exhibit 99.1, current report filed with the US Securities and Exchange Commission. Published November 2025. Accessed October 29, 2026. https://www.sec.gov/Archives/edgar/data/80420/000008042025000145/ex991-powlq4xfy2025earning.htm
6. MarketBeat. Powell Industries (POWL) Earnings Date, Estimates and Call Transcripts. MarketBeat. Accessed October 29, 2026. https://www.marketbeat.com/stocks/NASDAQ/POWL/earnings/
7. StockAnalysis. Powell Industries (POWL) Stock Forecast and Analyst Price Targets. StockAnalysis. Accessed October 29, 2026. https://stockanalysis.com/stocks/powl/forecast/