DK Street Journal

Omnicell told investors on July 30 that its sales pipeline was the largest it had seen in years, and in the same breath lowered the target for the recurring revenue that pipeline is supposed to turn into. The shares fell 10.4% that day, from $41.40 to $37.11.

That gap — a full funnel, a cut forecast — is the shape of the whole hospital-software question this year. Three vendors sell into the same buildings on three incompatible meters: a capital equipment budget, a pharmaceutical marketing budget, and an annual subscription line. What governs all three is the same balance sheet. Median hospital operating margin was -0.6% in January 2026 and only -0.3% year-to-date through February, with drug costs up 7.6% and supply costs up 7.8%, according to Healthcare Financial Management Association data. A Premier analysis cited by the same association projects the One Big Beautiful Bill Act will hit hospital revenue by an estimated $68.6bn across 2026 and 2027.

The capital meter

Omnicell, which sells automated medication-dispensing cabinets, central pharmacy robotics and 340B pharmacy programs to health systems, still books 56% of revenue as product: $175m of the June quarter's $312.2m, against $137m of service. Revenue grew 7.4%, gross margin reached 49.0% from 43.9%, and operating income went from $8.1m to $32.4m. Roughly a fifth of the quarter's non-GAAP EBITDA was a $15m tariff refund.

The forward meters moved the other way. Full-year product bookings guidance was widened to $425-560m, stretching the bottom end well below what was previously implied. Year-end annual recurring revenue guidance was cut to $660-680m from $680-700m, citing consumables opportunities taking longer to develop; against $636m at the end of 2025, that takes implied growth to roughly 4-7%. "Our pipeline exiting Q2 2026 is meaningfully larger than we have seen in recent years," chief financial officer H. Baird Radford told investors on the July 30 call. "However, our revised product bookings guidance range reflects our current view on the potential range of outcomes and timing for 2026 purchasing decisions."

The mechanism is specific. Management said this is the first time Omnicell and its largest competitor have launched new dispensing platforms at once — BD's Pyxis line handles more than 9.8 million transactions a day — pushing customers into multi-stakeholder evaluations that run quarters or years. The installed base being replaced is younger than the generation before it, so urgency is lower. Omnicell is expanding customer leasing programs to help hospitals fund the purchase. And memory chip constraints will add $6m of cost in the second half, worth 80 basis points of product margin.

The advertising meter

Phreesia sells patient check-in and payments software to practices, and separately sells drug makers the right to put branded messaging in front of those patients — a marketing budget, not an information-technology one. July-quarter revenue rose 10.4% to $129.5m, but the company paid $160m for AccessOne, a patient-receivables financier, in November. Fiscal 2027 guidance of $510-520m includes about $37m from that deal, implying organic revenue near $473-483m against $480.6m last year. That guide was itself cut from $545-559m set in December, while the adjusted EBITDA range of $125-135m was held.

The cut came from the pharmaceutical side: clients committed lower second-half spend on brand-specific and regulatory-policy dynamics, with visibility shortening. "[It is] very complex [with] a lot of different moving parts," chief financial officer Balaji Gandhi said on the fiscal fourth-quarter call, placing the impact "around the second half of the year, not the first half." Meanwhile quarterly adjusted EBITDA hit a record $32.9m from $22.1m, and average client count reached 4,744, up 6% — provider seats growing slower than reported revenue.

The subscription meter

HealthStream, which sells workforce compliance training and clinician credentialing on annual subscriptions, is the one raising. June-quarter revenue was a record $83.7m, up 12.5%, of which 8.3 points were organic and $2m was a one-time contingent-fee catch-up from a 2020 acquisition. Remaining performance obligations — contracted revenue not yet recognized — were $685m against $618m a year earlier, and full-year revenue guidance went up to $327-332m. "Our customers view HealthStream as a partner for solutions across their entire enterprise rather than a single application," chief executive Robert A. Frist, Jr. said on the August 3 call, crediting bundling.

Even here the budget shows through: HealthStream's large hospital customers flagged the expiration of Affordable Care Act premium tax credits and anticipated Medicaid reimbursement pressure on that same call.

What the prices did, and what they earn

The grouping fails first as arithmetic. Over twelve months Omnicell is up 8.1% and HealthStream 6.2%; the roughly 16% decline attributed to the group is Phreesia's 62.4% collapse alone, most of it a single 23.3% gap on the December 9 guidance warning. Over three months the ranking inverts — HealthStream up 18.0%, Phreesia up 14.4%, Omnicell down 17.2%.

Omnicell's fall is earned. At 15.3x forward earnings and a 9.0% trailing free cash flow yield it looks cheap, but consensus models earnings per share falling to $2.09 in 2027 from $2.27 this year, so the multiple sits on a shrinking base. Phreesia's bounce is the harder call: cheapest of the three at 2.03x forward gross profit against Omnicell's 2.85x and HealthStream's 4.05x, with a 10.6% free cash flow yield, but rising on margin off a wrecked base while organic growth went to nil. HealthStream, at 40.3x forward against 41.6x trailing, has almost no room between what it earns now and what it is expected to earn — and it is the only one whose backlog is paying for the price.

The test set out at the start was whether hospital budgets are reopening. Two of these three answered no, in their own guidance, six weeks apart; the third is growing by selling more products to customers it already has. Omnicell's pipeline may well be the largest in years. Nothing in its own forecast says which year it closes.

A contract signed years before the plant exists

Fervo Energy signed a 396-megawatt power purchase agreement with Google on September 1 for Cape Station, its enhanced-geothermal project in southwestern Utah. It is the largest such agreement on record, with power expected online in 2028 and an option that would take Google to nearly 1 gigawatt by June 2030. Fervo's binding contracted capacity rose about 60%, to 1,054 megawatts. It is the second deal between the two; Fervo's Project Red pilot in Nevada has fed Google's local data centers since 2023.

