DK Street Journal

Corning sold $2.07bn of optical fiber, cable, connectors and hardware in the June quarter — 49% of company sales, and up 32% on a year earlier. The enterprise half of that business, the part that ships into data centers, grew 65%, and AI data-center sales nearly doubled. Segment net income rose 77% to $438m on a 21% net margin, the best the segment has recorded. Carrier sales were flat.

That is now the company. Corning, which also makes glass substrates for panels, Gorilla-type cover glass, ceramic emissions substrates for vehicles and lab consumables, has become a fiber manufacturer with side businesses, and it says demand for its high-density optical products still runs ahead of what it can produce. Multi-year capacity agreements signed this year with Meta — up to $6bn — with NVIDIA and with Amazon are turning long-term contracts into the majority of the optical book. Which makes the basis on which all of it gets reported worth reading closely.

Two sets of books, both legitimate

On reported figures, Corning's June quarter produced revenue of $4.505bn, up 16.6%, operating income of $698m, up 21.8%, and diluted earnings of $0.64 a share. The figures management leads with are core sales of $4.74bn and core earnings of $0.78 — 5.2% and 22% above the reported lines. The gap principally reflects an adjustment for hedged currency exposures that Corning adopted prospectively on April 1, 2026, replacing the constant-currency convention under which it had translated the yen at ¥120 for 2025 and 2026, plus non-cash tax and restructuring items, per the quarter's earnings exhibit. The market rate is nowhere near ¥120: spot traded between ¥160.38 on September 2 and ¥154.36 on September 7.

This is not cosmetic, because the plan investors are underwriting is stated in sales. "In the second quarter, we delivered outstanding results, and we upgraded our Springboard Plan to grow sales to an annualized run rate of $20 billion by the end of 2026, $30 billion by the end of 2028, and $40 billion by the end of 2030," chairman and chief executive Wendell Weeks said on the July results. June-quarter core sales annualize near $19bn; the reported ones near $18bn. Corning's own risk-adjusted version of Springboard is lower again — $27bn by end-2028 and $35bn by end-2030 — with the timing of optical scale-up adoption named as the largest single adjustment factor.

The AI build pays one segment and taxes the other

Glass Innovations — display substrates and cover glass, merged into a single segment in the first quarter of 2026 — grew 1% to $1.46bn, with net income of $354m. The reason is one price: Corning expects handset units down a mid-teens percentage this year because memory prices have inflated the cost of building a phone. The same data-center spending that fills the fiber order book is what shrinks the market for phone glass.

The shares closed at $154.29, 39.7% below the June 29 peak of $255.69 and down 13% over three months, though still up 117% over twelve; the 50-day average has sat below the 200-day since mid-August. Forward earnings are 47.1x against 69.8x trailing, and trailing stood near 88x in early May. Consensus has not moved against the company — $3.28 of earnings this year and $4.32 next — but the sell side has split on price, with JPMorgan cutting to $170 at neutral, Barclays to $129, and UBS reiterating a buy at $196. The 5.68% gain on September 4 came with no Corning-specific news and on the day Broadcom reported AI semiconductor revenue of $16.7bn; the likelier reading is a sector-wide bid for optics rather than anything from Corning.

What the business earns and what it does not

The fall from June is premium removal, not a markdown of the franchise: revenue growth has decelerated only gently, from 20% in the March quarter to 16.6% in June, and margins expanded rather than compressed. What the business does not yet earn is the gap between the two plans and the two profit lines. An investor buying Corning at 47x forward earnings is buying management's internal Springboard rather than its risk-adjusted one, and quoting a per-share number that sits 22% above the one in the filing. Both are defensible; neither is the same as the other.

The September quarter is guided to $4.9-5.0bn — the first full period reported entirely under the new hedged-exposure adjustment, and the first clean read on how wide that wedge stays.

Quanta Services books hundreds of millions of dollars a quarter of data-center electrical work that never appears in its backlog, and management says roughly 95% of the power generation it expects to build is not in there either.

That matters because the order book is the number investors trade these companies on. Quanta, a Houston specialty contractor that self-performs 80-85% of its work building transmission lines, substations and distribution for utilities and technology customers, and EMCOR Group, the Norwalk, Connecticut mechanical-and-electrical contractor that wires and cools data halls and separately runs a recurring building-services business, both reported record signed work on July 30 and both raised full-year guidance. Both stocks are roughly a fifth below their May 6 levels, over a stretch in which the S&P 500 exchange-traded fund rose 5.0%. Late August took the pair down further before they steadied into early September; the question the order books answer is whether contracted work is thinning, or whether the same work is simply getting paid later.

Two numbers, one book

Quanta's June-quarter backlog was a record $53.4bn against remaining performance obligations of $33.6bn, per the company's release. The $19.8bn gap is the master-service-agreement layer: estimated orders under multi-year utility agreements, including estimated renewals, which do not qualify as accounting performance obligations. Neither figure is the ceiling. Quanta admits a data-center project to backlog only once a limited notice to proceed exists, and its data-center master agreements book and bill within the quarter without ever passing through the reported number. Technology and load-center work is now 15-20% of revenue.

