DK Street Journal

Salesforce bought a good part of its revenue growth last quarter. Informatica, the data-integration business it acquired and which was not in the year-ago base, contributed $456m of the $11.3bn the company reported on 26 August. Back that out of the reported figure and revenue grew about 6.4% against the 10.8% headline — a deceleration from the 13.3% posted in the April quarter, not the reacceleration the following session's 22.6% jump implied.

That session is why the composition matters. Salesforce, which sells sales-pipeline, customer-service, marketing, commerce and Slack collaboration software to enterprises and has historically billed by the seat, is the largest live test of whether agentic artificial intelligence adds revenue to an applications vendor or eats the seats it is sold against. The market answered on 27 August, before the answer is in the accounts.

The new segments hide the old question

Effective this fiscal year Salesforce stopped reporting Sales, Service, Platform and Marketing & Commerce separately. Its quarterly filing now splits subscription revenue two ways: "Agentforce Apps" — the seat-billed core of Sales, Service, Marketing, Commerce and Slack — at $7.2bn, up 8% in constant currency, and "Data 360, Headless Platform and Other" at $3.6bn, up 20%. The faster of the two is where Informatica's $440m of subscription revenue landed. Arithmetic on the disclosed figures leaves the rest of that line growing in the mid single digits. Neither reported segment grew organically faster than high single digits.

The earnings line was flattered from a different direction. Of $5.90 in non-GAAP diluted earnings per share, $2.53 came from gains on strategic investments, with roughly $2.7bn of unrealized gains attributed to the Anthropic stake. Excluding them leaves about $3.37 against a $3.27 consensus — an operating beat of some 3%. Gross margin fell to 76.65% from 78.10%, so gross profit grew more slowly than revenue, and GAAP operating income was flat at $2.3bn.

What is actually working is forward

Current remaining performance obligation — contracted revenue due within a year — reached $33.5bn, up 14% in constant currency, three points ahead of revenue and what management called the fastest bookings growth in four years. Agentforce annual recurring revenue passed $1.5bn, up over 240%, though the definition was widened this quarter to include Slackbot and other AI offerings, so that growth rate is not measured against a fixed product perimeter. It is still recurring-revenue arithmetic and customer counts rather than a revenue line.

The raise tells the same story. Full-year guidance went to $46.1bn–$46.4bn; of the $300m constant-currency increase, $200m is the pending Contentful and Fin acquisitions — Fin, the customer-agent company formerly called Intercom, at about $3.6bn — and $100m organic.

The other half of the session was Anthropic. "The number one AI in the world, Anthropic, and the number one CRM, Salesforce, coming together for the first time in an incredibly powerful way to build a new product called Claudeforce," Marc Benioff, chief executive, told investors on the 26 August call. Claudeforce puts Salesforce records and workflows inside Anthropic's Claude interface and makes Claude the default model across Slack and core products. "This SaaSpocalypse narrative has been such nonsense," Benioff said on CNBC the same day.

The re-rating did the work

Trailing four-quarter gross profit rose 2.1% when the new quarter replaced the year-ago one. The shares rose 22.6%. Price-to-trailing-gross-profit went from roughly 5.0x to 6.17x in a session, the top of its own range this year and up from 4.47x in late July. It remains far below ServiceNow, the workflow-automation platform now selling a customer-relationship product of its own, at 12.06x, and Microsoft at 16.98x. The 15.8x forward earnings looks cheaper than either — but that multiple rests on a fiscal 2027 consensus that embeds this quarter's investment gains; consensus has earnings per share falling 1.6% to $15.94 the following year.

So the backlog and the bookings earn some of this. Nothing in the income statement does. Organic growth slowed, both margins compressed, and the earnings beat net of an investment mark was three cents' worth of operating performance. What the market bought on 27 August was the forward book and a distribution deal.

That deal is the part worth watching, and not only for the revenue. Claudeforce makes the salesperson's habit form in Anthropic's window, with Salesforce's data supplied to it. General availability comes in September at Dreamforce.

The wholesale price of a .com domain has not changed since September 2024. VeriSign — which operates the .com and .net registries, maintains the internet's root zone and runs two of its thirteen root servers — grew revenue anyway, on 12.7 million new registrations in the June quarter, the most in any quarter in its history.

That matters because the standard reading of this company is a toll road that raises its toll. For most of the last decade it was: VeriSign exercised every permitted 7% increase in the prior contract cycle, taking the wholesale price from $7.85 to $10.26 in successive September steps. Then the toll froze, and the meter took over.

Units, while the price sat still

The combined .com/.net base ended June at 179.1 million names, up 3.05 million in the quarter. A year earlier it stood at 170.5 million and was shrinking — down 0.1% from the prior year. That is a swing to roughly 5% annual unit growth from nothing. On average bases across the two periods, revenue per domain per quarter rose from about $2.41 to about $2.45, near 1.6%, against unit growth of about 4.4%; the arithmetic includes .net and non-domain lines, so treat it as an approximation, but three-quarters of the quarter's 6.0% revenue growth is names, not rate.

"The 12.7 million new registrations are the largest we have seen for any quarter in our history," executive chairman and chief executive D. James Bidzos told investors on the July 23 call. Management credits artificial-intelligence tools that make it fast and cheap to find a name and stand up a site, registrars redirecting spend toward customer acquisition, and its own shift to registrar-specific marketing, with strength in the United States and Europe. The company raised 2026 domain-base growth guidance to 5.2–6.0% having guided 1.5–3.5% in February.

Reported revenue growth, meanwhile, decelerated — 7.6% in the fourth quarter of 2025, then 6.6%, then 6.0% — precisely because the price contribution ran out. The renewal rate slipped to 75.2% from 75.5%, and management expects it to drift lower as more first-time names come up.

The toll resumes on November 1, when .com goes to $10.97, the first of four permitted 7% steps running to 2030. "Our own modeling… is we would expect about 50% of November's 7% price increase on .com to be recognized in 2027 revenues," chief financial officer John Calys said on the same call. Consensus has 2027 revenue up 9.8% after 5.9% this year, which is that sentence turned into a number.

