DK Street Journal

onsemi spent September 16 telling investors it would be a larger and far more profitable company by the end of the decade. The maker of power and sensing silicon — metal-oxide field-effect transistors, silicon carbide modules and image sensors sold into electric vehicles, charging infrastructure, solar and industrial power — targeted roughly $11bn of 2030 revenue on a 12–14% annual growth model raised from 10–12%, a 53% non-GAAP gross margin against about 40% this year, and more than $2.5bn of AI data-center revenue against roughly $500m in 2026. The same afternoon, the Federal Reserve raised its benchmark rate a quarter point to 3.75%–4% on a 12-0 vote, its first increase since 2023. By the following close onsemi had fallen 7.3%, to $67.73.

That pairing is the whole story of the analog power complex this quarter. Every one of the six names that supply voltage regulation and signal chain into data centers, cars and factories peaked within days of June 18, and every one is lower: onsemi is down 44.3% over three months, Vicor — which builds power modules and converters — 34.7%, Microchip Technology, the microcontroller and mixed-signal maker, 28.7%, Monolithic Power Systems 25.2%, Texas Instruments 18.6% and Analog Devices 16.5%. The ranking does not follow AI exposure. The most automotive- and industrial-levered name fell hardest; the name whose data-center line is growing fastest fell least. It follows how far each had run and how much current earnings sit underneath the price — which is what a discount rate does. The 10-year Treasury yield climbed back to 5% after the decision, and this is a group carrying 22x to 44x forward earnings.

The per-socket name kept accelerating

Monolithic Power designs the multi-phase controllers, power stages and vertical power modules that feed current to accelerators, and it is paid dollar content per platform rather than by volume of a catalog. Its second quarter, reported July 30, was the third straight acceleration — revenue up 47.6% to $980.6m after 26.1% and 20.8% growth in the prior two quarters. Enterprise data revenue reached $380.6m, up 164.3% from a year earlier, and 38.8% of the company against 32.7% three months before. Management raised its full-year enterprise-data growth target to 130% from 85%, citing new socket wins and higher module content. Gross margin sat flat at 55.2% while operating margin went from 26.6% to 31.0%, and inventory fell to 140 days of current-quarter revenue from 157.

"We focus on growing the revenue and growing the net profit, EPS," chief executive Michael Hsing told investors on the July 30 call, describing a shift from chip supplier to full solution provider.

The shares nonetheless moved from a strong uptrend as recently as June to a decisive downtrend, their 50-day average now below the 200-day. The one dated, name-specific cause inside the window is a September 14 Edgewater Research note questioning whether Monolithic Power keeps its slot on Nvidia's next platform, which took the stock down 7.39% that session. It remains an allegation: no confirmation of a lost socket has been published, and secondary accounts differ on which platform is at issue. Estimates have not moved down with the price — consensus calls for FY2026 earnings of $27.42 a share, up 54.8%, and 25 analysts carry an average target near $1,830 against a September 17 close of $1,169.58. What compressed was the multiple: trailing price-to-gross-profit is 33.1x, against 38.5x on August 18.

The catalog name printed the best quarter and fell least

Analog Devices sells tens of thousands of long-lived data converters, amplifiers and power parts, mostly through distributors, into industrial, automotive and communications customers. Its July quarter was the first $4bn quarter in company history, revenue up 39.6% to $4.02bn, with GAAP gross margin at 67.3% and operating margin at 40.1% against 31.5% two quarters earlier. Industrial grew 53% and is 49% of revenue; communications grew 84%, and data center is now 80% of that segment. Book-to-bill is above one but, management said on the August 19 call, "not unusually elevated," with distributor channel weeks deliberately held below the 6-to-7-week target. Pressed on durability, the chief financial officer grounded the outlook in "bookings momentum, design activity, and pipeline visibility" rather than "hopes and dreams."

Texas Instruments, which owns its fabs and earns on loading them, shows the same pattern: June-quarter revenue up 22.8% to $5.46bn, gross margin recovered to 61.4% from 57.7%, operating margin 42.3% — and shares down almost a fifth in three months.

What the falls earn

On forward gross profit, the comparable measure when margins run from onsemi's 38% to Analog Devices' 67%, Monolithic Power is still the most expensive of the four at 26.1x, ahead of Texas Instruments at 18.8x, Analog Devices at 18.4x and onsemi at 11.1x. That ordering is the verdict. Monolithic Power's quarter-long fall is a level correction on a premium that survived it, not a response to anything in the print — the open question there is a research allegation about a future platform, which no filing has yet answered. Analog Devices was marked closest to its own numbers. onsemi is the one the market treated hardest while pricing it cheapest, and it is the name promising AI content it has largely not booked, with 9.2% revenue growth and a 38.4% gross margin last quarter; sell-side cuts after the investor day, Stifel to $75 and Truist to $91, both still sit above the price.

The common factor is the cost of money, and a group that re-priced together on a rate decision is a group that will not sort itself by socket content on the way back. Sixteen of eighteen Fed participants project at least one further increase. On September 18 both Monolithic Power and Analog Devices rose, the first session since the hike in which the direction was genuinely in doubt.

Stryker's chief financial officer spent part of September 8 explaining why a cyberattack that hit the company in March is still holding back sales. Recovery in peripheral vascular is taking longer than expected, Preston Wells told the Wells Fargo Healthcare Conference, costing 70 to 80 basis points of growth in the second and third quarters and probably the fourth as well, and hip sales had come in slower than planned — softer joint replacement in Europe, plus US summer seasonality. He did not cut guidance. Stryker's shares closed down 7.4% that day, alongside Boston Scientific, which said in a filing it would not meet third-quarter or full-year expectations after its own cyber incident.