The income statement is doing what the mechanism implies. Revenue grew 41.1% to $9.557bn, the fourth straight quarter of acceleration from 15.6% a year ago, and gross margin widened 296 basis points to 16.2%. Guidance went up, to revenue of $39.3-39.7bn. Asked on the July 30 call whether crews were the binding constraint, chief executive Earl "Duke" Austin said: "We're nowhere near capacity." The company spends about $250m a year training craft labor against a four-year journeyman pipeline.

EMCOR's book got longer

EMCOR's remaining performance obligations hit a record $17.14bn, up 44% year over year with 95% of that organic — against first-half revenue of $9.78bn, bookings of roughly 1.4 times what it billed. Operating margin reached a record 10.6% and full-year earnings guidance rose about 10% to $32.00-33.25 a share.

The honest complication is inside the mix. Electrical construction, the scarce-labor business, grew 24% and added 210 basis points of margin to 13.9%. Mechanical grew faster, at 31%, and lost 110 basis points to 12.5% as roughly 9-10% of that book moved to guaranteed-maximum-price and construction-manager forms where equipment passes through at thin markup. And the work pays out later. "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%," chief financial officer Jason Nalbandian told investors on July 30, citing project size and the water and wastewater mix. Management also said the record quarterly margin is unlikely to repeat. EMCOR bought five union electrical contractors this year — B&B Electric, Sidney Electric, Giles, Schmidt Electric and Connelly Electric, together $625m of revenue for about $750m upfront — buying crews rather than backlog.

What actually broke, and where

The fixed-price accident the sector fears did happen — at Primoris Services, the pipeline and renewables contractor, which cut 2026 adjusted earnings guidance to $4.80-5.00 from $5.80-6.00 on renewables cost overruns. It is the only name in this group down over twelve months.

IES Holdings, the electrical contractor serving data centers and housing, looks like a catastrophe and is not one. Its unadjusted price series shows a 52.7% one-day drop on August 24; the company distributed a two-for-one stock split after the close on August 21. Its June quarter grew revenue 40% with backlog of $4.5bn, and it trades at a higher price per dollar of gross profit than EMCOR does.

The verdict

Nothing in either order book earns the decline. Quanta's price against a dollar of its trailing gross profit fell from about 29.3x in early May to 19.86x while that gross profit grew 16% — the multiple did the falling. Part of that is a rational de-rating of an extreme: at 32.9x enterprise value to EBITDA and 37.3x forward earnings, Quanta is priced far above EMCOR on both measures, and consensus 2026 earnings sit inside management's own guidance rather than above it. EMCOR is the cleaner case, at 22.9x forward earnings — almost exactly where it stood in May, because guidance rose as the shares fell.

What the fall does earn is duration. EMCOR's signed work converts over a longer horizon than it used to, Quanta's largest transmission corridors are still in engineering, and a 30-year Treasury yield above 5.2% discounts cash that arrives later. No company-specific disclosure explains the August leg down; the likelier reading is a rate-and-sector repricing that arrived while the order books were still growing.

The biggest scopes in Quanta's book are not crews in the field but drawings — its largest corridors do not reach field execution until the second half of 2027. Until then, the buildout pays its contractors on paper, and paper is what a long bond marks down first.

Brookfield is dismantling the share class that made its two listed infrastructure vehicles awkward. On July 21 the boards of Brookfield Renewable, which owns contracted hydro, wind, solar and storage across the Americas and Europe, and of Brookfield Infrastructure, which owns regulated utilities, transport, midstream and data infrastructure, each agreed to convert their paired corporate share into a single publicly traded corporation, exchanging one share for one unit. Special meetings are set for October 14 and closing is expected in the fourth quarter.

A one-for-one exchange makes the arithmetic unavoidable: any premium the corporate share carried over the partnership has to go to zero. It has. Brookfield Infrastructure Corporation, whose economics are a claim on the same cash flows as Brookfield Infrastructure Partners, closed at 1.320 times the partnership on January 2 — a 32% premium — and at 1.006 times on September 4. Brookfield Renewable Corporation has run the same course, from 1.428 times to 1.011 times. Over six months each corporate share underperformed its own partnership by roughly 19 percentage points: the infrastructure share fell 23.4% against the partnership's 4.7%, and the renewable share fell 22.3% while its partnership rose.

The same cash flows, two trends

The cleanest proof that this is not a business event is that the paired securities moved in opposite directions for months. Brookfield Renewable Corporation has been in a sustained downtrend since May 11 — two months before the announcement — while the partnership was simultaneously in a strong uptrend, its shorter average above its longer. Brookfield Infrastructure Partners held that uptrend through September 3 while its corporate twin was falling. Two securities entitled to identical earnings cannot be telling two stories about the business.

And the business is not the problem. Brookfield Renewable reported second-quarter funds from operations of $421m, a record, at $0.62 per unit, up 11% per unit from a year earlier. It raised the quarterly distribution to $0.392, an annualized $1.57, against $2.14 of funds from operations per unit over the last twelve months — a payout near 73%, nowhere near the strain a levered yield vehicle shows when it is funding a dividend out of sales. Corporate debt is $3.5bn at an average 4.6% over 13 years and entirely fixed; 96% of the $32bn of non-recourse project debt is fixed too, leaving only a sliver floating into higher coupons. Brookfield Infrastructure grew funds from operations 10% per unit and raised its distribution 6%, at a 66% payout.