Investors have already paid for it. The shares changed hands at 31.6 times trailing earnings, against 25.9x at the end of 2024 and 27.6x at the end of 2025, and 29.8x forward on 2026 earnings-per-share growth of 7.4% — flattered by a 7% cut in the share count, funded by $1.17bn of buybacks and dividends over twelve months, more than the company's free cash flow. Costs are moving too: operating expense rose 14% to $138m on server and chip inflation the company called meaningful.

The registrar pays the toll

GoDaddy, which resells those domains to small businesses alongside hosting, website builders and payments, is the opposite trade. Its shares are down 33.9% over twelve months, in two earnings-day drops — 14.3% on February 25 and 16.7% on July 31, the latter after guidance was narrowed on what Reuters attributed to slower adoption of its AI tools and weaker customer acquisition.

The business did the reverse. Second-quarter operating income rose 31.7% and operating margin went from 21.9% to 27.1%. But customers numbered 20.5 million, up 22,000 in three months, while revenue per customer rose 9% to $250. Applications and commerce grew 11% to $515m at a 46.8% segment margin; the core platform of domains and hosting grew 4% to $783m, with domain revenue up 5% to $470m and the aftermarket up 9% to $129m. The attach is real. The customer base is not growing.

Bookings say the same more sourly: up 6% overall, with applications and commerce moderating to 7% and management flagging about a percentage point of headwind from products it is deliberately winding down. Its Airo AI assistant reached a $50m annualized bookings run rate, five times the prior quarter and about 1% of revenue. "We expect the need for traditional products like do-it-for-you services and template-based website builders to narrow and evolve over time," chief executive Amanpal Bhutani said on July 30 — a concession that part of the legacy base is being disintermediated.

What each side earns

VeriSign's operating result is earned by units; its multiple expansion is a bet on a price cycle that has not yet started billing and on .web, whose launch is expected late this year or early next and whose wholesale price, unlike .com's, nothing caps. GoDaddy's margin expansion is earned in cash — 13.2% trailing free-cash-flow yield at 13.2 times forward earnings, against 30.6x at the end of 2024 — but the de-rating is a verdict on whether a flat customer count can be monetized indefinitely, and peer Wix, down 36% over twelve months, suggests the market is judging the category rather than the company. The squeeze the setup invites — registry fees eating registrar gross margin — has not happened: the $0.71 increase arrives November 1 against a GoDaddy .com renewal list price of $22.99, roughly double the wholesale cost.

So the two ends of the same wire are being priced on opposite assumptions about the same demand. VeriSign is valued as though the new names keep arriving and the toll compounds on top of them; GoDaddy as though the people registering those names never become customers worth more than the ones it already has. The Q3 registration count settles the first question before the price increase settles the second.

Two mid-market payroll software vendors reported within a day of each other in the first week of August, and both told investors to expect less money next year from the same source: the interest they earn on employers' cash while it sits with them. Paycom guided its client-fund interest to about $105m for 2026 against $113.0m earned in 2025. Paylocity, reporting a day earlier, put fiscal 2027 at roughly $103m against nearly $120m.

That line is where most readings of these businesses stop, because it looks like the rate trade unwinding. It isn't. Strip the interest out of both companies' guidance and the software underneath is growing faster than the reported number — which makes the shrinking interest line a headwind these two are absorbing rather than a crutch they are leaning on.

How the money sits

A payroll processor collects withholding from an employer on payday and remits it to taxing authorities when due. In between it holds the cash and keeps the yield. Paycom, the Oklahoma City vendor of a single-database human-capital platform sold to small and mid-sized US employers on a per-employee subscription, ran an average daily client balance of about $2.9bn in its second quarter, up 9%. Balances are rising; the interest is falling. The decline is yield, and Paycom's original 2026 outlook of roughly $103m was built on an assumption of two rate cuts before the company nudged it to about $105m on 5 August assuming rates hold.

Interest was 5.5% of Paycom's 2025 revenue and is about 4.8% of the 2026 guide. Back it out and the raised guide of $2.197bn–$2.212bn, a headline 7–8%, implies roughly 8.3% growth in everything else. Paylocity, the Schaumburg, Illinois vendor selling payroll and human-resources software to mid-market employers through a direct sales force, is the starker case: its fiscal 2027 revenue guide of $1.880bn–$1.895bn is about 7% growth, roughly 8.1% ex-interest. Its adjusted EBITDA excluding client-fund interest reached $534.9m against $459.6m, up 16.4%, while the all-in figure grew 12.3%.

The two are not equally levered to rates. Paylocity's interest equals about 31% of its $386.0m of operating income; at Paycom the share is roughly 12%. And the mature comparison runs the other way: at Automatic Data Processing, the $113.8bn payroll and human-resources outsourcer, interest on client funds rose 13.7% to $403.9m in the March quarter as balances grew 8.5% to $48.3bn and the average rate earned rose to 3.3%. Its laddered bond book is still rolling into higher coupons while the smaller vendors' shorter money reprices down.

The meter itself is flat

These contracts bill per employee per month, so headcount is the meter — and the meter is not moving. ADP's pays-per-control, the employees on client payrolls it bills for, grew 1% in fiscal 2026 and is guided to 0–1%. Paylocity chief financial officer Ryan Glenn told investors on the 4 August call that the company is assuming flat client workforce levels in fiscal 2027, "a slight degradation from recent trends". Paycom management described client employment growth as stable with no acceleration.

Growth is therefore coming from somewhere else. Paylocity ended the year with about 44,400 clients, up 7%, and revenue per client of roughly $37,200, up more than 5% — new logos, price and module attach, including the spend-management business it bought in Airbase. Recurring revenue accelerated to 12.4% in the fourth quarter.