That session is the visible edge of something larger and stranger. Over the twelve months to September 17, Stryker — a Kalamazoo, Michigan maker of hip and knee implants, surgical robots, endoscopy and neurovascular devices — fell 25.2%, to $282.07. Zimmer Biomet, the Warsaw, Indiana implant house whose entire business is the thing supposedly being commoditized, fell 4.2%. Over three months the two moved in opposite directions: Zimmer up 8.8%, Stryker down 8.4%. The market has taken a quarter off the company whose reported numbers accelerated and left flat the company whose knee line stopped growing.

The numbers Stryker actually printed

Second-quarter revenue of $6.589bn was up 9.4% year on year. Gross margin widened to 68.3% from 62.3%, operating income rose 21.9% and net income 44.3%. The cyberattack damage sits in the first quarter, where revenue grew 2.6% and operating income fell 20.6%; the next quarter reaccelerated. Growth was not concentrated in implants — MedSurg and Neurotechnology sales of $3.6bn grew 9.2% organically against Orthopaedics' 8.6%, so more than half the company reprices on a cycle that has nothing to do with an artificial joint.

The machine underneath is placement, not price per part. Stryker's Mako installed base has passed 3,000 systems in 47 countries with more than 2.5 million cumulative procedures, and more than two-thirds of its US knee cases and about a third of US hips now run on a robot. "We can execute what depends on us, and that's what we're going to continue to do," Wells said on September 8. Canaccord trimmed its price target to $385 from $400 on the manufacturing problems — still about 40% above the current price, which is sell-side treating the move as an overshoot rather than a franchise mark-down.

On today's consensus of $15.00 for 2026, last September's Stryker price implied 25.1x forward earnings; it now trades at 18.3x, while consensus still has earnings growing 10.6% this year and 11.6% next. It is cheap against its own history and not cheap absolutely — 18.7x trailing enterprise value to earnings before interest, taxes, depreciation and amortization, and a 4.5% free-cash-flow yield.

What implant pricing actually costs

The price erosion that is supposed to define this business is disclosed, small and bounded. Zimmer's pricing headwind was 80 basis points last quarter against full-year guidance of up to 100, with 85% of the book already contracted. Mix can run the other way: "We get a 40% premium every time that we move from non-coated — non-iodine-coated hip to a coated hip," chief executive Ivan Tornos said on the August 5 call, describing a Japanese launch. China, the volume-tender risk, is declining about 20% and is 2% of revenue.

Units are the problem instead. Zimmer's knees grew 0.1% in constant currency last quarter, hips 5.1%, and revenue growth decelerated from 10.9% in the December quarter to 4.8% in June — yet the company raised its 2026 outlook to 2.25%-3.25% organic growth and $8.47-$8.59 of adjusted earnings, lifted the buyback to $1bn, and grew robotics 21.5% on record ROSA capital sales. At 11.2x forward earnings, 13.0x trailing EV/EBITDA and a 9.9% free-cash-flow yield, Zimmer's multiple is where last September's price implied it would be, near 11.7x. It carries roughly $6.9bn of net debt after the Paragon 28 and Monogram deals, and consensus has its EBITDA falling 3.7% this year.

The volume story has a named source. HCA Healthcare's preliminary second quarter flagged weak elective surgical volumes concentrated in under-65 orthopedic and spine patients losing exchange coverage, with management declining to assume a second-half rebound and full-year revenue guidance cut to $77.0-79.5bn. On that news the whole operating-room complex fell — GE HealthCare 7%, Intuitive Surgical 6%, Stryker 5.4% — and Zimmer fell least, at 3%.

The two other names in this corner point the same way. Smith & Nephew, the UK group in orthopedics, sports medicine and wound care, cut full-year growth to about 4% with orthopedics down 1%, and chief executive Deepak Nath attributed the US shortfall mostly to itself: "Fundamentally, it's Knees. We're not able to participate in the fastest-growing part of Knees, which is Cementless." It trades at 12.6x forward earnings. Globus Medical, the spine and navigation maker, raised 2026 earnings guidance to $4.95-$5.05 at 14.8x forward earnings — and watched enabling-technology revenue fall 26% as hospitals shifted from buying capital equipment to leasing it, even as deployed units rose 25%.

The verdict

Part of Stryker's de-rating is earned: one wrecked quarter, a supply drag management now expects to run into the fourth quarter, and European hips that slowed. None of that explains seven turns of forward earnings against accelerating revenue, a six-point gross-margin gain and reaffirmed guidance. And the mechanism the price move is usually blamed on — contracted implant prices grinding down a point a year — is the one variable both companies quantified and bounded. What is being marked is the hospital's willingness to place capital and the under-65 patient's willingness to book an elective knee. Zimmer, on that reading, was never de-rated at all; its estimates are the flat part, and its shares have simply tracked them.

The risk nobody has priced is the one nobody has measured. A June study suggested sustained weight-loss drug use could prevent thousands of knee replacements a year; the trial testing those drugs in hip and knee arthroplasty patients only begins recruiting in October.

The merchants are still there. Yelp ended its second quarter with roughly 510,000 paying advertising locations, down 1% from a year earlier, and Cars.com carried 19,343 dealer rooftops, essentially unchanged. What left was the traffic. Yelp's ad clicks fell 5%; Cars.com's average monthly unique visitors fell 14%.

That split matters because of how these businesses are actually paid. Yelp — the local-review platform that sells restaurants, plumbers and dentists a monthly advertising budget per business location — prices that budget internally as clicks times cost-per-click. Cars.com, which sells franchised and independent dealerships a monthly marketplace subscription bundled with websites, appraisal software and reputation tools, is paid per rooftop regardless of how many shoppers arrive. In both cases the merchant contract is the revenue and the traffic is the inventory consumed inside it. With zero-click Google searches reaching 68% in early 2026, the erosion turns up in volume long before it turns up in the customer list. What happens next depends on what else is in the bundle.