One number does support the worry. Roughly $175m of that $421m quarter — about 42% — sat in the hydro segment as other income, gains on developed and noncore assets rather than electricity sold, and management said it expects that line to scale as recycling becomes systematic. Chief executive Connor Teskey framed the growth on the first-quarter call as capturing development margin: building wind and solar in-house and selling them down to buyers with a lower cost of capital. Some $2.2bn of sale proceeds were agreed or closed in the quarter, $630m of it net to Brookfield Renewable, at or above target returns.

What actually re-rated the units

Brookfield Renewable's own trend broke only in the first days of September, after a 14.2% three-month decline that took the units to 14.7 times trailing funds from operations and a 4.99% distribution yield. Against a 30-year Treasury that stood at 5.25% on September 4 — having reached its highest level since 2007 in mid-August on government-spending and inflation worries — that spread is gone. The whole category went with it: Ormat, the geothermal operator, fell 26.0% over three months and Clearway Energy 21.2%. NextEra Energy fell 2.6%.

At the parent, Brookfield Corporation is down 10.5% over three months on 14.5 times forward earnings while distributable earnings before realizations grew 15% per share, fee-bearing capital rose 19% to $672bn and fee-related earnings rose 20%. "We were active through the first six months of the year—raising $98 billion of capital, deploying $100 billion into large-scale opportunities, and monetizing $40 billion of assets at attractive returns," president Nick Goodman said on August 13. Brookfield Asset Management, the pure fee vehicle, rose 8.3% over the same three months. Whatever is being marked down at the parent sits in its balance-sheet, insurance and carried-interest half rather than the fee engine.

The verdict

Three declines, three causes, none of them the funding mechanism the structure was supposed to hide. The corporate shares lost a premium by design, and that loss is complete rather than ongoing. The partnerships lost a valuation cushion to the long bond, a discount-rate move applied to contracted cash flows that grew through it. The parent lost ground its own operating disclosure does not explain. The residual worry is the composition of Brookfield Renewable's earnings: a recycling program running at record volume is a real profit stream, but it is lumpier than a power purchase agreement, and the more of the distribution it covers, the less the payout ratio tells you.

The conversions leave Brookfield's ownership, the preferred units, the public debt and the management fees exactly where they were. What they remove is the second price. After October 14, holders of the same assets stop arguing about which wrapper they are worth more in — and start arguing about the assets.

SolarEdge Technologies, which sells string inverters paired with per-panel power optimizers, plus batteries and monitoring software, to installers and electrical wholesalers, told investors on August 5 that its September quarter would bring in less revenue than its June quarter. The shares fell 28.1% that day, from $48.76 to $35.08, on 13.3m shares against a normal two to four million. The results underneath that guidance were the best the company had printed since 2023.

The gap between the quarter and the guide is the whole argument on this stock, and it gets its next test on Thursday, when the company webcasts a 2026 investor day at 10:00 EDT at which management has said it will detail safe-harbour transactions in residential and commercial solar and how it thinks about revenue from its solid-state transformer work.

The quarter behind the guide

June-quarter revenue was $346.2m, up 19.6% year on year and 11.5% sequentially — a fourth consecutive sequential gain off the $170.7m trough of December 2024. Reported gross margin reached 27.5%, against 11.1% a year earlier and deeply negative at the trough. Non-GAAP gross margin of 28.6% marked a sixth straight quarter of expansion, and non-GAAP operating income of $10.2m was the first quarterly profit in nearly three years. Free cash flow was $3.1m, against $601.6m of cash.

Two things sit inside that margin. The first is $13.3m of tariff refunds, roughly four points, the same kind of non-demand credit that flatters inverter income statements across the industry. The second is what is absent: management said there was no material safe-harbour pull-forward in the quarter, so the growth reflects equipment installers actually needed.

The September guide of $310-340m assumes about $15m of European seasonality, continued US residential weakness and no repeat of the refunds — though $11.5m arrived in July. That is how a 28.6% margin becomes a 22-26% guide.

Why the US half is stuck

US residential revenue of $154.9m fell 2% sequentially, blamed on slow tax-equity funding and unresolved foreign-entity rules squeezing installer cash flow. The mechanism is federal. The Section 25D credit that paid homeowners 30% of system cost expired outright on December 31, 2025, while the commercial 48E credit survives for leases and power purchase agreements, so federal money now flows only to systems somebody else owns. Wood Mackenzie and the Solar Energy Industries Association see US residential installations falling 21% this year, with third-party-owned share rising to 65% from 44%. The financiers who now buy the hardware cannot close until they know the content rules: Treasury's interim guidance in Notice 2026-15 requires at least 40% of project value from non-prohibited foreign sources, with fuller regulations still pending.

Europe is carrying the company in the meantime: revenue of $154.4m rose 36% sequentially and more than doubled year on year, and the new Nexis three-phase platform shipped more than $60m there in the quarter, with a single-phase version due in the first quarter of 2027.