Paycom's answer is its own cost line. Second-quarter revenue of $531.2m grew 9.8% while operating income of $168.5m grew 50.0%, taking margin from 23.2% to 31.7%. A 2025 data-center build now saves more than $100m a year in research spend and over $30m in third-party artificial-intelligence token fees; more than 500 roles went on 1 October 2025, concentrated in support and implementation, leaving 5,770 employees. Chief executive Chad Richison put it plainly on the 5 August call: "So our focus is product automation and that drives cost efficiencies in many areas, including labor." Diluted shares fell to 45.9m from 56.3m after $1.4bn of buybacks year to date.

What the shares have and have not earned

Paycom is up 74.4% over three months to $238.53 but only 5.7% over twelve, and a single session did the work: a 23.8% gap on 6 August, on 3.65m shares against a daily norm nearer 500,000, the day after the raise. Paylocity is up 42.4% over three months with no comparable session and remains down 12.0% on the year. ADP rose 30.3% and Paychex 33.6% over the same three months, both still lower year on year. All four were in downtrends in May; by late August Paycom and ADP had their 50-day averages above their 200-day, while Paylocity and Paychex had only flattened.

On valuation the pair sit below the incumbents. Paycom trades at 19.7 times forward earnings against 25.3 times trailing, on consensus earnings growth of 30.9% that is mostly the vanishing share count. Paylocity is the cheapest of the four at 17.9 times forward against 31.6 times trailing, but on forward earnings growth of only 9.0%. ADP is at 23.2 times forward and Paychex 21.3 times.

The verdict splits the pair. Paycom has earned the profit half of its move and none of the revenue half: growth has decelerated three years running, from 23.2% in 2023 to 8.9% in 2025, and the 2025 operating margin of 27.6% was below 2024's, so this year's surge is a recovery rather than a new peak. Its retention, at 91% annually, is the lowest of the group against Paylocity above 92% and ADP's record 92.1%. Paylocity has earned less on margin and more on the top line, but carries nearly triple Paycom's dependence on rates.

Both companies have now told investors that hiring will not help them. Both assume the Federal Reserve cuts twice. What is left is price, modules and the vendors' own payrolls — and a cost base can only be cut once.

INNIO, the Munich maker of Jenbacher and Waukesha gas engines carved out of General Electric, told investors on July 28 that its equipment order intake had more than quadrupled from a year earlier, to $2.3bn. Its shares fell 17.0% the next session.

What the market read was not the order line but the margin underneath it. Engines and fuel cells are being bought at all because heavy-duty gas turbine slots are sold out into the end of the decade, and buyers are paying for delivery dates. The question that separates these three suppliers is whether a megawatt sold today is paid for on shipment, on a service contract years later, or not really paid for at all.

INNIO is buying its backlog

The orders are genuinely incremental. Data-center equipment order intake went from $27m in 2023 to $2.28bn in 2025, which is not a relabelled European cogeneration replacement cycle. Backlog reached $6.6bn, up 279%, including a 1.1 GW behind-the-meter prime-power order. "Combining our equipment order backlog with our slot reservations, we have more than 15 GW of committed business as of Q2 2026," chief executive Olaf Berlien said on the July 28 call; roughly 64% of it is data-center work.

The cost shows up one line down. Equipment adjusted segment margin fell from 18.7% a year ago to 13.8%, management attributing the drop to self-funded capacity and ramp-up cost, while Services earned 29.8% on 39% of the quarter's $937.7m of revenue. Every incremental data-center megawatt currently dilutes group profitability. This is not balance-sheet stress — net leverage improved to 2.7x from 3.6x at end-2025 — and consensus rebuilds EBITDA margin to 21.0% next year on the assumption that ramp costs anniversary. Q2 put that assumption in question.

Wärtsilä is paid, but not yet for this

Wärtsilä, the Finnish engine and power-plant group, is the only one of the three whose profit is already annuitized: a €9.0bn order book, a record €2.6bn combined service order book, and a rolling 14% comparable operating margin that met its own financial target — "which is our financial target," chief financial officer Arjen Berends told investors on the July 21 call. Second-quarter orders rose 33% to €2.8bn, with Energy up 82% on 1.2 GW of firm data-center orders across two projects. "The gross margin of the energy equipment order book has improved by more than 500 basis points," chief executive Håkan Agnevall said on the same call.

Reported sales, meanwhile, fell 9.3% in the quarter, because eleven divested business units dropped out of the base, and management flags roughly four years — deployment plus ramp — before engines shipped now generate material service revenue. Half-year service order intake actually fell 4%. The annuity is real and flat; the growth is booked and unshipped.

Bloom converts, expensively

Bloom Energy, whose solid-oxide fuel cells convert gas to electricity without combustion, is the only one whose margins moved with its volumes: revenue up 166% to $1.065bn, product gross margin of 37.2%, operating income up 737%, and full-year revenue guidance raised to $3.9–4.2bn. Its service segment, historically dilutive, earned 22%. The offsetting entry is the stack-replacement obligation: warranty and product-performance liabilities built $71.6m in the first half against $13.8m of spending.

Pricing is where they part. Bloom trades at 48.8 times forward gross profit against INNIO's 12.0 times — roughly four times as much for a dollar of gross profit, at gross margins within a point of each other. Wärtsilä, whose gross-profit reporting is not comparable, is the cheapest on every other measure: 25.9 times trailing and 25.8 times forward earnings, 14.7 times EV/EBITDA, and a 7.7% trailing free-cash-flow yield. A forward multiple equal to the trailing one means the market has priced in essentially no earnings growth.

What the shares did

Between August 17 and 28 the whole listed data-center power complex sold off together — GE Vernova down 15.5%, Cummins 11.7%, Bloom 7.6% — while INNIO fell 25.9%, to $20.83, some 22.9% below its June IPO price of $27.00. No company-specific INNIO disclosure was discoverable in that window; the likelier reading is a sector de-risking that hit the least-seasoned name, one still 86.2% owned by an Advent- and ADIA-led group, hardest. Wärtsilä is down 17.5% over three months and up 17.9% over a year.