Yelp: the merchant stayed, the inventory did not

Yelp's revenue grew 1.4% to $375.5m, a fourth straight quarter of roughly flat growth. Underneath it, advertising revenue from restaurants, retail and other categories fell 10% to $102m while Services advertising was flat at $241m. Average cost-per-click rose only 1%, so the price side did not offset the 5% decline in clicks. Operating margin went from 14.4% to 11.6% and operating income fell 18%; gross profit actually shrank slightly, diluted by the $270m Hatch acquisition.

Management's framing is demand, not disintermediation: Yelp expects the difficult environment for local businesses to persist through 2026, and chief financial officer David Schwarzbach attributed better net revenue retention to onboarding, ad targeting and upselling, changes he called structural. The buyback that removed 15% of diluted shares over the past year has been paused to repay the revolver drawn for Hatch, with a resumption flagged for 2027.

The shares closed at $19.63 on 18 September, their lowest in a year. At 9.3x trailing earnings against roughly 14–15x a year ago, the de-rating is real — but the forward multiple of 10.9x sits above the trailing one, which is arithmetic for falling profits: consensus has 2026 earnings at $1.80 a share against $2.24 last year.

Cars.com: same traffic loss, opposite result

Cars.com reported the same erosion and the reverse outcome. Traffic fell 12% and unique visitors 14%, yet dealer revenue rose 3% and monthly average revenue per dealer hit a record $2,500. "Marketplace revenue grew over 7% year-over-year in Q2," chief executive Tobias Hartmann told investors on 6 August. "Outside of 2021's pandemic-related recovery, this is the fastest marketplace growth rate in our public company history." Operating income rose 83% to $27.9m on a top line up less than a percent, as the 2017 customer list finished amortizing and operating expenses fell.

The bundle is doing the work, unevenly. Premium Plus is the fastest-growing tier, but AccuTrade appraisal subscribers were flat-to-down sequentially and Dealer Inspire website units declined, with a new general manager given two to three quarters to stabilize them. Automaker and national advertising — the part not under subscription — fell 18% to $13.6m. Net debt of about $417m against guided adjusted profits implies roughly two times leverage, and first-half free cash flow of $43.5m did not cover $57m of buybacks.

What the market is actually marking down

It is not contracted per-merchant revenue. Both companies held their merchant counts and Cars.com raised its price. What is being marked down is monetizable volume, and the pass-through differs by meter: Yelp's subscription is consumed in clicks, so a click shortfall becomes a revenue shortfall within the quarter, while Cars.com's rooftop contract absorbs a traffic decline as long as leads still convert. Cars.com is up 41% from its March low; Yelp is at a 52-week low, and the eight-name local-marketplace group fell about 10% in thirty days against a flat market, with a September rate rise and a 10-year Treasury yield near 5% compressing every low-growth small cap at once. CarGurus, the nearest auto-marketplace comparison, is down under 7% over twelve months — this is not one uniform verdict on referral traffic.

Yelp is now paid by the thing eroding it. In July it licensed 330 million reviews to OpenAI for surfacing inside ChatGPT, and Schwarzbach told investors on 6 August: "We are targeting an annual run rate of $250 million in Other revenue by the end of 2028." That was the fastest-growing line Yelp sold last quarter. It is also still the smallest.

A Guggenheim analyst withdrew his price target on Accenture on 18 September, thirteen days before the company closes its fiscal year and reports, arguing the shares had come back without the demand. Applied across the three largest listed systems integrators, that test breaks the story the market has been telling about this industry all year.

The accusation hanging over the sector is substitution: that code assistants replace the billable offshore seat, which is why the Nifty IT index has fallen nearly 40% from its peak. If that were what was being priced, the businesses most levered to headcount would be marked hardest. The numbers say something else, and with Accenture's full-year print due on 1 October the difference matters to what the market thinks it is buying.

The per-employee test fails

Over the trailing four quarters, Accenture — the Dublin-based professional services firm that both advises clients and runs their applications and infrastructure under multi-year contracts, with 799,000 staff — generated $91,490 of revenue per employee. Infosys, the Bengaluru outsourcer behind the Topaz artificial-intelligence suite and the Finacle banking platform, generated $61,869 across 328,062 people. Cognizant, the Teaneck, New Jersey services firm with 356,700 employees, generated $60,673 — marginally less leverage on headcount than Infosys, off the same offshore delivery pyramid.

Cognizant recovered 41.5% from its 18 June low, within a fifth of a point of Accenture's 41.7%, and is down 12.0% over twelve months — the shallowest decline in the group. Infosys recovered 2.4% and is down 38.4%. Delivery model does not separate them.

What Infosys's quarter actually did

The June quarter took Infosys's operating margin to 21.09% from 20.81%, with gross margin up as well, on revenue of $5.082bn. Large-deal total contract value was $3.6bn, of which 61% was net new work rather than renewal — the opposite of a franchise living on vendor consolidation — with artificial-intelligence work at 8.2% of revenue and headcount down 532 sequentially. Management narrowed fiscal 2027 constant-currency guidance to 1.5%–3.0% from 1.5%–3.5%, leaving the margin range untouched. Its American depositary receipts broke to a new low in their twelve-month downtrend on 17 September regardless.

Two things outside the quarter explain the marks. The rupee is down 8.7% against the dollar over twelve months, roughly 6 points of the receipt's 38.4% fall; the same quarter grew earnings per share 14.9% in rupees while dollar net income rose 1.3%. And Infosys carries a higher revenue multiple than either peer, at 2.19x trailing sales against Accenture's 1.52x and Cognizant's 1.29x, with a forward price-to-earnings of 13.60x sitting above its 13.37x trailing — consensus modeling no earnings growth at all.