The mirror

Enphase, whose microinverters convert power at each individual panel, is the opposite case. Its June revenue fell 19.6%, US sell-through fell 34%, and $84.3m of the quarter — 29% — was safe-harbour equipment for projects whose revenue is not recognized until 2028. "We are not stopping and waiting. We are not waiting for things to improve," chief executive Badri Kothandaraman told investors on July 28. The shrinking company still carries the richer multiple: Enphase at 7.71x trailing gross profit and 8.63x forward — the forward reading above the trailing one, meaning gross profit is priced to fall further — against 7.04x trailing and 7.11x forward for SolarEdge. Both stood near 16x in mid-May. The duopoly they once split is also gone: Wood Mackenzie's 2025 review put Enphase at 31.7% of US residential inverters, SolarEdge at 31.3% and Tesla Energy at 29.6%.

What the derating earns

SolarEdge earns part of its fall on the guide itself. A business whose September quarter shrinks sequentially after four gains is not compounding yet, and its margin promise leans on refunds it declines to forecast. The rest it does not earn alone: measured from June 5 to September 4, eight of eight listed solar names fell between 23% and 46%, so most of what happened to this multiple happened to everything wearing the label. The likelier reading is broad compression laid on top of a real, policy-driven hole in US demand. What keeps the recovery case fragile is leverage — $337m of 2.25% notes due 2029, whose conversion conditions were satisfied during the June quarter, leaving them convertible at holders' option through September 30.

Enphase has already put a number on its safe-harbour book: roughly $1.1bn of agreements, most of the revenue arriving in 2028. SolarEdge has said only that it had none in the June quarter. Thursday is the day it says what it has instead.

Two meters, one speed

Okta bills enterprises for every employee who logs in, and bills developers using its Auth0 product for the monthly active users of their apps. In the quarter ended July 31, reported August 26, those two books grew 11% and 13% in annual contract value — and the split between them, 59% workforce and 41% customer identity, was identical to the prior quarter, when the same lines grew 10% and 12%. Both halves accelerated by exactly one point. The mix did not budge.

That is the fact the last six months of share price does not obviously accommodate. Okta, the largest independent identity vendor and the one company whose front door most large enterprises pass through, has been re-rated on the premise that AI agents will multiply the number of identities it charges for, breaking the link between its revenue and its customers' headcount. Chief financial officer Brett Tighe, asked on the August 26 call what agents contribute today: "Still immaterial. Still very small. We're very early innings… for FY '27, we don't think it's going to be material. But '28 and beyond, we do think that there is a real possibility for this to be material for the business in the long run."

What is actually expanding

The expansion Okta can document is module attach at its biggest accounts. Identity Governance, Privileged Access and Identity Threat Protection made up 30% of bookings in the quarter, at an average uplift of roughly 40% to deal value. Customers above $1m of annual contract value grew 22%, past 600. But customers above $100,000 grew only 6%, to 5,255. This is large-enterprise consolidation — existing whales buying more shelf — rather than a proliferation of new identities being metered.

The bookings are genuinely better than the revenue. Current remaining performance obligations rose 14% to $2.585bn and total obligations 17% to $4.858bn, against reported revenue growth of 10.6% to $805m — the third straight quarter of decelerating revenue, from 11.6% two quarters ago. Net retention held at 107%. Free cash flow was $227m, a 28% margin against 22% a year earlier. Gross margin widened almost three points to 79.6%. The seven-point jump in reported operating margin, to 13.3%, is mostly stock compensation and acquisition amortization rolling off: on a non-GAAP basis, operating margin improved about half a point, to 28.2%. And Okta's own third-quarter guidance takes that 14% bookings growth back to 11–12%.

The re-rating was three days

The six-month gain of 111% was not a grind. Three earnings sessions — May 29, June 1 and August 27 — supply roughly nine-tenths of it; the August 27 session alone ran $129.35 to $172.91, a 33.7% close-to-close move, after CNBC reported a 20% gain in extended trading the evening before. The 50-day average has sat above the 200-day since June 8.

What the shares now pay for that is roughly double what they paid in spring. Price to trailing gross profit is 11.80x, against 5.97x on May 3 and 10.29x on July 29, on gross profit that grew 14.5%; forward earnings are 43.4x against consensus revenue growth of 10.9% this year and 9.9% next. The mean analyst target, near $172, is level with the price. Okta led a broad security re-rating rather than standing alone — Palo Alto Networks rose 101.9% and Fortinet 86.8% over the same window, while CrowdStrike fell 50.3%.

SailPoint, the identity-governance specialist and the only live independent comparable since Palo Alto closed its $21.1bn purchase of CyberArk in February, grew revenue 21.6% to $280.1m last quarter — about twice Okta's rate — but at a 64.7% gross margin and a 28.5% operating loss, and at 14.96x trailing gross profit. It rose 14.8% on August 27 having disclosed nothing of its own.

The verdict

The business earns part of this. Bookings running three and a half points ahead of revenue, retention off its trough, and cash conversion at 28% are real, and the governance and privilege modules are being bought. What nothing in the disclosure yet earns is the doubling of the multiple, because the specific thing that doubling assumes — a usage-metered book pulling away from a seat-metered one — has not started. Okta's two halves are growing at the same pace, in the same proportion, and its own guidance says the bookings figure steps back next quarter.