The verdict divides cleanly. INNIO's de-rating is earned by its own accounts: the orders are real, and so is the five-point margin cost of winning them. Wärtsilä's is not earned by anything in its P&L except a divestiture it disclosed in advance. Bloom's decline from May is the one case where the business improved and the price still had further to fall, because 48.8 times forward gross profit was pricing more than a doubling of revenue.

Wärtsilä says the engines it ships to a data center today start paying service revenue about four years out. Whoever is still buying megawatt scarcity then will find out which of these three was selling a product and which was selling a delivery slot.

The jet-engine aftermarket boom is not reaching the companies that own pieces of the engines. IHI, the Tokyo-listed heavy-industry group that holds fixed program shares in the IAE V2500, Pratt & Whitney's PW1100G geared turbofan and several GE Aerospace engines, lifted civil-engine revenue 27% to ¥109.4bn in the June quarter; operating profit at the aerospace, space and defense segment that houses it rose from ¥27.9bn to ¥29.1bn, a gain of 4%.

The gap is mechanical. A risk-and-revenue-sharing partner buys a fixed percentage of a program and is paid that percentage on everything the program sells — high-margin spare parts and newly built engines shipped on thin build economics alike. Departures were flat across the first half of 2026 while Safran's civil spare-parts sales rose 27.9% in dollars, and that deeper-repair cycle does flow through to partners automatically, with no customer relationship required. So does the opposite: LEAP deliveries hit a record 1,802 units in 2025 and rose 41% again in the first half of 2026. Owning both sides of that mix is what capped IHI's segment margin and pushed Kawasaki's aerospace profit down.

Paid without a customer, charged the same way

IHI holds workshare on GE's CF34, GE90, GEnx and Passport 20 alongside the V2500 and PW1100G — partner positions across all three Western propulsion primes. The V2500 and PW1100G aftermarket operations alone added ¥4.3bn to the quarter's profit, with spare parts up 15% in dollar terms before currency effects. Group operating profit rose 250.5% to ¥73.2bn and full-year guidance went up to ¥1.84trn of revenue and ¥250bn of operating profit, helped by a roughly ¥40bn property gain. "Operating profit reached a record high even excluding the real estate sale," chief financial officer Hiromi Oshima said on the August 5 call.

The same automatic share is why IHI's accounts once looked like this: in the year to March 2024, gross margin of 10.95%, a ¥70.1bn operating loss and a ¥68.2bn net loss. IHI produces the PW1100G high-pressure rotor, and when the powder-metal contamination defect surfaced it guided to an ¥80bn operating loss in place of a profit forecast. Gross margin has since recovered to 23.09%. A separate unquantified item sits open: subsidiary IHI Aerospace drew a five-month JAXA bidding suspension on June 2 over falsely reported costs.

Kawasaki's good quarter came from everything else

Kawasaki Heavy Industries — military airframes for Japan's defense ministry, rolling stock, robots, motorcycles — holds comparable partner positions, and they were the drag. Aerospace Systems profit fell on "a temporary decline in revenue" and "increased shipments of newly manufactured commercial aircraft engines". The record ¥35.7bn of quarterly business profit and the raise of full-year guidance to ¥180bn came from precision machinery, rolling stock and energy, plus tariff refunds and a weaker yen; powersports guidance was cut. Whatever Kawasaki is at 4.84x trailing gross profit — the cheapest of the three — it is not an aero-propulsion story this year.

Kratos is buying the installed base it doesn't have

Kratos Defense & Security Solutions builds low-cost expendable turbines and unmanned aircraft for the Pentagon, and it is the counter-case: revenue up 30.5% to $458.8m, gross margin down to 21.8%, and a GAAP operating loss of $1.6m. Guidance calls for $125–135m of capital spending and $85–105m of free cash flow use this year. "We are currently placing initial orders with our supply chain for the components for 3,000 small Kratos TDI Spartan turbojet engines," chief executive Eric DeMarco told investors on August 4, at an average selling price near $50,000. The GEK800 turbofan won a military type designation and a development contract as second-source propulsion for the Joint Air-to-Surface Standoff Missile on August 18. The shares fell 15.5% in the eight sessions after. Kratos trades at 29.1x trailing gross profit, roughly double the 14.95x paid for the unmanned-aircraft maker AeroVironment, and it is the only one of the three with no fleet to be paid from.

What the shares did

All three fell with the defense complex between August 14 and August 28 — L3Harris down 9.9%, Kratos 18.9%, IHI 10.2%, Kawasaki 8.2%, with AeroVironment the worst at 22.7%. The Japanese names' weakest session was August 19, when the Nikkei 225 fell 3.16% on a broad technology selloff. IHI is 36.8% below its February peak and trades at 15.97x forward earnings against consensus expecting earnings per share up 43.9% in the year to March 2027; Kratos is 43.2% below its own February high at 62x forward earnings.

The verdict the disclosures support is narrow. IHI's aftermarket annuity is real and arrives without effort — but the delivery ramp that is the primes' best news is a margin dilutant one tier down, and the same sentence that trimmed IHI's segment growth turned Kawasaki's aerospace profit negative. Neither Japanese name's August decline is explained by anything either company disclosed.

Being paid automatically means being charged automatically too. When Pratt & Whitney's powder-metal problem surfaced, IHI's share of the remedy arrived without anyone asking. Kratos does not have that exposure yet, for the same reason it has no annuity: there is no fleet out there carrying its name.

Bharat Heavy Electricals' power division — coal boilers, steam-turbine islands and hydro packages — earned ₹562.81 crore before interest and tax in the June quarter, against a loss of ₹510 crore in the same quarter a year earlier. Power revenue rose 51.8% to ₹5,919.50 crore, and the group as a whole swung to a consolidated net profit of ₹376.71 crore from a ₹455.50 crore loss, Business Standard reported.