That pilot is also, at present, the whole revenue line. Fervo booked $113,000 in the June quarter against an operating loss of $28.7m. Geothermal is the only zero-carbon resource that runs around the clock regardless of weather, and there are exactly two US-listed pure-plays selling it — Fervo, a Houston driller applying shale horizontal-well technique to hot rock, and Ormat Technologies, which owns and operates 1,355 megawatts of commissioned geothermal, solar and recovered-energy plants and manufactures the binary turbines inside them. Both signed data-center offtake with Google inside seven months. Neither has a contracting problem. Both have a commissioning problem, and that is the entire distinction between a backlog and a business.

Fervo: the cost curve is real, the calendar is the risk

"This agreement reinforces that EGS is ready to power the next generation of computing infrastructure," chief executive Tim Latimer said on September 1. "This new PPA is part of our repeatable commercial model." The engineering supports him. Fervo's Sawtooth 7 well reached 19,500 feet and 460°F in 21 days, matching wells that were on average 35% shallower, and the company has cut per-foot drilling costs roughly 70% since 2022. Phase I of Cape Station costs about $7,000 per kilowatt; Phase II targets $5,500 on the wider, hotter Fervo 3.0 well design.

The calendar is where it bit. On its August 12 call Fervo gave first hard revenue guidance of $60-80m for 2027 — against a $7.2bn contracted backlog — blaming transmission curtailment caused by a grid operator adding a separate asset. Consensus has already marked 2027 to $72.9m, inside management's own range. Cape Station Phase I, 100 megawatts in three blocks, still targets first power in the fourth quarter of 2026. Liquidity is not the issue: $2.1bn of cash against $228.4m of debt, with first-half operating burn of only $43.8m. The spending is capital, guided at $850-900m for the second half.

Ormat: the growth came from batteries

Ormat's second quarter grew 10.6% to $258.8m with gross margin up 220 basis points to 26.5%, and full-year guidance rose to $1.15-1.2bn. Underneath, contracted geothermal Electricity revenue grew 5.8% to $169.3m, while merchant Energy Storage rose 195% on strong PJM pricing at a margin management expects to normalize toward 40-50%. Operating income fell 3.2% to $34.2m. Electricity guidance was cut about $5m to $710-725m on one-to-two-month commissioning delays at two Caribbean projects. Of 155 megawatts added year to date, 15 were geothermal — a 5-megawatt Cove Fort upgrade and the 10-megawatt Dominica plant. All 202 megawatts under construction are contract-backed, and roughly 190 megawatts of legacy contracts expiring 2031-2034 at a weighted-average $86 per megawatt-hour reprice into a market above $100.

Ormat carries $3.4bn of debt at a 3.9% average rate, 4.3 times adjusted earnings before interest, taxes, depreciation and amortization, with $449m of 2026 capital spending still to come. "We are speaking with different hyperscalers, data centers, and utilities about EGS projects," chief executive Doron Blachar told investors on August 5. Its own enhanced-geothermal answer — the Ormega100 turbine and a Desert Peak pilot with SLB — begins drilling in the fourth quarter.

What the prices have and have not settled

Ormat is 27.5% below its June 2 close of $144.48, a break dated to Bernstein's June 16 Underperform initiation, which argued it would lose the enhanced-geothermal race to Fervo by name. It is nonetheless up 15.1% over twelve months and 6.8% over the past thirty days. Fervo, which raised $1.89bn at $27 a share in May, trades a third below its offer price despite gaining 27.5% in the single September 1 session on 22.3m shares, six times normal volume; press accounts put that close at $19.75, up 28.41%. The wider listed power complex fell together over the summer against a 30-year Treasury yield at a 19-year high, but that backdrop does not explain a company rising while its band signals turned negative.

On valuation the two are barely comparable. Ormat fell from about 56 times forward earnings at its June peak to 40.8 times now, on an unchanged $2.57 consensus for 2026 — a de-rating that did rational work without making the shares cheap, since consensus has 2027 revenue growing 0.5% and earnings per share falling 2.1%. Fervo has no usable earnings or sales anchor; it trades at 1.90 times book with $2.1bn of net cash inside a $5.16bn market value, leaving roughly $2.9m of enterprise value per contracted megawatt. Policy sits on both sides of the ledger favorably: geothermal qualifies for the full federal investment or production credit if construction starts by 2033, a decade longer than solar and wind.

The verdict is symmetrical and uncomfortable. The demand premium in the geothermal story is real and is being written into contracts by the same buyer for both companies — but it has reached neither revenue line. Ormat's earned growth this year came from merchant batteries and an equipment delivery, not from the contracted megawatt-hours the thesis rests on; Fervo's contracted book is genuine and its drilling costs are genuinely falling, yet nothing in it converts to cash before a grid interconnection that a third party controls. The de-rating priced delay, and delay is exactly what both disclosed.

Ormat lays out a roadmap to 2.6-2.8 gigawatts at an investor day on September 8. Fervo's first commercial block is due to energize before the year ends. Until it does, the cleanest firm-power story on the market is still one where the megawatts exist on paper and the electricity does not.

Science Applications International Corporation, which sells engineering, IT modernization and logistics almost entirely to the Army, Navy, NASA and the intelligence agencies, reported its July quarter on August 31 and did something it had not managed all year: it raised full-year revenue guidance to $7.2-7.3bn from $7.0-7.2bn, with adjusted earnings guidance lifted to $10.65-10.75 a share. That is back above the level it cut to on February 11, when a pre-announcement blaming procurement delays and lost enterprise-IT recompetes took 16% off the stock in one session and dragged Leidos and CACI down with it. It remains below the $7.35-7.55bn the year began with.

These companies do not sell software. They sell cleared people billing hours against multi-year task orders, which makes one line item decisive: funded backlog, the slice of the order book with appropriated money behind it. A total backlog number can sit on a shelf for years under a stopgap; funded backlog is what can be billed now. On that measure the sector bear case has the direction wrong.