Okta's cross-app access protocol went generally available with Anthropic, and agent single sign-on now ships inside the standard edition with a consumption cap built in but not enforced. Chief executive Todd McKinnon said on the call that agent identity "could be the biggest category of cyber" in the fullness of time. The pricing today is still a per-user uplift — the plumbing for a per-agent meter is installed, and the meter is not yet switched on.

Grand Canyon Education signed a new fifteen-year contract in July with the single university that supplies almost all of its revenue, restructuring its fee to 60% of tuition and academic-related charges. The agreement took effect on July 1 — the same day the federal government stopped issuing Grad PLUS loans to new graduate borrowers.

That coincidence is the story of this corner of the market. Neither listed company here is a university in the way its label suggests, and both are meters on federal student aid rather than on classroom demand. What the meters register, and what can turn them off, has almost nothing to do with the enrollment headlines that have been moving the share prices.

A fee, not a school

Grand Canyon Education stopped owning a school in 2018. It now sells learning-management technology, curriculum design, admissions, marketing and back-office work to universities, and through its Orbis Education arm supports healthcare programs at 27 of them. But 89.4% of its service revenue in the first half of 2026 came from Grand Canyon University alone, the same share as a year earlier.

The amended master services agreement runs through June 2041 with up to three automatic five-year renewals and removes the university's right to terminate for convenience. It also narrows the fee base: 60% of tuition and academic fees, with ancillary revenue staying with the university. Management put the revenue cost at roughly $20m a year and the operating-income cost at under $1m a quarter, because a reimbursement the company had been paying the university disappears alongside it.

The operating base underneath is not deteriorating. Second-quarter service revenue was $264.0m, up 6.7%, with operating margin of 22.0% against 20.9% a year earlier. Operating income has outgrown revenue in each of the last four quarters, and the diluted share count is down 6.8% year on year. Hybrid campus enrollment grew 8.5% across 47 locations. "The revenue per student of these students is more than three times that of an online student," chairman and chief executive Brian Mueller told investors on July 30 — though 22 of those sites will not add new students this autumn because of state regulatory caps.

The rule that actually bites

The legal basis for the whole arrangement is the Department of Education's 2011 bundled-services exception, which permits paying a third party a share of tuition per enrolled student when recruitment is bundled with other services; without it, federal law bars compensation tied to securing enrollment. The department has been reviewing that exception, and separately proposed a rule holding programs accountable for graduates' earnings. More immediate is the end of Grad PLUS for new borrowers, with lifetime federal graduate borrowing capped at $100,000 — a hard ceiling on what the university's large master's and doctoral base can finance, and therefore on the 60% of it that gets booked as service revenue.

American Public Education runs the mirror image: it owns American Public University System, Rasmussen University and Hondros College of Nursing outright. Its military school's 90/10 ratio was 89% in both 2024 and 2025 — one point of headroom under the federal cap, after tuition assistance and GI Bill money were moved to the federal side of the calculation. The business itself is repairing: second-quarter operating income rose 92.5% to $13.5m, the nursing segment grew revenue 11% on 7% enrollment growth, and on August 10 the company raised full-year guidance to $690–698m of revenue and $2.48–2.79 in earnings per share.

What the group was actually selling

The July and August declines came from elsewhere. Lincoln Educational reported new student starts up just 1% against 19.5% a quarter earlier, and on August 6 Universal Technical Institute cut its full-year adjusted earnings guide to above $135m from above $155m, blaming roughly 70% of it on its high-school recruiting channel. Over the last thirty days those two fell 40.6% and 46.9%. Grand Canyon Education rose 1.9%; Perdoceo and Strategic Education, the other obvious read-across names, barely moved.

So the verdict splits. Grand Canyon Education's twelve-month fall from $205.14 is multiple compression and nothing else — trailing earnings of $8.27 a share are within a cent of a year ago, and the stock trades at 18.3x trailing and 14.9x forward against roughly 25x last September. Part of that is earned: a $35.0m litigation reserve blew up the third quarter of 2025, and the shares gapped 8.9% lower on October 31. But the Education Department rescinded its $37.7m proposed fine with prejudice in May 2025, and the fee contract is now locked for fifteen years. American Public Education's thirty-day drop is harder to defend at all: it came after a beat and a raise, and nets against $133.9m of cash above debt, about 16% of the market value, leaving under 7x guided adjusted earnings on an enterprise basis.

What neither price reflects, in either direction, is the federal plumbing. A graduate borrowing cap that took effect in July shows up in autumn applications, not in a June income statement; a 90/10 ratio at 89% is a compliance line, not a demand signal. Both businesses are running well on rules that changed nine weeks ago.

Ambarella closed its July quarter with the fifth consecutive slowdown in its rate of revenue growth, and every product it announced alongside the result bills later. Revenue was $108.1m for the quarter ended July 31, up 13.2% year on year, the company reported on September 3; the guidance midpoint for the current quarter, $119.5m, implies about 10.2% growth, a sixth step down. Adjusted earnings of $0.18 a share beat estimates.