What produced that swing is a coal cycle, not a gas one. BHEL closed the March year with an order book of roughly ₹2.4 lakh crore — about $27bn, and more than seven times FY26 revenue — split 81% Power, 18% Industry and 1% Export, after ₹75,916 crore of inflows during the year. Roughly ₹59,000 crore of that intake was power work, dominated by supercritical coal boiler-turbine-generator packages: 3x800 MW at Telangana Stage-II, 2x800 MW for Adani at Raipur, 1x800 MW for DVC at Durgapur, 2x660 MW for NTPC at Talcher. India's power ministry has planned an additional 97 GW of coal and lignite capacity, taking thermal installed capacity to 307 GW by 2034-35. The binding question for the equity is no longer order intake. It is conversion.

The licensee

BHEL's heavy-duty gas capability is borrowed. It builds advanced-class frames under a technical assistance and licence agreement with GE dating to 1986, and its most defensible gas asset is aftermarket service on more than 190 installed machines through BGGTS, its joint venture with GE running since 1997. The disclosed frame wins are historical and modest: the largest, a 726.6 MW combined-cycle plant for ONGC Tripura Power, was worth about ₹2,200 crore. Against a ₹2.4 lakh crore book that is a rounding error, and no sizeable new combined-cycle or data-center award was discoverable this year. The gas supercycle that has occupied this page — GE Vernova's $176bn backlog, Siemens Energy's 14.2% quarterly margin — passes BHEL by.

Conversion is working; cash has not arrived

The execution evidence is good. FY26 revenue rose 23.5% to ₹33,782 crore, net income tripled to ₹1,600 crore from ₹534 crore, and operating margin doubled to 6.0% from 3.0%. Net working capital fell to 164 days from 241, which sell-side coverage paired with warnings on valuation, execution capacity and whether the margin gain holds. One line is going the other way: the Industry segment grew revenue 12% in the June quarter while its segment result fell 20.9% to ₹242.61 crore.

The shares have more than doubled over twelve months and are up 63.8% in six. They trade at 61.6x trailing and 45.5x forward earnings and 35.1x trailing EV/EBITDA, on a trailing free-cash-flow yield of minus 0.94% — reported profit has not yet become cash. Consensus makes the arithmetic explicit: net income near ₹3,291 crore in FY27 and ₹4,679 crore in FY28, earnings per share going from ₹4.60 to ₹13.45 in two years.

The contrast next door

Hindustan Aeronautics, India's sole-source military airframe integrator, rose 25.2% over the same six months on a constraint that is physical rather than financial: only about six of 99 contracted GE F404 engines had arrived against 30 Tejas Mk1A airframes built, flown and tested but undeliverable, until the first jet was handed to the Air Force this month. Management has guided to 10-12% revenue growth in FY27. At ₹4,899.00 against FY26 earnings of ₹136.30 a share it trades near 36x — about where it traded a year ago. That advance was earnings.

BHEL's was not. The direction of the business is real and coal-driven, and the working-capital improvement falsifies the standard complaint that its backlog growth is being eaten by receivables. What the June quarter does not underwrite is a price that already assumes several years of flawless execution on fixed-price legacy contracts, at a company whose gross margin last year was 11.9%. Two-thirds of the equity story rests on profits nobody has yet booked.

The boilers now in BHEL's order book will burn coal into the 2040s. Whether the ₹2.4 lakh crore behind them turns into cash is a question the next four quarters answer, and the price has already voted.

Harbin Electric, which builds boilers, steam turbines, hydro sets and nuclear island components for Chinese power stations, sold RMB873m of gas-fired equipment last year. That was 1.9% of group revenue of RMB45.70bn, and it fell 55.3% from the prior year. Over the same period the shares behaved as though the company were levered to the global scramble for gas turbines: up more than double over twelve months, then down 34% in six.

The stake is what investors thought they owned. Artificial-intelligence data centers have made heavy-duty gas turbines the scarcest capital good in power, and anything with a turbine attached was marked up through 2025. In China the money is somewhere else entirely, and the sell-off since March has not distinguished between a company whose earnings are compounding on coal and nuclear and companies whose earnings have stalled.

What the Chinese order books actually contain

Harbin is the only one of the three that breaks gas out. Coal-fired equipment was RMB18.53bn of 2025 revenue, 40.5% of the group and up 16.2%; nuclear grew 3.5%, hydro fell 8.1%. The gas line had been RMB1.95bn in 2024, so 2025 more than halved something already small. New power-equipment order intake rose 26% to RMB44.0bn, with coal orders up 39.4% and nuclear up 81.6%.

That mix follows the country's construction schedule. China commissioned a record 22.4 GW of gas-fired capacity in 2025, a third of the world's total, yet gas is still only about 4.5% of the national fleet. Meanwhile China is due to commission 85 of the 104 coal units starting up worldwide in 2026, roughly 55 GW, after 30 GW came online in the first half alone.

Shanghai Electric's RMB172.8bn of 2025 new orders itemizes coal at RMB26.59bn, wind at RMB22.97bn, storage and nuclear — and no gas line at all. Elevators, motors and intelligent manufacturing took RMB44.48bn, a quarter of the book. Dongfang buries turbines inside an "energy equipment manufacturing" segment that was RMB58.0bn, or 73.8%, of 2025 revenue.

The technology is borrowed, and one route just closed

Dongfang has been a Mitsubishi licensee since 2003; the Japanese company marked cumulative orders of 150 units through Dongfang Turbine in October 2024 and runs a Guangzhou aftermarket joint venture with it. Its indigenous G50 is a 50MW F-class machine with about ten units booked. At the annual meeting in Chengdu on 29 May, Dongfang disclosed an export agreement signed 30 April with a North American customer for a first phase of ten G50s, and a target of 20 to 30 indigenous turbines a year by early 2028. Management attributed the group's thin margin to fierce wind-power competition and described gas turbines as a business where the long money is earned in service.