What Accenture's rally is made of

Accenture lost 4.73% on 18 September on 11.5m shares, roughly four times the prior session's volume, after Guggenheim's Jonathan Lee cut the stock from Buy to Neutral and withdrew his price target, his argument being that the shares had rebounded more than 52% from the June low against a 2% gain in the S&P 500 with channel checks showing "no corresponding improvement in demand."

He had material. June-quarter new bookings were $19.32bn, down 2% on a book-to-bill of exactly 1.0, with consulting filling more of the order book than the managed-services half growing twice as fast. The restructuring the company calls business optimisation still cost $0.40 of nine-month adjusted earnings per share. In August staff were told they could carry unused vacation into the new fiscal year to keep selling through the 31 August close; "our shareholders are counting on us to deliver a strong quarter," chief executive Julie Sweet wrote in the memo. The stock trades at 12.35x forward earnings against 21x to 37x at each of its last five fiscal year-ends — cheap against its own history, and cheaper still in late July at 10.1x, the repair since coming on flat-to-falling estimates.

The verdict

Accenture's advance is multiple repair rather than order-book recovery, which is precisely the trade Guggenheim closed. Infosys's break to new lows is earned by its price rather than its quarter — margins rose, deal wins skewed net new — and a sixth of the twelve-month decline in the receipts is currency the business never touched. What is sorting these three is the direction of estimates and the currency of the listing; the one company whose operating income actually grew, Cognizant, is also the cheapest and the best held. Labor substitution is a good story. It is not what these charts are measuring.

On 1 October Accenture closes the argument with a full year of bookings and its first fiscal 2027 guidance. In June a book-to-bill of exactly 1.0 was survivable because the shares had already been marked down for it. Nothing has been marked down since.

EMCOR Group, the Norwalk, Connecticut contractor that designs, installs and maintains the electrical and mechanical systems inside buildings — power distribution, cooling, clean-room ventilation, high-purity piping — told Morgan Stanley's Laguna conference on September 17 that it now serves about 50% to 60% of the electrically relevant data-center markets nationwide and 40% to 50% of the mechanically relevant ones. It is the first time the company has sized its own share of that work, and chief executive Tony Guzzi said it deep into a selloff in the shares.

The share matters because the content per site is rising. A traditional cloud data center runs 20 to 75 megawatts; an artificial-intelligence campus is often 200 megawatts or more across several buildings over three to five years, and at that scale EMCOR told the conference the mechanical scope runs 1.5 to 2 times a cloud project's and the electrical scope 1.25 to 1.5 times, mostly because of cooling. More than half a market, on jobs worth more per megawatt, is the whole investment case — and it is being marked down.

The book grew; the guide went up

June-quarter revenue rose 19.8% to $5.155bn, the fourth straight quarter near 20%, and operating margin reached a record 10.62%. Remaining performance obligations — signed, unbilled work — hit $17.14bn at June 30, up 43.9% and roughly 95% organic, with the largest gains in network and communications on data-center contracts. Against first-half revenue of $9.78bn, that is bookings running about 1.4 times billings. On July 30 the company raised full-year earnings guidance to $32.00–33.25 a share from $29.75–30.75. Consensus for the year sits at $32.95, inside the raised range. Estimates have not been cut.

The shares fell 12.8% in the thirty days to September 17 and sit 20.7% below their 2026 high, still up 21.4% over twelve months. What changed in the past week was the price of money and the mood around the buyer. The 10-year Treasury yield touched 5.014% on September 14, its highest since October 2023, and the buildout complex sold off after Anthropic chief executive Dario Amodei called for a global slowdown in frontier-model development — while Alphabet, Microsoft and Meta, the companies writing the checks, rose.

Where the business is genuinely exposed

Duration. EMCOR's chief financial officer told investors on the July 30 call: "Historically, we would say that 85% or so of our RPOs burn in 12 months. where we're sitting today, it's more like 75% or 76%." A longer book is a more distant cash flow, and a 5% long bond bites hardest there. Mix, too: mechanical construction margin fell 110 basis points to 12.5% as roughly a tenth of that book shifted toward guaranteed-maximum-price forms where equipment passes through at thin markup, even as electrical widened 210 basis points to 13.9%. And the recurring maintenance business is no ballast — US building services grew 5.6% to $837.7m at a 7.6% operating margin, against $3.96bn of construction trades earning 13.1%.

Quanta Services, the Houston utility contractor whose transmission and distribution work is billed largely under multi-year master agreements to regulated utilities, fell 12.1% over the same stretch on revenue up 41.1% to $9.557bn and a record $53.4bn backlog. Its 2025 annual report discloses no customer at 10% or more of revenue, the largest at 8% — so the fear that one hyperscaler's pause empties the book has no anchor in either company's filings.

The verdict

The business earns none of this drawdown and the estimate sheet proves it; what the market repriced is the rate at which a lengthening book of contracted work is discounted, plus its willingness to pay for anything with artificial intelligence attached. EMCOR's forward multiple has gone from roughly 26x in mid-August to about 22x — the same anchor it carried in May, after a year in which signed work grew 44%. On forward gross profit it trades at 8.1x against Comfort Systems' 17.9x, a gap that gross margins of 19.8% versus 25.9% only partly explain. Quanta, at 36.9x forward against EMCOR's 22.1x and Comfort's 34.5x, has come down a third from the roughly 53x it carried in May but remains the expensive one of the three.

Guzzi sized the franchise on the day the shares were the cheapest they had been since spring. The share of the market is not in doubt; what a dollar earned in the fourth year of a five-year campus is worth now is.