The funded money is arriving

CACI International — signals intelligence, cyber, electronic warfare and enterprise IT for the defense and intelligence agencies — closed its fiscal year on June 30 with total backlog of $32.0bn, up 1.9%, and funded backlog up 28.6% to $5.4bn. Leidos, which builds national-security systems, modernizes Federal Aviation Administration air-traffic control and runs disability examinations for veterans, showed the same split more sharply: funded backlog grew 44% to $10.2bn against 5% growth in total backlog, on a book-to-bill of 1.1 times. CACI's full-year bookings ran at the same 1.1 times on more than $10bn of awards.

SAIC is the exception, at 0.6 times for the quarter and 0.8 times over twelve months, with $3.8bn funded out of $22.1bn. Management's explanation is that large recompetes slipped and that extensions protect near-term revenue while deferring the booking; one major award landed two days after the quarter closed. The corroborating figure is on-contract growth of 9% in the quarter, roughly double the prior-year pace, with programs ramping from recent wins tracking to $500m this year against $350m planned. "Our recompete win rate of over 90% this quarter creates an easier path to on-contract growth," chief executive James C. Reagan told investors on August 31.

What is actually breaking

The damage is real but it is about who bills, not whether anyone does. The Defense Health Agency is replacing Leidos as lead integrator of the military health record, awarding sole-source contracts directly to the underlying technology vendors, with transitions running through July 2027. The Veterans Benefits Administration has suspended incentive payments across its disability-examination program for the rest of 2026, where Leidos's unit drew a $392.7m delivery order this fiscal year. SAIC lost the RITS enterprise-IT recompete, worth about 350 basis points of second-half revenue, and the $1.4bn CASTLE-NET Army Corps order to Accenture Federal Services after the Government Accountability Office denied its protest in May.

The contract-mix shift the bear case leans on is happening slowly. Fixed-price work is 15-18% of SAIC revenue against roughly a third of its pipeline, and its Civil group earns north of 15% EBITDA margins on fixed-price jobs. "The move to more outcome oriented fixed price is real," finance chief Prabu Natarajan said on the same call, "but they are gradual." Leidos chief executive Tom Bell told investors on August 4 that outcome-based work is "in our wheelhouse".

The split in the shares

Over twelve months CACI rose 27% and SAIC 7.5% while Leidos fell 26%, and the peer group sits with Leidos: Booz Allen Hamilton lost a third, Parsons more. CACI earned its side — revenue growth accelerated to 17.6% in the June quarter and fiscal-2026 operating income rose 20.4% on revenue up 10.9% — but at 25.3x trailing earnings and 18.7x forward it now trades above the 19-24x range of the past two years and above the 21.0x it carried in early May. Leidos is the harder case: revenue accelerated to 7.2% and 2025 net income grew 16.1%, yet operating margin fell to 11.1% from 13.4% and quarterly operating income declined 11.2%. At 12.4x trailing and 10.8x forward against roughly 16x a year ago, with a 12.9% free-cash-flow yield, the market is pricing the health-record loss as permanent while consensus still models 6% revenue growth next year. SAIC sits between them, and on enterprise value against EBITDA at 7.8 times it is the cheapest of the three.

The verdict the funding data supports is narrower than a sector rolling over. Appropriated dollars are reaching these contractors faster than headline backlogs suggest; what has changed is that agencies are reallocating those dollars — insourcing integration, consolidating vehicles, handing incumbents' work to rivals. That is a market-share problem with names attached, and it argues against reading Leidos's de-rating as a budget verdict when its own funded backlog grew fastest of the three. It also means CACI's re-rating through the top of its own band now depends on the second-half acceleration management has guided to, with first-quarter organic growth flagged as low single digits.

About 15-20% of annual federal contract obligations land in September alone, and a stopgap has removed the October shutdown risk through early December. The awards are coming. Which of these three names appears on them is the only question the backlog disclosures cannot answer.

BWX Technologies, the sole-source supplier of nuclear reactors and fuel to the US Navy's propulsion program, ended June with backlog of $8.4bn, up 40% from a year earlier, its quarterly filing shows. On August 3 it raised its full-year revenue, adjusted profit, earnings and free cash flow guidance. The shares have fallen 24.4% since early March.

BWXT and Cameco, the Saskatchewan uranium miner that owns 49% of reactor builder Westinghouse, were both re-rated over the past two years as ways to own electricity demand from artificial-intelligence data centers. Neither books much of it. BWXT earns on multi-year Navy and Department of Energy contracts recognized as work is performed; Cameco sells uranium out of a long-term contract book with floors and ceilings struck years ago, and takes Westinghouse income on project milestones. What the market has removed since March is a demand premium that was never in either income statement. What it hit underneath differs by name.

At BWXT, the growth is real and the margin is not

June-quarter revenue rose 18% to $901.6m. Two-thirds of that — $601.3m — was Government Operations, which grew 2% at a 17.6% segment margin, while Commercial Operations grew 72% to $302.5m at 8.0%, the company reported. The commercial jump came from nuclear components, field services, fuel handling, medical isotopes and the Kinectrics acquisition; advanced nuclear volumes inside the government segment were lower. There is no small-modular-reactor revenue line doing the work.

The bears have one real number: GAAP operating income fell 12.2% to $89.9m on the 18% revenue gain, with gross margin down to 22.4% from 25.1% as BWXT builds capacity. On the August 3 call it cut its commercial margin guide to about 13% from about 14% and pushed meaningful recovery to 2027. Management also said it expects "at least one new build nuclear equipment order by year-end" — which means none had been booked. Its Army microreactor selection for Fort Campbell targets the early 2030s, and it agreed on July 31 to sell its medical business to Nordic Capital in a deal valued at up to $800m.

At Cameco, the drawdown made the stock dearer

Cameco's June-quarter revenue fell 7.2% to $814.1m and net income fell 92% to $25.2m. The cause is not price. Sales volumes fell 18% on "lower planned 2026 sales delivery volumes, reflecting contracting discipline," the company said, while the Canadian-dollar average realized uranium price rose 15% and Cameco raised its realized-price and consolidated revenue outlook. Delivery commitments still average about 28 million pounds a year through 2030, and production guidance of 19.5m to 21.5m pounds was left unchanged.