Scale the ambition. "We almost doubled the number of gas turbines sold globally, from 100 units in 2024 to 194 units in 2025," Siemens Energy chief executive Christian Bruch said of his own company. Shanghai Electric's window into advanced frames was its 40% stake in Ansaldo Energia, a former Siemens licensee; Ansaldo is now 99.6% owned by Italy's state investor CDP Equity.

Which de-rating is earned

Harbin's numbers went the other way from its share price: first-half revenue up 10.1% to RMB25.0bn and attributable profit up 63.5% to RMB1.718bn, gross margin 15.66% against 11.91%. It trades at 9.6x trailing and 10.1x forward earnings, 1.69x book, with a 19.2% trailing free-cash-flow yield; GE Vernova trades at 25.9x trailing. Huatai keeps a buy with a HK$24.94 target, about 13.6x estimated 2026 earnings.

Dongfang's case is different. First-half revenue grew 1.24% while profit rose 42%; second-quarter gross profit fell 6.4% and margin slipped to 14.95%. Consensus still models full-year revenue growth of 11.3%, which needs about 22% growth in the second half. The stock is down 38% from its March peak and still near 18.6x trailing earnings, roughly double Harbin's. Shanghai Electric guided first-half profit up 12% to 22% and credited government subsidies and gains on disposing of subsidiary equity; at a net margin near 1%, its 80x earnings multiple says little, and on price to gross profit it is no cheaper than Harbin.

The move was not orderly. Harbin rose 23% on 21 July and fell 9% on 24 August; one Hong Kong session saw power-equipment shares lead declines on the view that higher oil and gas prices would suppress turbine demand, with Dongfang's H-shares off almost 10%. GE Vernova, Siemens Energy and Mitsubishi Heavy Industries did not participate in the de-rating, which makes this a China-specific repricing of a China-specific story.

Two of the three de-ratings are earned by the accounts: Dongfang's top line has stopped growing and Shanghai Electric's profit is partly borrowed from disposals. Harbin's is not — but the reason is uncomfortable for bulls too. What came out of the price was an artificial-intelligence gas narrative that never appeared in its revenue. What is left is a coal and nuclear equipment business growing faster than the multiple implies, priced as though Chinese coal boilers have no future in a year when China is building most of the world's new ones.

Doosan Enerbility is the fifth company in the world to independently commercialize a large gas turbine, and the only one outside GE Vernova, Siemens Energy and Mitsubishi Heavy Industries selling a frame in the H-class — the largest and hottest-running machines utilities buy. Open its first-half accounts, though, and most of the gross profit comes from skid-steer loaders built in North Dakota.

That matters because of what the shares are priced against. Doosan Enerbility trades at 20.1x trailing gross profit, between GE Vernova's 29.1x and Siemens Energy's 14.7x, and reads as a reasonable price for a challenger in a market where the incumbents are sold out into 2029 and turbine prices are heading toward roughly $600 per kilowatt by end-2027, about 195% above 2019 levels on Wood Mackenzie's numbers. The multiple is only reasonable because a construction-equipment subsidiary is padding the denominator.

What is actually in the profit pool

Doosan Enerbility still owns 46.06% of Doosan Bobcat and consolidates all of it — the 2024 plan to hive the stake off and fold Bobcat into Doosan Robotics collapsed in December that year after proxy adviser ISS told shareholders to oppose it. Bobcat reported first-half revenue of $3.16bn and gross profit of $777.9m at a 24.6% gross margin. Doosan Enerbility's consolidated first half was ₩8.99tn of revenue and ₩1.55tn of gross profit, a 17.3% blended margin. Converted at this year's dollar-won range of roughly ₩1,428 to ₩1,559, Bobcat is about half the revenue and roughly 70 to 74% of the gross profit — which also means the power businesses run a materially thinner margin than the loaders do.

The turbine business is real, and one part of the challenger case has already been overtaken. Doosan's H-class order book is no longer domestic: it now holds contracts for twelve gas turbines with US customers, including seven 380 MW units agreed on March 6 for monthly delivery from May 2029, with the buyer undisclosed and industry sources pointing to Elon Musk's xAI. What is not there is the supplier-to-the-oligopoly trade: the named customers for its ultra-large forgings are China's Laiyang units 5 and 6 and NuScale's small modular reactor program, with no disclosed forging sales into the three incumbents' supply chains.

The businesses got better while the shares halved

First-half operating profit rose 32.4% to ₩547.8bn, orders rose 89.6% to ₩7.12tn and backlog reached a record ₩26.35tn. The stock fell 55.9% from its May 7 close of ₩136,400 to ₩60,200 on July 30, then rebounded 46.5%. Over the same slide GE Vernova lost 13.9%, Mitsubishi Heavy 17.4% and Siemens Energy 25.2%, and none bounced more than 11%. Eugene Investment & Securities cut its target 26.7% to ₩110,000 in late July while keeping a buy rating; its analyst, quoted July 28, said that "given Doosan Enerbility's solid earnings and order momentum, U.S. investment in the second half, along with expanded supply of SMR and nuclear equipment, will drive the share price higher." The cut was attributed to peer de-ratings, not to anything at Doosan.

The cause sits offshore of the turbine cycle. Korean stocks fell 5.98% on July 29 and triggered consecutive trading halts, and the index went from bear to bull market inside a month on the Samsung Electronics and SK Hynix artificial-intelligence trade. Leverage went out and came back.

Hanwha Aerospace was never the turbine bet

Hanwha Aerospace makes engine components for aircraft and marine vessels, and it traced the same 47.3% fall and 42.8% rebound. Its aero-engine division booked ₩760bn of second-quarter revenue and ₩37bn of operating profit, against ₩9,292.9bn and ₩1,365.5bn for the group; ground defense alone delivered ₩2,107.5bn of revenue and ₩533bn of profit on K9 howitzers, the Cheonho anti-aircraft gun and shipments to Poland, Egypt and the Middle East, with backlog at ₩38.3tn after an April contract with Finland. The independent gas-turbine service business PSM belongs to affiliate Hanwha Impact, which unified it with Hanwha Power Systems under the Hanwha Power brand in March; Hanwha Aerospace shareholders own none of it.