Ultragenyx Pharmaceutical, a Novato, California company that sells four rare-disease biologics and develops gene and antisense therapies, received full approval on 17 September for FAYUVI, the first treatment ever cleared for Sanfilippo syndrome Type A. It was the second approval in a month, after GENGLYCOS for glycogen storage disease type Ia on 19 August, the company's first gene therapy ever authorized. Across the same thirty days the stock fell 47.4%.

That is not a contradiction so much as an accounting lesson. An orphan-drug company earns money in three ways that look identical on a revenue line and are valued nothing alike: product it books itself, patient by patient; a royalty on a partner's sales; and pipeline assets that are worth a great deal until the day they are worth zero. The market pays roughly 12.0x forward gross profit for Ascendis Pharma, 4.9x trailing for BioMarin and 2.3x forward for Ultragenyx — a fivefold spread on gross margins of 91%, 80% and 84%, close enough that the comparison is fair.

What was marked to zero

On 3 September the Phase 3 Aspire trial of apazunersen missed both its primary cognitive endpoint and its key secondary responder index, with no difference between treated and control groups. Shares closed down 44.0%, from $26.53 to $14.85. "Based on everything we observed in the robust Phase 1/2 clinical development program and long-term extension study, we are disappointed by the Aspire result," chief executive Emil Kakkis said that day. Analysts had modelled peak sales above $1.8bn; Evercore ISI cut its target to $16 from $34. Management now plans "significant expense reductions" and a path to profitability in 2027.

The de-rating of that line item is earned. The base underneath it moved the other way. Second-quarter revenue of $214m grew 28.5% year on year, against a 2.4% decline in the first quarter. Crysvita, for X-linked hypophosphatemia, contributed $156m of it — and US and Canadian commercialization passed to Kyowa Kirin in April 2023, so $94m of North American revenue and $8m of European is royalty rather than sales Ultragenyx makes. First-half royalty revenue rose 11%, to $149m, on more patients treated. The owned franchises are genuinely ultra-orphan: about 675 reimbursed North American patients on Dojolvi against $27m in the quarter works out near $160,000 a patient a year.

The royalty also leaks. In November 2025 Ultragenyx took $400m from OMERS for an additional 25% interest in the same US and Canadian stream, with payments beginning January 2028 and capped at 1.55 times the purchase price. Reported royalty revenue therefore overstates what the company will keep.

The other kind of dollar

Ascendis Pharma, the Danish maker of long-acting hormone replacements built on its TransCon prodrug chemistry, is the clean version of the annuity. Product revenue reached €315m in the second quarter, up 105%, of which €252m was Yorvipath for hypoparathyroidism — 93% of the top line is drug sold to patients. Roughly 1,000 new US patients start each quarter and 95% remain on therapy after five years. Operating margin swung from −33.5% a year ago to +18.4%. "The majority of the drop-off is during that titration period," Jay Wu, president of Ascendis U.S., told investors on 13 August, naming the dosing ramp rather than payer rebates as the leak.

For that, the market pays 10.9x forward sales against 12.3x trailing — the forward figure sitting below the trailing one is the growth being bought, with consensus 2026 revenue of $1.37bn. Shares are up 20.4% over twelve months and 3.9% over three, a fraction of a doubling in revenue, and 9.5% of that was given back on 15 September after Novo Nordisk's metabolic rights reverted, leaving Ascendis to fund obesity programs itself. BioMarin, which sells seven rare-disease products on $3.22bn of 2025 revenue, grew 19.9% last quarter and sits between the two on price.

The verdict

Across the orphan names the three-month spread between best and worst runs about 70 points, and each of the largest moves traces to a dated single-company event rather than a shift in rates. The market got the direction right on Ultragenyx and has not yet re-examined what it left behind: 1.9x forward sales, $436m of cash, and a business whose fastest-growing line is a royalty it has partly pre-sold and does not control the selling of. Guidance of $730m–$760m for 2026 excludes both new launches entirely.

What it also left behind are two tradable priority review vouchers, awarded with the August and September approvals, in a market where single vouchers cleared at $180m and $205m this year — against a company the market now values at $1.4bn.

Centrus Energy raised half a billion dollars on 9 September and the market has spent the seven sessions since deciding what it cost. The Bethesda, Maryland company — which sells separative work units, the unit of enrichment service, rather than uranium — priced an underwritten offering of Class A stock and warrants at a combined $199.64 per unit, with common warrants over up to 6,992,382 further shares at exercise prices from $226.8625 to $362.98. Against the 22.5m diluted shares Centrus reported for the fourth quarter of 2025, that overhang is roughly 31% of the company. The stock closed at $185.38 on 8 September and $146.00 on 17 September.

That is the news, and it landed in a week when everything around it fell too. Uranium spot was $90 a pound on 14 September, the long-term contract price near an 18-year high, and Kazatomprom — the world's largest producer — has said it will cut about 8 million pounds from 2026 output, roughly 5% of global supply. Supply tightened; the equities broke. Between 8 and 17 September, Energy Fuels fell 19.6%, Uranium Energy Corp 16.1%, Cameco 8.9% and BWX Technologies 8.7%. What moved in that window was the cost of money: the 30-year Treasury yield reached a 19-year high in August and the 10-year pushed above 5.04% on 15 September, its highest since 2007, which is the discount rate applied to cash flows that begin in the 2030s.