That matters beyond one print because Ambarella and CEVA are the two listed companies whose whole business is moving AI inference into devices, and they bill on opposite meters. Ambarella designs the low-power vision chips inside internet-protocol security cameras, dashcams and automotive driver-monitoring systems, and is paid when silicon ships. CEVA never ships a chip: it licenses digital signal processor, neural processor, Bluetooth and Wi-Fi blocks to chipmakers, taking a fee upfront and a per-unit royalty on every device a licensee manufactures, reported a quarter in arrears. Both meters have now been read, and neither shows edge inference arriving as revenue yet.

Ambarella: the ramp is real, the new money is FY2028

The operating leverage is genuine. Gross profit rose 13.4% on revenue up 13.2% with operating expenses held, narrowing the operating loss to $8.1m from $22.0m and taking operating margin from -23.0% to -7.5%. Automotive set a revenue record on commercial vehicles, and the 5nm CV75 and CV72 chips are in a steep ramp.

The announcements, though, are dated. The X7 — Ambarella's first standalone AI accelerator, entering a field management itself described as including NVIDIA, Qualcomm and roughly 50 startups — is only sampling. The semi-custom 2nm CV8 expects first production revenue in fiscal 2028. Two seven-year channel partnerships were sized by chief executive Fermi Wang on the September 3 call: "When we talk to both CapGemini and Macnica, the range of revenue we are expecting from this collaboration is $0.5 billion with each one of them" — with meaningful revenue two to three years out.

What is not dated is cost. Non-GAAP gross margin of 59.3% missed the company's own guided midpoint, and management flagged memory-supply risk for the fourth quarter. Memory makers have shifted wafer capacity from conventional DRAM to high-bandwidth memory for AI accelerators; data centers now absorb an estimated 70% of world memory output, with contract prices up more than 50% quarter on quarter entering 2026. The AI buildout is currently reaching Ambarella as an input cost.

CEVA: sixteen percent more chips, one percent more money

CEVA's licensees shipped 567m units in the June quarter, up 16%, and paid $10.8m of royalty — up 1%. That is roughly 1.90 cents a device against about 2.19 cents a year earlier, because the growth came from low-rate Wi-Fi, up 28%, while higher-earning Bluetooth units fell 16%.

The reported line is otherwise improving. Licensing revenue rose 21% to $18.2m, the best in three years, on 10 agreements including two with first-time customers; full-year growth guidance went up to 13-15% and non-GAAP operating margin to 11% from 3%. "We are seeing increasing demand for our technologies across AI, connectivity, and sensing, strong adoption of broader hardware and software platforms, and continued diversification of our royalty base," chief executive Amir Panush said on August 10. But the flagship win — its NeuPro-M neural processor selected for custom AI silicon by a large computing platform company — runs 18 to 24 months to production, so royalties start in 2028. The identified fix for the rate per device, Bluetooth HDT, ramps meaningfully in 2027-28.

What the shares have done

Ambarella closed at $62.89 on September 4, down 27.5% from its August 7 high and 15.1% below the $74.09 it closed at before the Financial Times reported NXP Semiconductors was in talks to buy it for more than $3bn — the entire takeover premium and more, given back. Its 10.7% fall over September 1-4 was its own: NXP rose 1.8% and Arm 5.5% across the same sessions. CEVA, at $26.95, is down 45.6% since early June, but so is the small-cap chip complex around it — Himax -42.8%, onsemi -43.6%, Arm -35.9%.

Gross margins differ by nearly 30 points, so price against forward gross profit is the only comparison that holds: Ambarella at 10.7x, down from roughly 14.8x at its August peak, and CEVA at 6.9x from about 12.8x in June. Ambarella carries 78.2x forward earnings on consensus growth of 13%; CEVA 49.2x, falling to about 34x on 2027 estimates.

Ambarella's de-rating is earned. Five quarters of deceleration and a guide to a sixth, on a chip that has to be manufactured through a memory squeeze, do not support a multiple set for acceleration. CEVA's is not earned by its reported numbers, which are going the other way — but the royalty rate per device is real deterioration, and the company's own remedies for it are 2027 and 2028 events. Two businesses, two meters, one answer: what edge AI is worth to either of them will not be measurable this fiscal year.

The cleanest test costs nothing to run. CEVA reports its licensees' unit shipments and its royalty dollars every quarter. When 16% more devices start producing more than 1% more money, the story will have arrived.

Grupo Aeroportuario del Sureste, which holds the concessions for nine airports in southeastern Mexico including Cancún, plus San Juan in Puerto Rico and a Colombian cluster around Medellín, closed its acquisition of Motiva's airport interests on September 1. The price was R$5.1bn, about $936m, for 20 airports — 17 in Brazil plus Quito, San José and Curaçao — in what the company calls the largest international expansion in its history.

It bought them at the low point of its own operating cycle. That is the stake: a company whose franchise asset is contracting has just added a portfolio carrying roughly 45 million passengers a year, and none of those airports appear in the earnings the market is currently pricing.

Cancún is losing seats, not visitors

Cancún International handled 2.11 million passengers in June, down 11.5% year over year, with international traffic down 13.1%; first-half volume of 14.76 million was 4.7% below a year earlier. The cause is upstream. US carriers removed more than a million seats from Mexico–US routes for the summer 2026 season, with Alaska Airlines cutting about 240,000 over the past year, American more than 170,000, and Spirit Airlines' bankruptcy taking roughly 260,000 transborder seats out altogether.