At 23.8x forward earnings and 9.4x forward gross profit it is the cheapest of these four names, on 21 to 22% return on equity. The reason is visible: Korea's defense export orders ran near $8bn by end-July against $15.4bn for all of 2025, the big four contractors' combined backlog slipped sequentially to ₩98.5tn, and consensus models a 42% earnings decline in 2029.

The verdict

Neither drawdown was earned by the operating businesses, both of which posted record results through it. But the recovery should not be read as vindication of a turbine thesis either. Doosan's first US frames do not ship until 2029, its 63.5x trailing enterprise value to EBITDA rests on a 0.5% net margin, and the gross profit supporting its multiple is mostly compact equipment sold to American contractors — a cyclical business with no connection to data-center power. Hanwha's cheapness is defense cheapness, priced against an export pipeline that has halved.

The useful test is narrow. If Doosan's power segments cannot lift their margin above the loaders', then the challenger premium is being paid on a subsidiary the company twice tried to give away.

Baker Hughes shipped no more equipment and services out of its turbine business in the June quarter than a year earlier, and made a sixth more money doing it. Industrial & Energy Technology revenue was $3.29bn, flat year on year, while segment earnings before interest, tax, depreciation and amortization rose $93m to $678m — attributed in the company's second-quarter release to price, productivity, cost-out and currency, partly offset by lower volume and inflation.

That is the answer to the question hanging over the middle rung of the turbine economy — the aeroderivative units and packaged power islands bought by liquefied natural gas trains, pipelines and, newly, data centers that cannot wait for a heavy-duty frame slot. The worry was that equipment sold at fixed prices into rising casting and alloy costs would be the loss leader, with the profit sitting in the installed-base annuity of parts and overhauls. Baker Hughes' own backlog says both legs are paying. Contracted Gas Technology Services work stood at $16.7bn against $15.0bn for Gas Technology Equipment; the annuity book is now the bigger one. And the equipment leg is not being squeezed, because supply is the binding constraint. Wood Mackenzie expects turbine prices to reach $600 per kilowatt by the end of 2027, up 195% since 2019, and a turbine ordered today on average does not begin operating until 2031.

The order book is running years ahead of the revenue

Baker Hughes' quarterly IET orders doubled to a record $7.1bn, taking remaining performance obligations to $40.1bn with backlog up 19%. Power Systems orders were $2.6bn covering 2.7 gigawatts of generation, of which management put $2.2bn down to data centers; LNG equipment added $1.8bn across three projects. Chief executive Lorenzo Simonelli told the Bernstein conference on May 27 that replacement lead times have stretched to three-to-five years and that off-grid AI power is pulling demand toward modular 150-to-300 megawatt gas-turbine blocks, committing to more than $40bn of IET orders across 2026 to 2028. The company plans to double turbine capacity by 2028 and targets roughly $5bn of annualized Power Systems revenue at full utilization, against about $1bn in 2025.

The clocks differ, and management said so: services growth moderates in the second half on planned outages, and equipment booked this year converts to revenue materially only after 2027. Meanwhile the shares are down 4.8% over six months, and Baker Hughes trades at 9.5 times trailing gross profit and 13.2 times EBITDA — about two-thirds of Caterpillar's 14.6 times gross profit.

Caterpillar priced through the tariff bill

Caterpillar, which sells reciprocating gensets and Solar Turbines industrial units alongside its machines, reported Power & Energy sales of $8.24bn, up 17% at a 24.6% segment margin, 250 basis points wider, with power generation up 72% and total backlog at $72bn — close to double a year earlier. It did that while absorbing $400m of tariff cost in the quarter against a full-year headwind of roughly $2.2bn, offset in part by $392m of recoveries. Chief executive Joe Creed said on the August 4 call that customer enthusiasm for AI-era power showed "no one is slowing down", with turbine orders extending into 2029 and 2030. The company is restarting a 10-megawatt medium-speed engine platform mothballed in 2022, 1.5 gigawatts of it, with shipments from the fourth quarter.

The de-rating came from elsewhere. On July 29 Baird cut Caterpillar to Neutral and its target to $900 from $1,200, citing New York State's moratorium on data-center construction and a wave of state-level bills, while saying near-term fundamentals remain robust. The shares fell 6.9% that session; GE Vernova and Cummins fell with them. Caterpillar trades at 34.3 times trailing and 29.4 times forward earnings, against roughly 16 times over the past decade — the anchor David Giroux used in Barron's midyear roundtable. Solar Turbines itself is not broken out, so the fastest-growing turbine franchise in the group has no visible split between machines and service.

Sulzer sells the aftermarket without the machine

Sulzer, the Winterthur pump, mixer and separation group, is the independent servicer in the picture. Its Services division grew sales 4.4% and lifted its EBITDA margin to 17.7% from 16.7%, against 13.3% at Flow Equipment, and booked new work at a 40.6% order-intake margin while the group's slipped to 35.7%. It has raised capacity at its Houston gas turbine service shop by 20%, with skilled labor the limit. But that US turbine-service line is about CHF 150m of a CHF 3.55bn group — an option on the annuity rather than the engine of it. The drag is Chemtech, where order intake fell 22.7%; executive chair Suzanne Thoma said on the July 28 call that for the core of that business, "that should be, according to everything that we know, the bottom." Sulzer is the cheapest of the three at 17.8 times trailing and 16.5 times forward earnings, 10.2 times EBITDA.

What the businesses earn and what they do not

The annuity thesis is half right and the useful half is the other one: these companies are being paid for the queue. Flat volume producing 16% more segment profit at Baker Hughes, and a wider power margin at Caterpillar after a $400m tariff bill, are both pricing outcomes, and pricing holds only while lead times do. Caterpillar's slide is a multiple compressing off a decade-high reading on a dated worry about the 2027-28 order rate, with nothing in the reported numbers to match it. Baker Hughes' six-month drift has no counterpart at all in its disclosures. Sulzer's gain was margin: group order intake fell 3.9%.