Two ways to earn a fuel-cycle dollar

The two protagonists are paid on different meters, and both meters read better than their share prices. Centrus's June-quarter revenue rose 14% to $176.1m while gross profit fell 7.4% to $49.9m and operating income fell 69% to $10.4m. The growth was mix: enrichment revenue rose 22% to $153.4m on uranium resales of $53.4m, with actual separative work volumes down 23%, pricing up 3% and unit costs up 13%. Its government-funded half — Technical Solutions — fell 21% to $22.7m after the Department of Energy high-assay low-enriched uranium operations contract that paid for it expired on 30 June. Set against that: total backlog nearly doubled to $4.5bn stretching to 2040, $2.4bn of the enrichment piece now under definitive rather than contingent agreements, $1.9bn of cash, and the same-day long-term partnership with Radiant to fuel Kaleidos microreactors. Commercial production at Piketon is guided to 2029. Management told the August call that utilities remain in a "wait-and-see" posture until centrifuges are installed.

Cameco, the largest listed uranium producer, is the opposite case: its realised price rose while its stock fell. Second-quarter realisation was C$93.13 a pound, up 15% — about US$68, against that $90 spot — with 2026 unit costs guided to $63.00–$67.50 and term contracts covering average deliveries above 28 million pounds a year for five years. "We are back into a mid-'90s long-term uranium price on its way to three digits likely, and this is in the absence of replacement rate demand," chief operating officer Grant Isaac told investors on the 31 July call. What fell was the other half of the company: Cameco's share of Westinghouse adjusted earnings before interest, taxes, depreciation and amortisation dropped to US$163m from US$352m, and the company withdrew its framework outright — "we are no longer providing a five-year growth outlook for Westinghouse."

What the break did and did not discriminate

If the market were separating contracted pounds from government funding risk, BWX Technologies would have been spared: the naval-reactor and medical-isotope maker grew revenue 18% to $901.6m, carries $8.4bn of backlog, up 40%, and raised all four 2026 guidance lines. It fell with the rest. The likelier reading is a single rates-and-artificial-intelligence-capex trade in names held as data-center power proxies.

So the businesses earn part of this and not the rest. Cameco's per-pound economics are intact and its shares are still up 13.5% over twelve months; what nothing in the contract book explains is a forward earnings multiple of 60x, against roughly 56x in May, on consensus revenue growth of 4.2%. Centrus is the harder case: expensive at 58.9x forward earnings on consensus profit falling to $2.55 a share this year from $3.90, with free cash flow negative while Piketon is built, and now with a third of its share count sitting in warrants above $226.

The backlog belongs to the 2030s. The dilution is dated, priced and already on the record.

Roper Technologies buys software companies for a living. This year, its largest purchase was itself.

The owner of Deltek (project accounting for government contractors), Vertafore (insurance agency systems) and Aderant (law-firm practice management) has put $3.2bn into its own stock in 2026, retiring 9.0m shares, including 3.6m for $1.2bn in the June quarter alone, according to its second-quarter disclosures. Diluted share count fell from 108.4m a year earlier to 102.6m. That 5.4% reduction, rather than any change in what the market will pay for the assets, accounts for most of what a shareholder earned per share.

This matters beyond one quarter of capital allocation because buying is the model. Roper describes itself as 29 businesses, 21 of them software, about $8.5bn of revenue, 40% EBITDA margins and free-cash-flow margins in the low 30s. Reported growth is capital deployment stapled to whatever the existing units produce on their own. Take deployment away and the underlying figure is visible: reported revenue growth has stepped down from 14.3% in the September 2025 quarter to 8.5% in June, of which five points were organic.

The arithmetic that pointed inward

Roper's stated test is that an acquisition must underwrite at roughly six to eight times year-five EBITDA, against a buyback that management frames as equivalent to about ten times. Few private assets met the standard this year. Net debt now sits at 3.4 times EBITDA, against roughly $5bn of annualized deal capacity, and the near-term plan is to rebuild that room by reducing debt. Chief executive Neil Hunn told the Goldman Sachs Communacopia + Technology Conference on 9 September that sponsor conversations have become "decidedly more constructive" as sellers accept the new valuation reality, and that he expects mergers and acquisitions to beat buybacks over a five- to seven-year horizon. At the same appearance management conceded that organic growth "has been harder to execute than hoped."

The operating business is not deteriorating. Second-quarter free cash flow rose 11% to $447m, trailing twelve-month free cash flow is $2.6bn, and full-year adjusted earnings guidance was raised for the second time, to $22.15–$22.30 a share with organic growth "in the 6% range." The Application Software segment grew 8% to $1.18bn on 5% organic growth, with enterprise gross retention in the mid-90s. Reported net income is less useful: the quarter's $1.169bn includes a $995.9m fair-value gain on Roper's stake in Indicor, whose instrumentation sale should bring about $1.3bn before tax, cash guidance excludes.

What the market has not done

Roper trades at 16.9 times consensus 2026 adjusted earnings of $22.22, against roughly 26.6 times on its 2025 average share price and 2025 earnings — a de-rating of about a third that this year's advance has not repaired. Measured against gross profit, the standard comparison when margins differ across a group, the shares cost 6.57 times trailing profit today versus 6.51 times in early May. The move off the June low is arithmetic, not appetite.

The de-rating was never about Roper. UBS downgraded ServiceNow on 10 April and cut its target to $100 from $170, arguing that artificial-intelligence infrastructure spending is crowding out core software budgets, with more than half of enterprise conversations now including talk of containing non-AI software spend. ServiceNow, which sells workflow automation to large enterprises, grew revenue 24% in the June quarter while its gross margin fell to 70.7% from 77.5% and operating income halved to $162m; at 34 times forward earnings it remains the most expensive name in this group. Hunn's answer is that Roper has seen no seat compression across its 21 software units, and that it is moving toward charging credits drawn down per task performed by software agents, on a use-it-or-lose-it basis.