The damage sits at Westinghouse. Cameco's share of its adjusted profit fell to $163m from $352m, the prior year having carried roughly US$170m from the Czech Dukovany milestone, and the company eliminated its five-year Westinghouse growth outlook entirely, guiding 2026 to US$370-430m. Westinghouse has since confidentially filed for an initial public offering.

The split

The two did not fall together. BWXT is down 10.4% over thirty days and its 50-day average has been below its 200-day since early June; Cameco rose 7.4% over the same stretch, helped by an August 21 jump on the Westinghouse filing and a raised price outlook, and its own trend has begun to mend.

So the verdict divides. BWXT's de-rating carried past the premium and into the contracted business: 32.8x forward earnings against roughly 47x in May, with backlog and book-to-bill still rising and free cash flow guided to $345-360m. Cameco's did not de-rate at all — at 62.6x forward against about 56x in May, earnings fell faster than the shares, and consensus does not expect the step-change until 2027.

Both now depend on the same unbooked thing: a large-scale AP1000 order cycle. BWXT has told investors one equipment order should land before the year is out. Cameco has stopped forecasting Westinghouse five years out and is letting an IPO price it instead.

Leggett & Platt sold its last share on August 27. The company, founded in Carthage, Missouri in 1883 and never a seller of finished mattresses — it makes the steel rod, drawn wire, innersprings, specialty foam, foundations and adjustable bases inside somebody else's bed — was absorbed the day before into Somnigroup International, the owner of Tempur Sealy and Mattress Firm and, for nearly half a century, Leggett's own supply counterparty.

That matters beyond the delisting notice. A component supplier is paid on units, and the units are gone: the Mattress Recovery Council's industry group ISPA reports US mattress unit shipments fell 13.2% in 2025 while the market's total value fell only 6.5%, as average selling prices for domestically produced beds rose nearly 5% and buyers who did show up traded up. Stationary foundation shipments — the steel-heavy product Leggett sells — dropped more than a quarter. Fewer beds, less content in each. Leggett's answer, after two years of plant closures, a gutted dividend and divestitures, was to be bought by the customer.

The margin repair was real

June-quarter revenue of $999.7m fell 5.6% year over year, yet gross profit rose 7.4% and operating income rose 13.3% to $80.3m. Gross margin reached 20.3% against 17.9% a year earlier. That is genuine self-help, and it is one quarter old: in the March quarter operating income fell 29% and margin was 5.0%. Revenue has declined every year since 2021 and is down about 20% from that peak.

Bedding Products trade sales were roughly flat at $386.9m on volume down 7% — retailer merchandising changes, lower volume with an adjustable-bed customer, soft US and European demand, and a decision to walk away from a financially challenged US spring customer — partly offset by higher trade rod and wire sales. Steel is both the input and, at these prices, part of the offset: US hot-rolled coil reached $1,002 a ton in March against $694 at the start of 2025 after Section 232 tariffs doubled to 50%. Net debt stood at 2.6 times trailing adjusted earnings before interest, taxes, depreciation and amortization, the quarterly dividend was held at $0.05, and full-year guidance had been withdrawn on account of the merger. The company held no earnings call.

The price was arithmetic

With the exchange ratio fixed at 0.1455 Somnigroup shares, Leggett's quote was its acquirer's, multiplied. On August 27 it closed at $9.20 against a conversion value of $9.08. On August 6, when Somnigroup cut guidance and fell 7.2%, Leggett fell 6.7% the same session on no news of its own. The 30-day slide that looked like a demand rollover was one company's earnings call transmitted through a ratio; La-Z-Boy fell 18.6% over the same stretch and Mohawk Industries rose 13.7%, so nothing segment-wide was happening. The deal was worth about $2.5bn at April's announcement and about $2.3bn at completion, the difference being Somnigroup's own share price.

"By bringing a successful supply partner into our group, we accelerate our ability to deliver differentiated, consumer-centric innovation," Somnigroup chief executive Scott Thompson said when the deal was struck. On the August 6 call he was blunter about the demand: "There's no question there's a K. Entry-level bedding has been the hardest hit by far, and luxury bedding, we'll call it, has been very resilient and at times strong."

What the buyer now owns

Somnigroup is running the same pattern at four times the scale — June-quarter revenue down 3.0% to $1.82bn, operating income up 12.1% — and its North American Tempur-Sealy gross margin jumped to 61.8% on Mattress Firm integration synergies. But it guided full-year adjusted earnings to $2.85–$3.15 a share on an assumption that the global bedding industry is down mid-single digits, upgraded from low-single digits, and the shares are down 27% year to date. They have de-rated from about 26.6 times trailing earnings to 22.3 times forward; on consensus 2027 earnings of $3.62 the same price is 18.7 times. Combined leverage was roughly 2.8 times adjusted EBITDA at close, with run-rate synergies raised to $75m.

The verdict splits cleanly. Leggett's restructuring earned its margin expansion, and none of it solved the unit problem — analysts model 2027 revenue growth of 1.9% and flat earnings for the standalone business, which is to say the base it was cutting toward never stopped shrinking. Vertical integration is the industry's answer: capture the component margin inside a branded retailer rather than sell it into a market that buys 13% fewer beds. Whether that works is now a question about one share price, because owning the innerspring business means owning the mattress retailer.

The listed component supplier no longer exists. What is left is a bet that the world's largest bedding company can earn more per bed at a time when the industry keeps selling fewer of them.

The substrate business that was supposed to be Wolfspeed's moat is the part that broke. Materials revenue fell 45% year over year in the June quarter, to $43.3m, on what the company called a tightening demand environment and increased competition, while power devices rose 6% sequentially on AI data-center orders.