The orders that decide 2028 revenue are being written now, while states argue about whether the data centers get built. A turbine bought in that argument still does not run until 2031.

The world's airlines are not flying much more this year than last, and the companies that repair their engines just reported the strongest aftermarket numbers of the cycle. GE Aerospace, which builds and maintains jet engines through its Commercial Engines & Services arm, told investors that departures were flat across the first half of 2026 and that it expects only modest growth in the second. Spare-parts revenue grew 25% regardless.

That gap defines the season. The maintenance annuity — decades of shop visits and parts sales on an installed base chief executive Larry Culp described to investors as "80,000 engines and growing, and more than 2.3 billion flight hours" — is supposed to be geared to flying. This year it is not. The airline industry's trade body expects global passenger demand to grow 2.1% in 2026, a sharp deceleration from 5.3% in 2025. What is driving reported growth instead is heavier work per shop visit, price, and — newly — the material availability to finish the heavy jobs. All three convert into billed cash. None compounds the way traffic does, and each carries a date on which it stops.

Safran got the parts it needed

Safran, GE's equal partner in CFM International and the clearest read on the narrowbody fleet, reported civil engine spare-parts sales up 27.9% in dollars and civil engine services up 40.4%, the latter mostly on LEAP rate-per-flight-hour contracts. Propulsion recurring operating income reached €2,253m, up 28%, at a 24.5% margin — above the 22-24% the company guides to for the year. "Spare part sales for civil engines increased by 28% in dollar value. And civil engine services also grew by more than 40%," chief executive Olivier Andriès told investors on July 28. Asked how long the heavy-workscope mix lasts, he said Safran is "confident that the trend that we see now… is going to continue for at least the 2 next years in 2027 and 2028."

The honest caveat is that the record margin came from material availability unlocking scopes that were rationed in 2025, plus an August 2025 price increase. And half the narrowbody book has not yet inflected: LEAP-1B profit recognition on flight-hour contracts is still gated on the Maverick turbine blade, expected late 2026 or early 2027. Safran trades near 33 times forward earnings against consensus for 26% earnings growth this year — the least expensive of the group per point of near-term growth.

GE's mix went the wrong way

GE's services revenue rose 26% and internal LEAP shop visits 50%, with roughly $170bn of commercial services backlog behind it. But group gross margin compressed nearly three points to 35.0%, and the engines-and-services margin fell 1.6 points to 27.3% on original-equipment-heavy mix and LEAP still earning below the portfolio average — parity only "by the time we get to 2028," chief financial officer Rahul Ghai said. Ghai also disclosed that spare-parts shipments delayed by material shortages grew 20% sequentially. GE raised guidance across the board and the shares fell 4% on the print, the third consecutive post-earnings decline. It asks roughly 44 times guided 2026 adjusted earnings of $7.65-7.85, above Rolls-Royce near 35 times and Safran near 33.

Rolls-Royce raised on accounting timing

Rolls-Royce, the widebody specialist with no narrowbody engine, grew Civil Aerospace revenue 29% organically to £6.19bn at a 25.3% margin, with large-engine flying hours at 113% of 2019 — and trimmed its full-year flying-hours outlook toward the lower end of 115-120%. The profit raise to £4.7-4.9bn was explicitly driven by higher long-term service-agreement margins and an increased level of contract catch-ups: £372m of catch-ups plus £497m of contractual margin improvement, against £1.57bn of divisional profit, with the company stating the level is not sustainable into the second half. Meanwhile upgraded Trent 1000 turbine blades more than double time on wing — richer visits, fewer of them. The shares are up 44.6% over twelve months.

Pratt & Whitney bought volume back

RTX ran the same trade in reverse: Pratt & Whitney's commercial aftermarket grew 25% while commercial original-equipment sales fell 8%, a deliberate decision to send material to the repair network. Grounded A320neos are down almost 40% from peak and full geared-turbofan recovery is targeted through 2030; the powder-metal remediation still requires repetitive disk inspections every 2,800-3,800 cycles. RTX is the cheapest of the four at 29 times forward earnings against 37 times trailing, with a 4.2% free-cash-flow yield.

The one that could not ship

Honeywell Aerospace makes auxiliary power units, avionics and business-jet engines rather than large commercial turbofans, and it supplied this sector's only violent session: from $203.64 to a $156.47 close on August 6, a 23.2% fall on 20.6m shares against a norm near 4m, after cutting organic growth guidance to 4-5% from 7-9%. Demand was fine — commercial aftermarket still grew 8% — and the constraint was castings, forgings and bearings from about 2% of its suppliers. "One component coming from this one supplier at being about $15 million of past due actually unlocks hundreds of millions of dollars of revenue output," chief executive Jim Currier told investors. At roughly 21 times its cut guidance, it trades at less than half GE's multiple.

Over thirty days the four large-engine names went nearly nowhere — Rolls-Royce up 10.9%, Safran up 1.3%, RTX and GE down about 2% each — while Honeywell Aerospace fell 20.9%, so the group's decline is one name. All four did drop together over three sessions in mid-August with no company news discoverable; the likelier reading is sector-wide repricing.

The verdict is that the annuity is real and converting into billed work at all four engine makers, but it is being paid for depth and price on a fleet that is not expanding its hours. Safran and RTX are earning their advances: one on a margin that beat its own guidance, the other on a recovery still visible in grounded-aircraft counts. GE is asking the highest price of the group while its services margin compresses. Rolls-Royce, the best performer, carries the largest slug of profit that its own release attributes to contract timing.

The engines will get repaired either way — more than 95% of GE's third-quarter spare-parts revenue was already sitting in backlog when it reported. The open question is who books the takeoffs that create the next round of work.