Contrast the meter at Procore Technologies, which prices construction-management software against the annual building volume a customer runs through the platform rather than by user: customers above $100,000 of annual recurring revenue rose 14% to 2,871 and now supply 68% of that revenue. Its shares cost less per dollar of gross profit than in the spring, 6.46 times forward against 7.96 times on 3 May. Improving disclosure, shrinking price.

So the honest reading of Roper's year is that the market has re-priced neither durability nor its absence. It has simply been handed fewer shares. Buybacks financed at these levels are accretive only while the de-rating persists, and 3.4 times leverage sets a limit on how long that can run. If sponsor prices come in where Hunn expects, capital swings back to deals and the compounding resumes on the old terms; if they do not, mid-single-digit organic growth is the entire story, and it is the number management has already called harder than hoped.

Roper reports third-quarter results on 22 October. The line to read first is not earnings per share — the denominator has been doing that work — but the organic growth beneath it.

The one place AAR Corp's margins slipped last quarter was the business of buying used jet parts, and the reason was that airlines have stopped retiring airplanes. AAR — which distributes new and used aircraft components under exclusive manufacturer authorizations, runs heavy-maintenance and paint hangars, and overhauls landing gear, wheels and brakes — saw its Parts Supply segment sell $424m in the quarter ended 31 May, up 39%, while segment adjusted earnings margin fell 250 basis points. Management's explanation on the fourth-quarter call was that asset availability was constrained, so AAR had to pay more for the airframes and engines it harvests, narrowing the spread between what it buys and what it sells.

That is a shortage of feedstock, and it is the inverse of the story the market bought in September. Teardown material is scarce because operators keep flying ageing narrowbodies instead of parting them out — which is the same fact that keeps hangars full and shop visits booked. The distinction matters because the mid-September repricing of this entire industry rested on a forecast that flying itself is about to slow.

What actually happened in September

Guggenheim initiated coverage with Neutral ratings on AAR, HEICO, StandardAero, TransDigm and VSE, forecasting global passenger traffic growth of 1.8% in 2026 against 5.4% last year and warning that higher fares and further retirements could weaken parts demand within six to twelve months. Days earlier Melius Research cut GE Aerospace, HEICO, Honeywell's aerospace arm, TransDigm and Woodward to Hold, expecting aftermarket growth to decelerate to high single digits next year; its TransDigm target went to $1,331 from $1,477. Both calls landed on a week in which the 10-year Treasury yield reached 5.04%, the highest since 2007, after the Federal Reserve's first rate increase in three years.

The result was uniform. Over the thirty days to 17 September, HEICO fell 19.4%, Howmet 21.7%, Astronics 29.5%, StandardAero 17.3% and TransDigm 12.6%, while the S&P 500 exchange-traded fund fell 1.6%. AAR's 19.6% and VSE's 26.2% are the same trade, not a two-company problem.

The meters that pay these companies

AAR's fiscal 2026 revenue rose 19% to $3.308bn and operating margin widened from 6.66% to 8.18%; fourth-quarter sales of $928m rose 23% with adjusted earnings before interest, taxes, depreciation and amortization up 27%. Net leverage ended the year at 2.03x. Guidance for the current quarter is sales growth of 21% to 23% and an adjusted EBITDA margin of 12.25% to 12.75%, and chief executive John M. Holmes reaffirmed nothing less at the Jefferies industrials conference on 9 September.

VSE Corporation, which sold its fleet and federal-services arms to become an aviation parts distributor and repair shop, grew second-quarter revenue 65% to $449.1m — 14% of it organic — and expanded gross margin by 562 basis points to 17.02%. "Adjusted EBITDA nearly doubled year-over-year, significantly outpacing revenue growth and adjusted EBITDA margins reached a record 19.2% in the quarter," chief executive John Cuomo told investors on 5 August, the day the company raised full-year revenue guidance to 61–64% growth. The shares rose 5.3% on that print and made an all-time high twelve days later. The entire de-rating is post-beat.

The third meter is engine values, and it has not broken. Willis Lease Finance, the engine lessor and spare-parts trader, sold 21 engines at a 14.2% margin in the June quarter, lifted its average lease rate factor to 1.03% from 1.00%, and told investors its portfolio appraises about 20% above book. Its warning sits elsewhere: utilization fell to 85.0% from 87.2% and short-term maintenance reserves to $39m from $50.2m as operators cut hours on fuel-thirsty CFM56 and V2500 fleets. Fewer legacy hours flown now means fewer legacy shop visits later — the one honest strand of the bear case, and a slow one. The replacement wave is real but gradual: Boeing delivered 418 jets through August and Airbus 475, while Pratt & Whitney still expects roughly 350 geared-turbofan aircraft grounded per day on average through the end of 2026.

What the businesses earn and what they don't

AAR's forward multiple has fallen from 25.9x next-year consensus earnings at its 18 August peak to 20.6x, against consensus growth of 17.3% — roughly the growth it implies, with a trailing free-cash-flow yield of 1.16% as the offset, because buying authorizations and carrying inventory consumes cash. VSE is the name where the markdown has a foundation: forward earnings of 26.6x this year against about 36x at the peak looks cheaper, but trailing enterprise value is 33 times EBITDA, trailing free cash flow is nil, and $872m of net debt at 2.4x sits partly on a $900m floating-rate term loan raised for the $2.025bn Precision Aviation Group deal. A 5% funding curve is a real cost there in a way it is not at AAR.

So the split is this: nothing in either company's disclosed numbers deteriorated, and the de-rating is a multiple event built on a traffic forecast for a year that hasn't started. What the businesses have genuinely earned is narrower — a squeezed used-parts spread at AAR, and at VSE a balance sheet that converts almost none of a record margin into cash while rates rise.