That leaves group sales running roughly a quarter below the volume management says it needs to break even on gross margin — fixed-cost arithmetic, and depreciation on two 200mm fabs does not care about it. Shares in both listed wide-bandgap pure-plays roughly halved over three months, but only one of the businesses deteriorated: Navitas grew revenue 22.5% sequentially and guided higher, and still trades near 80 times trailing sales.

Wolfspeed — the Durham, North Carolina maker of silicon carbide and gallium nitride materials and power devices — reported its fiscal fourth quarter on August 19, and its two halves moved in opposite directions. Power products, the switches sold into electric vehicles and increasingly into data-center power supplies, brought in $106.3m, up 6% from the March quarter. Materials, which sells silicon carbide substrate wafers to everyone else in the industry, brought in $43.3m, down 45% from a year earlier — a decline the company attributed to a tightening demand environment and increased competition.

The materials line is the one that matters beyond a single quarter, because it was the profitable, vertically integrated end of the business: the Mohawk Valley device fab fed by the roughly $5bn John Palmour Manufacturing Center in Siler City, the only 200mm silicon carbide chain of its kind at scale. Those buildings depreciate whether or not wafers move through them. Management put gross-margin breakeven at roughly $800m of annual revenue and named factory utilization the primary lever. September-quarter guidance is $140m to $160m of revenue with non-GAAP gross margin still negative — about a $600m run rate. Roughly $200m of annual revenue has to appear before the first dollar of gross profit does.

Four quarters down, capital spending switched off

Group revenue was $149.6m, down 24% year over year, at a GAAP gross margin of minus 25% and a net loss of $145m. It was the fourth consecutive quarterly decline, and the guide makes a fifth. Full-year revenue was $665.1m against $757.6m in fiscal 2025. Capital expenditure was cut to $5m in the quarter from $38m against about $1.1bn of cash and roughly $600m of net debt, the legacy of a prepackaged Chapter 11 that closed on September 29, 2025 and took borrowings from about $6.7bn to about $2bn. Chief executive Robert Feurle said the quarter's achievements "strengthen our technology leadership and confidence in our long-term growth opportunities."

The cheap-Chinese-substrate explanation needs a caveat. Six-inch prices in China collapsed from roughly $1,500 to about $500 a wafer, with Tankeblue and SICC taking the number two and three share positions — but Digitimes reported in July that six-inch prices had bottomed and begun recovering as supply tightened. The likelier reading of a 45% decline is lost share and weak electric-vehicle demand landing on top of a price reset that already happened.

What is growing is small and undisclosed. Wolfspeed said AI data-center revenue more than doubled from fiscal 2025 to fiscal 2026 and rose about 20% sequentially, without giving a dollar figure, and on August 6 announced its silicon carbide had been qualified into LITEON's 800-volt direct-current sidecar platform for hyperscale customers.

The fabless side of the same bet

Navitas Semiconductor, the Torrance, California gallium-nitride and silicon-carbide designer that owns no fabs and employs 190 people, is running the opposite cost structure and posted the opposite trend: June-quarter revenue of $10.5m, up 22.5% sequentially, with third-quarter guidance of $13.5m at a 39.7% non-GAAP gross margin, and $557m of cash and no debt. Its data-center content arrives in four dated tranches from the second half of 2026 through 2028, with the largest — in-rack 800-volt conversion — in 2027. "We are also actively engaged with hyperscalers, merchant power customers, data centers, OEM, ODM on multiple program ramping in the second half of 2026 that will accelerate throughout 2027," chief executive Chris Allexandre told investors on the July 27 call. Wolfspeed sued Navitas in Delaware on July 7 over five wide-bandgap patents; Renesas, a large Wolfspeed shareholder, followed on July 22 with a trade-secret claim. Navitas countersued in August and called the suits a sign competitors are losing share.

What the fall does and does not explain

Over the three months to September 2, Wolfspeed fell 55% and Navitas 58% — but silicon-incumbent Vishay fell 41%, onsemi 40% and STMicroelectronics 27% while the S&P 500 rose about 1%. The 20 largest chip stocks shed more than $1.3trn in the days to July 29. This was an unwind of AI-power momentum that hit the pure-plays hardest because they had the least earnings underneath.

Wolfspeed earns its share of it. The shares now change hands at 2.08 times trailing revenue and 2.16 times forward — the forward multiple sits above trailing because consensus has fiscal 2027 revenue falling 15% to $642.8m — and price-to-book is 1.48 times, close to the accounting value of the fabs. Navitas is a different case: the business accelerated and the halving only takes it from roughly 190 times trailing sales to 80, with 61 times forward, against consensus $500m of revenue in 2030 that would still be 5.9 times today's market value. Its fall repriced a story; Wolfspeed's recorded a shrinking one.

Both companies now point at the same prize — 800-volt racks in 2027. Wolfspeed has to fund four more quarters of depreciation to be there to collect it. Navitas has to ship.

New Jersey Natural Gas, the regulated utility serving about 564,000 customers across six New Jersey counties, asked the state Board of Public Utilities on June 1 for a $157.6m increase in delivery revenue — and filed, the same day, to take roughly $158 a year off the average residential bill by cutting the pass-through cost of the gas itself, an 8.9% reduction ahead of the coming winter.

That pairing is the entire economics of a local distribution company. It buys the molecule and hands it to households at cost, earning nothing on it; every dollar of profit comes from the pipe, recovered in periodic rate cases and, between them, through surcharge mechanisms. Which means the commodity it does not profit from sets the bill that determines whether regulators will let it earn more on the pipe it does. Cheap gas is regulatory headroom. The question this year is who has it and who has spent it.

Buying room

New Jersey Natural Gas's ask rests on a $4,046.1m rate base, a 10.10% return on common equity and a 55.5% equity layer, and seeks to reflect roughly $950m of system investment not currently in rates. The company timed implementation for the end of the 2026-2027 heating season, when usage falls, and notes the board's review of base-rate filings typically runs nine to twelve months and typically ends in a settlement. On parent New Jersey Resources' August 4 call, management called the case "plain vanilla" and described the package as leaving customer rates "nearly flat." It is flat only because the commodity line fell far enough to absorb the delivery increase.