"The summer is lost, and we are expecting the recuperation process up to the end of the summer season," chief executive Adolfo Castro told investors on the July 24 call. "Winter season, we see a better outlook, and I would say more seats, more offered seats than what we had last year."

The margin problem is structural

Second-quarter adjusted EBITDA margin fell 560 basis points to 62%, with Mexican EBITDA down 9% and Puerto Rico down 17%. Administrative expenses rose about 30% and medical insurance costs 39% on Mexican tax reform, both described as recurring. Sitting underneath all of it is the federal concession fee on gross revenue, raised from 5% to 9% in January 2024 — a permanent subtraction no traffic recovery reverses. Operating income has fallen year over year for four consecutive quarters.

The commercial annuity is thinning too. Group commercial revenue per passenger rose 13% to MXN 153, but only because a newly consolidated US concessions business joined the numerator; Mexico's figure fell to Ps.145.7. Castro was blunt about what was added: "Today's EBITDA margin in the U.S. operations is around 9%… It is a completely different business in comparison with what we have in Mexico, Puerto Rico, and Colombia."

What the discount measures

ASUR trades at 8.02x trailing EV/EBITDA against OMA at 9.20x and GAP at 10.25x, with the shares at 257.33 against a twelve-month range of 255.23 to 381.16. JPMorgan cut the stock to Underweight on weak traffic, trimming its target to MXN 615 while acknowledging the 14% discount to peers. Cash generation is not the complaint: first-half operating cash flow of MXN 7.3bn was up 21%, net debt is 0.9x EBITDA, and two extraordinary dividends of MXN 10 a share are proposed for November 24 and December 15.

Grupo Aeroportuario del Pacífico, which runs twelve Mexican airports including Guadalajara and Tijuana, is the same regulator and the opposite result. "Passengers traffic declined by 5.6%… EBITDA grew by 8.4%, and EBITDA margin expanded by 230 basis points to 69.3%," chief executive Raúl Revuelta said of the second quarter, with non-aeronautical revenue up 23.9% on parking, advertising and the Cross Border Xpress bridge it operates itself. August traffic rose 0.5%, with Guadalajara up 10.5%. Its shares still fell 15.7% over twelve months, and its trailing free-cash-flow yield is 2.01% against OMA's 6.00% — the arithmetic of a MX$52bn 2025–2029 build program, close to MX$19bn of it at Guadalajara, that consumes cash years before the regulated tariff recovers it.

So the two names fell together for unrelated reasons, and the common explanations do not survive. It is not the Mexican market: the iShares MSCI Mexico ETF rose 21.5% over the same twelve months in which ASUR fell 23.3%. It is not translation either — the peso closed at 16.87 per dollar on September 4, up roughly a fifth since January 2025, which means each peso of earnings converts into more dollars and ASUR's decline understates the local de-rating. ASUR's discount is earned by a real margin break at Cancún and San Juan. GAP's is a cash-flow timing charge on a build it has committed to.

What is unexplained is the denominator. The multiple the market is marking ASUR down on reflects a company of nine Mexican airports, one Puerto Rican and a Colombian cluster. Since September 1 it has been something else, and management has told investors to expect "business as usual" from the new airports — no significant synergies, no partial divestment. Cancún's Terminal 1 opens in the fourth quarter. The next monthly traffic release will say whether Castro's lost summer ended when he said it would.

Paycom's operating profit grew 50% in the June quarter on revenue up 9.8%, and the extra dollars did not come from more workers on the platform it bills. Paycom Software sells single-database payroll, time, benefits and talent software to small and mid-sized American employers, and it is paid largely per employee per month — a meter that only moves when clients hire. Operating income reached $168.5m on revenue of $531.2m, lifting operating margin to 31.7% from 23.2% a year earlier.

The margin came from inside the company. Gross margin rose to 83.2% from 81.9%; management told the August 5 call that the 2025 build-out of its own data centers now yields more than $100m of annual research savings plus over $30m less in third-party artificial-intelligence token fees. Below that, the diluted share count fell 18.5% to 45.9m after $1.4bn of repurchases year to date. That is why consensus has Paycom's earnings per share up 30.9% this year to $12.09 while revenue grows 7.7% — and the company guided 2026 revenue to $2.197–2.212bn, or 7–8% growth, slower than the quarter just reported.

The meter underneath

Paycom stored 7.4 million employee records at the end of 2025, up 5%, across roughly 20,300 parent-company clients. Chief executive Chad Richison told investors on the August 5 call that "client employment growth would have just been stable, same, consistent with -- as it's been every year in the past with the exception of when it went down about 14% during COVID." Stable is the operative word: Paycom's automation products deliberately remove the service work it once billed for, and the same 10-K that lists ADP, Paychex, Paylocity, Dayforce, Oracle, SAP, UKG and Workday as competitors rests the defense on that automation rather than on price.

The other high-margin dollar is float. Paycom holds client tax withholdings and payroll cash for days and invests it, earning $26.0m of interest in the quarter on an average daily balance near $2.9bn, with about $105m guided for the year. That is 15.4% of quarterly operating income. A quarter-point of easing costs roughly $7m annualized — under 1% of consensus operating income, real but second-order. ADP, the industry's largest processor, shows why the direction is not automatic: it guides fiscal 2027 interest on client funds up to $1.54–1.56bn on a portfolio yielding about 3.7%, because a laddered book keeps rolling old low-coupon paper higher even as the front end falls, while guiding the workers it actually bills to 0–1% growth.