AAR reports its first fiscal quarter after the close on 29 September. The line to read is not the sales number, which is already guided, but whether the buy-sell spread on used material widened again — which will depend on whether airlines have started parting out their old narrowbodies at last.

Home Depot's biggest baskets are still being rung up. Comparable transactions over $1,000 — the roof replacement, the kitchen, the riding mower — rose 2.4% in the quarter reported on 18 August, according to trade coverage of the results, and 13 of the retailer's 16 merchandising departments posted positive comparable sales. Monthly comps improved through the quarter, from 0.5% in May to 2.2% in July.

That is not the record of a business the market has marked down by roughly a quarter in a year, and the gap is the story of the whole repair-and-remodel complex right now. What has been repriced is the rate at which American houses change hands, not what the two largest home-improvement retailers reported. Inside that repricing, exactly one of the two growth engines investors paid up for has actually been impaired — and it is not the $18bn one.

The basket has two halves

A home-improvement dollar is earned per basket, and the halves move independently. Home Depot's comparable transactions fell 1.0% while comparable average ticket rose 2.8% — fewer projects, each costing more, with the average transaction at $92.50. Lowe's, the more do-it-yourself-weighted of the two, saw transactions fall 2.1% as customers chose repair and maintenance over renovation.

What opens a big basket is a house changing hands, and houses are not changing hands. Existing-home sales fell 2.0% in August to a 3.98m annual rate, a third straight monthly decline and a 14-month low. Home Depot management puts turnover at roughly 3% of the housing stock. Financing moved the wrong way through the period: Freddie Mac's 30-year fixed averaged 6.76% in the week to 10 September against 6.35% a year earlier, with daily screens above 7%, after the 10-year Treasury crossed 5% and the Federal Reserve raised rates a quarter point on 16 September.

"While consumer uncertainty and housing affordability continue to pressure demand for larger home improvement projects, we remain focused on what we can control," chief financial officer Richard McPhail told investors on 18 August.

The engine Home Depot bought is running

SRS Distribution, acquired for $18.25bn in 2024, is not a store: it sells roofing, pool and landscape materials per job to contractors on trade credit and delivers to the site from hundreds of branches. SRS then bought drywall distributor GMS for $5.5bn, closing in September 2025, taking Home Depot's spend on stapling a wholesale distributor onto a retailer's income statement close to $24bn. Revenue growth has accelerated for three quarters, from −3.8% to 4.8% to 5.7%, and SRS comped above the company average with positive sales in every vertical.

The acceleration is bought, and the price shows. Acquired distribution diluted gross margin by 60 basis points, operating margin slipped to 14.29% from 14.48%, return on invested capital fell to 24.8% from 27.2%, and inventory turns eased to 4.5x. A $730m refund of tariffs collected under the International Emergency Economic Powers Act added 145 basis points to gross margin and masked input-cost pressure; McPhail told the same call he expects fourth-quarter gross margin "to be relatively flat versus last year." The full-year outlook first issued in December 2025 — comparable sales flat to 2%, earnings per share flat to up 4% — has not been cut.

Floor & Decor slowed the compounding

The hard-surface flooring specialist is the genuine growth-engine casualty. Comparable store sales fell 2.1% in its quarter ended June, and the reported blowout — 48.2% gross margin, operating income up 51.4% — is a $56.2m tariff refund worth 450 basis points. Adjusted gross margin was 43.9% against 43.8%. "We are pleased to have delivered adjusted diluted earnings per share of $0.58, unchanged from the prior year period, despite a 2.1% decline in comparable store sales driven by continued softness in large discretionary flooring projects," chief executive Brad Paulsen said on 30 July.

Operating expenses rose to roughly 38.3% of sales from 37.1% — the fixed occupancy and pre-opening cost of putting up warehouses into negative comps. So the company cut the opening rate to 20 new stores for 2026 from 25 in 2025, on a 281-store base against a stated 500-store target, and shrank the format to about 55,000 square feet. Capitalized at roughly 45x earnings a year ago, it trades at 21.8x trailing and 23.7x forward — the forward multiple sits above the trailing one because the market is discounting an earnings base inflated by refunds. Guidance still contemplates comps anywhere from down 4% to flat.

The market did not discriminate

Lowe's comped 0.2% against Home Depot's 1.7% and moved full-year adjusted earnings to about $12.25 a share, the low end of its range; its shares fell 27.0% over twelve months against Home Depot's 26.8%. Home Depot trades at 21.3x trailing and 20.3x forward against roughly 28x a year ago; Lowe's at 16.6x and 16.0x. Over the trailing month Builders FirstSource fell 14.2%, Installed Building Products 17.2% and RH 28.7%, against 2.4% for the S&P 500 tracker. Home Depot's own slide from $344.30 on 19 August to $305.48 on 15 September came in nineteen sessions with none worse than about 2%, and no company news to pin it on beyond a Bernstein target cut to $344 on 9 September and chief executive Ted Decker's temporary medical leave announced on 12 August.

Floor & Decor's de-rating is earned: flat adjusted earnings, negative comps and a self-cut unit-growth rate are what a compounder losing its compounding looks like. Home Depot's is not earned on the quarter it reported — but consensus has it earning $15.01 in the year to February 2027 against $14.23 actually earned last year, so 20x is cheap only against its own history, not against the growth being guided to. The sharper tell sits outside both names: POOL Corp and SiteOne, the listed pure-plays in two of SRS's three verticals, are down 47.9% and 34.2% over twelve months. Per-job contractor distribution is being marked down harder in public than Home Depot's own disclosure of it suggests.

The November quarter is the test of whose meter is right. Home Depot has kept a guidance range intact through three rate moves against it; if the transaction that triggers a renovation keeps not happening, the ticket half of the basket has to carry a second year alone.