ONE Gas, which is essentially all regulated distribution — 2.2 million customers in Oklahoma, Kansas and Texas — got everything it asked for. An Oklahoma administrative law judge approved Oklahoma Natural Gas's full $28.7m base-rate request, worth about $1.53 a month to the average household. Texas Gas Service won a $36.9m statewide infrastructure surcharge in July. The Kansas legislature went further and loosened the mechanism outright, raising the maximum monthly residential surcharge to $1.35 from $0.80 and cutting the review clock to 90 days. A separate Texas statute allowing deferral of depreciation and taxes on capital awaiting rate treatment is worth about $0.42 to 2026 earnings per share.

The result was second-quarter adjusted earnings of $0.82 a share against $0.54, on weather 25% warmer than normal, and full-year guidance raised to $4.89–$4.95. Rate base averages $6.3bn this year on roughly $800m of capital spending. The bill for that comes in dilution and cash: the diluted share count is up 4.5% year over year and trailing free cash flow yield is minus 3.4%.

Where the room ran out

Elsewhere the affordability squeeze is real and visible. The Indiana Utility Regulatory Commission denied NiSource's $741m five-year gas modernization plan 3-0 on August 5, finding the utility "failed to provide sufficient evidence for the commission to determine that each proposed eligible improvement is cost-justified". Commissioners Bob Deig, Anthony Swinger and David Veleta added: "We decline to accept NIPSCO's suggestion that cost-justification can be inferred because the projects were selected by internal subject-matter experts." Two days later Indiana convened a technical conference on returns and trackers. NiSource reaffirmed 2026 adjusted earnings guidance of $2.02–$2.07 anyway; its shares are down 13% over six months.

Southwest Gas, the Arizona-Nevada-California distributor, shows the same pressure as an earnings gap: an 8.1% return earned over twelve months against 9.89% authorized, with Nevada intervenors recommending about $40m of a $74m request at a 9.3% return, an order targeted for October. Its forward multiple of 21x is the highest here; the trailing figure is distorted by a one-off 2025 gain. Atmos Energy, the largest pure distributor, is the counter-case on throughput: 51,000 net customer additions in the year to June and $396m of annualized operating income increases already implemented.

What the de-rating is actually pricing

Every one of these companies has been marked down. ONE Gas trades at 17.2 times trailing and 16.3 times forward earnings, against 20.2 times in early May. Atmos is at 19.8 times both ways, against 24.4 times in May. New Jersey Resources sits at 14.8 times trailing against 17.4 times in May — but its forward multiple of 15.0 times sits above trailing, because consensus 2027 earnings of $3.40 are 5.1% below this year's $3.59. That is not the regulator. Roughly 38-41% of New Jersey Resources' earnings come from outside the utility, per its own segment guidance: unregulated Energy Services at 21-23%, Clean Energy Ventures at 10-13%, midstream at 8-11%.

So the two pure-plays are not one trade. Over the past month ONE Gas rose 3.4% while New Jersey Resources fell 6.5%; over six months the order reverses, ONE Gas down 8.6% against 1.1%. Nothing in the rate-case record explains either. What does explain most of it is the discount rate — the 30-year Treasury above 5.3% in August, with the utilities sector down about 5.6% on the month — against an authorized return fixed in nominal terms for the nine to twelve months a case takes. Affordability is throttling recovery in Indiana and Nevada. At ONE Gas and New Jersey Natural Gas it is not yet touching the rate base; it is being managed with someone else's money.

That money is borrowed from a commodity price nobody at these companies controls. The Energy Information Administration's July outlook put Henry Hub at $3.67 per million British thermal units for the year; a colder winter or a tighter market takes the headroom back, raises the bill without adding a cent to the utility's return, and forces the next filing to argue for the pipe on its own merits.

Salesforce was the last of the front-office software names to turn, not the first. Braze and Freshworks — both selling customer messaging and support software to mid-market and enterprise buyers — had already crossed into uptrends, their 50-day averages above their 200-day, by 3 August. Salesforce was still in a downtrend that day. It reported on 26 August and gapped 22.6% higher the next session, from 205.62 to 252.05 on 55.5m shares against a baseline near 10m a day the week before.

That sequence changes what the move means. The re-rating was not Salesforce dragging a sector behind it; it was a broad unwind of the "SaaSpocalypse" positioning that had priced recurring-revenue software for the moment AI agents replaced the humans holding the licenses. CNBC's account of the 27 August session has Salesforce up alongside CrowdStrike, Adobe and Autodesk on the same day. The interesting question is which of these businesses actually earned it — and the two largest here run opposite meters, which makes them a clean test.

The usage meter delivered

Twilio, whose programming interfaces let developers embed voice, messaging and verification into their own apps, has never sold a seat. June-quarter revenue of $1.499bn grew 22.0%, the third consecutive acceleration from 14.3% and then 20.0%; organic growth, stripping incremental US carrier surcharges, was 17%. Gross profit growth accelerated in step, from 10.4% to 17.7% to 20.4%, so the volume is reaching the gross line. Dollar-based net expansion — what existing customers spend this year against last — reached 116%. Guidance went up twice, with full-year organic growth now 13-13.5% against 9.5-10.5% before.

The one blemish is a non-GAAP gross margin of 49.1%, down 160 basis points, which management attributes entirely to roughly $71m of incremental US carrier pass-through fees billed on at cost; absent them margin would have risen, and operating margin gained a point anyway. Chief executive Khozema Shipchandler called the quarter "exceptional" and put AI-related demand in "very early innings," most visible so far in voice. There is still no AI revenue line; the disclosure is triple-digit growth in AI-native customer cohorts and an eight-figure deal with one AI company.

The seat meter did not

Salesforce's July quarter grew 10.8%, decelerating from 13.3%, and about four points of that came from Informatica, consolidated this year and absent from the year-ago base — near 6.4% organic. Gross margin fell about a point and a half to 76.7%. Operating income of $2.331bn was flat year over year. Net income rose 86.9% and per-share earnings more than doubled, arithmetic supplied by a diluted share count falling from 962m to 821m under a $25bn accelerated repurchase that management expects to shrink the count at least 14%, at an average near $176.

Two lines genuinely confirm demand. Current remaining performance obligation — contracted revenue due within a year — reached $33.5bn, up 14% in constant currency, three points faster than revenue. Agentforce annual recurring revenue passed $1.5bn, up more than 240%. But that is roughly 3.5% of a subscription run-rate near $43bn, and the seat-billed core — Sales, Service, Marketing, Commerce, Slack — was $7.2bn of the quarter's $10.82bn subscription revenue, growing 8% in constant currency. Marc Benioff pointed on the 26 August call to what consumption looks like at scale: "The Army Human Resources Command expects to drive up to 55 million Agentforce conversations every single month." Management's stated near-term lever, though, is a seat upgrade: only 5% of sales and service knowledge workers sit on premium tiers carrying a 60-80% price premium.

What is paid for, and what is not

Salesforce trades at 23.4x trailing and 15.6x forward earnings, but the forward year is the flattering one — consensus has fiscal 2028 earnings of $16.03 a share, below fiscal 2027's $16.43. Twilio's 38.3x forward sits above its 30.2x trailing, because a one-off non-operating gain produced $1.067bn of quarterly net income on $84.5m of operating income; against its own ten-year history it remains cheap on sales. Because Salesforce converts three-quarters of revenue to gross profit and Twilio under half, the comparable measure is what each costs per dollar of gross profit: Twilio costs about twice what Salesforce does. HubSpot, the remaining holdout, only reached an uptrend on 27 August and is down 47.9% over twelve months.

The verdict splits. Twilio's advance is backed by accelerating volume through a meter AI agents make busier, though its own third-quarter guide steps organic growth down to 11-12%. Salesforce's is backed by backlog and a fast-growing consumption tier that is still too small to move the revenue line, sitting on top of an income statement where the operating business stood still and the shareholder gain came from the share count. Buying either one requires believing the relief trade was right about the category before Salesforce's numbers arrived.

Up to 55 million conversations a month, billed by the action rather than by the login — that is the business Salesforce says it is becoming. Its income statement has not shown it yet.

Atmus Filtration, the Fleetguard filter business separated from Cummins, built its case on things trucks use up rather than things truckers buy new: roughly 86% of 2025 net sales came from the aftermarket, replacement fuel, oil and air filters for vehicles already in service. In the June quarter that stream was flat year over year globally — no more filters sold — even as North American Class 8 net orders ran 120% ahead of 2025 year to date through July.

Those two facts only look contradictory until you ask what is fixing the freight market. It is subtraction. The ATA's for-hire tonnage index fell 1% in July to 113.5 and sat 0.5% below a year earlier, with the association crediting the recovery to excess capacity leaving rather than to freight growing; tighter Department of Transportation enforcement of non-domiciled commercial driver licences and English-proficiency standards has hastened the exit. Surviving carriers get better rates and order equipment. The total miles run across the national fleet — the only variable that consumes a filter — do not move. An annuity indifferent to the build cycle turns out to be indifferent to the upswing as well.

What the quarter actually contained

Atmus reported record revenue of $527.9m, up 16.4%, and raised full-year sales guidance to $1.975bn–$2.030bn. Most of the growth was bought: the Industrial Solutions segment created by the Koch Filter acquisition contributed $42m on its own. The core engine-filtration business tells the plainer story. "Power Solutions delivered sales of $486 million compared to $454 million in the prior year. An increase of 7%," chief financial officer Jack Kienzler said on the August 7 call. That 7% was three points of price, two of currency and two of volume. Aftermarket, at 85% of the segment, was flat — stronger in the United States and Mexico, softer across Europe, the Middle East, Africa and Asia outside China.

The second half looks thinner. Pricing is guided to moderate to about 0.5% from more than 2% in the first half, commodity inflation in chemicals and plastics tied to Middle East conflict is hitting both costs and Atmus's Indian joint-venture income, and the adjusted EBITDA margin range was narrowed to 19.75%–20.25% with the top end cut. Tariffs are the one thing not doing damage: Kienzler told investors the net EBITDA impact should be "substantially neutral." Chief executive Stephanie Disher said the company has "yet to see a significant inflection" in freight and expects the aftermarket to stay roughly flat. Full-year segment guidance assumes volume between zero and 2%.

Shares fell 7.9% on the day of that raised outlook, are down 12.2% over the past month, and sit 28% below their 52-week high of $65.57. JPMorgan cut the stock to Neutral with a $46 target, raised from $44, on valuation rather than demand.

The uncomfortable comparison

Donaldson, the diversified filtration maker that competes directly in engine aftermarket parts, grew fiscal-2026 aftermarket sales 8.1% with growth in every region and a double-digit increase in the independent distribution channel — the same channel where Atmus is currently signing new distributors for Fleetguard and Koch. Same parc, same idle miles, opposite result. Atmus guides to 1%–2% of aftermarket share gain for the year and claims roughly 90% share in US medium-duty trucks; the June quarter did not show it.

The dealer rung corroborates the mechanism. Rush Enterprises, the largest North American commercial-truck dealership network, took $605m of parts, service and collision revenue in the June quarter, up 1.5% and 64% of total gross profit, at a 130.8% absorption rate. New-truck demand is the hot end: "Our backlog is as big as it has been in a couple years, to be honest with you, where we sit right now. And I will tell you, we are basically sold out," chief executive W. Marvin Rush told the July 29 call. Rush's own quoted price dropped by a third on September 1 — that was the three-for-two stock split taking effect, and adjusted for it the shares are up 29.2% over twelve months, on trailing profits that are still falling.

The verdict

Atmus earns a good part of its de-rating. Price is rolling off, input costs are running hotter for longer than management first thought, the margin band came down, and the acquisition that flatters the top line is already in consensus — analysts model revenue growth decelerating to about 5% in 2027. At 16.0x forward earnings against 18.0x trailing and 12.8x trailing EV/EBITDA, it trades well under Donaldson's 20.6x forward and 18.5x EV/EBITDA, and the gap is not free money: the cheaper company is the one losing relative ground in the channel both are contesting.

What is wrong is the premise that the truck economy is rolling over. Class 8 build slots for 2026 are oversubscribed, the dealer aftermarket is growing, and one of the two filter makers grew its replacement business by 8%. Atmus's problem is narrower and more specific: it owns a toll on miles at a moment when the industry is fixing its economics by parking trucks instead of running them.

Atmus has not confirmed a date for third-quarter results; the last two years put the print on the first Friday of November. The line to read will not be revenue, which the Koch deal will lift again. It is whether aftermarket volume finally moves off zero.

Sold before it was produced

Expand Energy, the former Chesapeake Energy renamed after its 2024 combination with Southwestern, has already sold most of next year's output. As of late July the company had 66% of 2026 volumes and 41% of 2027 volumes hedged: 500.4 billion cubic feet of swaps struck at $3.91 per thousand cubic feet, and 878.9 Bcf of costless collars floored at $3.49 and capped at $4.75. Front-month Henry Hub settled at $3.001 on 2 September. Those positions are currently adding to what Expand gets paid; the only constraint is the $4.75 ceiling, which binds only in a rally.

That is the mechanism worth holding onto, because a gas producer's revenue is not the futures price. It is Henry Hub, minus a basin discount set by pipeline capacity, plus or minus whatever was pre-sold. Over the past month the commodity did all the work — front-month gas rose from $2.758 on 3 August — and the two largest members of the Appalachian gas group captured less of it than the smaller ones: Infinity Natural Resources gained 23.2%, Gulfport 15.9%, Diversified 13.8% and Antero 10.9%, against EQT's 5.6% and Expand's 6.7%. No company-specific disclosure from either large name fell inside the window; the last material events were the July second-quarter releases and Expand's 27 July acquisition agreement.

What the basin actually pays

EQT, the largest US natural gas producer, holds roughly 2.0 million gross acres with 1.7 million in the Marcellus. Its June-quarter realized pricing differential was -$0.67 per unit, better than guidance even though basis widened during the quarter, which the company credited to marketing optimization and its curtailment strategy. Since then the discount has been closing: the September Texas Eastern M2 basis contract printed -$0.95 in early August after touching an all-time high of -$0.90, with winter 2026/27 and 2027/28 basis trading about 15 cents above May levels. The physical relief is dated: FERC has authorized construction of Transco's 1.6 Bcf/d Southeast Supply Enhancement for late 2027. Takeaway, not geology, has been the binding constraint — one interstate pipeline has been completed out of Appalachia in roughly twelve years.

EQT's quarter showed the split between price and performance clearly. Revenue fell 29.2% year over year to $1.81bn and operating margin compressed to 25.1% from 44.3%, with gas averaging $2.89 per million British thermal units. Volumes went the other way: 634 Bcfe beat guidance, full-year output was raised about 90 Bcfe, capital spending was cut $25m, and free cash flow attributable to EQT was $330m. The demand it has contracted is real and late — a ten-year, 325 million cubic feet a day agreement with Competitive Power Ventures for a West Virginia power facility starting as early as 2031, priced off PJM power rather than a gas index and sized by management at roughly $100m of annual free cash flow, plus a five-year liquefied natural gas offtake beginning 2028 worth about $45m.

Louisiana gas in an Appalachian basket

Expand produced 7.48 Bcf/d in the June quarter, and 42.6% of it came from the Haynesville, where its Gen3 completion design has pushed asset breakevens below $2.75/Mcf. It holds roughly 70% of remaining core Haynesville inventory, the swing supply for Gulf Coast export terminals. Its $1.25bn purchase of gas marketer Twin Eagle Holdings, agreed 27 July and expected to close this quarter, adds more than $200m of annual EBITDA and cuts a corporate breakeven of roughly $2.70/Mcf by another 5 to 10 cents before synergies.

Management has been buying its own stock against that math. "Prop-month natural gas prices dipped after the first quarter, and we were prepared to act decisively when our stock price dislocated from our mid-cycle price view of $3.50 to $4," interim chief executive Michael Wichterich told investors on 28 July. Expand repurchased $850m of shares in the quarter, about 4% of the float, with a further $1bn authorized.

What the move earned and what it did not

The rally was the commodity's, and neither large producer kept pace with it. What the businesses earned is narrower and more durable: Expand's realizations are being lifted by a hedge book struck above the market and by Haynesville costs falling, and EQT's volume beat and differential defense are its own doing. What nothing in the accounts explains is the price of the two relative to each other. EQT carries a 67% premium on trailing enterprise value to EBITDA — 6.53x against 3.91x — on almost identical free-cash-flow yields of 10.8% and 11.1%, and its forward price-to-earnings of 13.47x sits above its trailing 12.26x because consensus models 2026 earnings below the last twelve months and 2027 earnings down 5.2%. Its trailing multiple was 11.28x in early May; the share price fell and the multiple did not.

The grouping itself is the last casualty. The largest name in an Appalachian pure-play basket takes most of its gas from Louisiana and sells it toward export demand; the other is a Marcellus producer whose upside is written into power contracts that begin in 2028 and 2031. They move together because the same futures curve prices them, and that is the only thing they share.

Expand's own view is that 3.5 Bcf/d of new Permian associated-gas egress by year-end keeps the market oversupplied through at least the first half of 2027. The Energy Information Administration expects end-October inventories at a record 3,985 Bcf. Everything both companies have contracted arrives after that overhang has to clear.