Workday's two backlogs disagree

Workday, which leads enterprise human-capital software buyer consideration at 41.9% against SAP's SuccessFactors at 32.3%, reported subscription revenue up 13.9% and twelve-month subscription backlog of $9.03bn, up 14.2%. Its total subscription backlog, the multi-year commitment, grew 8.0% against 17.6% a year earlier, and fiscal 2028 subscription growth is pre-guided to about 11% with margin doing the work. Chief executive Aneel Bhusri said on the August 27 call that "AI products alone drove more than 100 million of new ACV, which accounted for more than 25% of all new ACV closed in the quarter" — new annual contract value — though only 200 customers had signed the flex-credit contracts that meter agent work, against 5,500 already running agents. Headcount stayed flat at 20,900 and a $5bn buyback finished six months early.

What was paid for it

Since the April 10 low Paycom has doubled and Workday risen 74%; over twelve months the group is still down about 18%, with Workday off 15%. Paycom's August 6 gap of 23.8%, the session after its print, did more than the rest of the last 30 sessions combined. Workday's quarter compresses into a late-July rotation, the report that Silver Lake was in take-private talks that moved the shares 12.5% on August 14, and a 5.8% post-results session; the drift between was negative.

So the businesses earn part of this. Paycom's 8.5 points of margin expansion are cash, and its cost program is documented. Workday's shorter meter is steady. What neither earns is the price change: Paycom at 19.2x forward earnings against 24.6x trailing, versus roughly 9.4x implied in April on the same $12.09 estimate, and Workday at 17.7x forward against 39.6x trailing — against ADP's 23.1x forward, the mature comparison. Workday now costs 6.66 times its trailing gross profit, against 4.31 times on May 3, while that gross profit grew 13.2%.

Both companies are being paid a growth price for doing more with the same number of employees underneath them. Cost programs lap, share counts can only shrink once, and the August employment report counted 22,000 new jobs. The next quarter has to come from somewhere the meter can see.

Vail Resorts sells next winter before it snows. The company that runs 37 mountain destinations under the Epic Pass — Vail, Breckenridge, Park City, Whistler Blackcomb — collects most of a season's lift revenue in a spring-and-autumn selling window, banks it as deferred revenue, then recognizes it whether or not the snow arrives. For fiscal 2025 that meant roughly 2.3 million guests committed in advance across its North American, Australian and European resorts, over $975m of revenue and about 75% of all paid skier visits.

The mechanism was stress-tested this year and passed. What failed is the meter that tells you about next year: pass product units for the 2026/27 North American season were down approximately 10% through 26 May 2026, days sold down 8%, and sales dollars down about 5%. Price is covering roughly half the unit loss, and the committed dollar base is now shrinking too.

The winter did what the model said it would

"Weather conditions remained extremely unfavorable in the third quarter, adding to what had already been one of the most challenging winters in history across the western U.S.," chief executive Rob Katz said in the June 8 results release. Skier visits in the April quarter fell to 7.276 million from 8.609 million, a drop of 15.5%, while effective ticket price rose about 12% — pass holders had already paid, so lift revenue barely flinched. The pain migrated to everything bought on the mountain: ski school revenue fell 11.5% to $141.8m, retail and rental 8.3%, dining 10.7%. Resort Reported EBITDA for the quarter fell $61.3m.

The leak is in the conversion funnel. Fewer visits this winter means a smaller pool of skiers to sell next winter's pass to, and new-passholder sales came in weaker than renewals. A bad snow year does not hurt the season it ruins; it hurts the one after.

Price has been doing the work, and it is running out

Unit declines were 3% into 2024/25, 1% in the spring 2025/26 window and 3% again through September 2025 before this year's 10%. Short-term deferred revenue was $467.0m at 30 April 2026 against $468.6m a year earlier — flat.

Management's answer is cost and product. The Resource Efficiency Transformation Plan delivers an incremental $45m this year, $106m annualized, which does not fill the roughly $100m hole the guidance cuts opened. Katz's July 14 "Epic Experience" reset put money into food, private lessons, gear and snowmaking, framed as a break with the past: "the Pass and acquisitions were not the end goal," he said. The 2026/27 Epic Pass starts at $1,089, up about 3.6%, with a new 20% discount taking under-30 buyers to $869 — a price cut aimed at the demographic being lost.

What the shares have already conceded

The stock is down 15% over twelve months, and fell 11.7% in the ten sessions to September 4 without a discoverable company announcement; the likelier reading is de-risking into the fiscal-year print, though Goldman Sachs initiated at Sell with a $132 target on August 13, arguing the company's 5–7% organic EBITDA growth ambition sits against roughly 1.5% actually compounded from 2019 to 2025.

At 10.4x trailing enterprise value to EBITDA, against a thirteen-year median near 15.9x and a low near 9.0x, the multiple has done its de-rating. The earnings have not finished doing theirs: consensus has fiscal 2028 earnings per share at $7.15, still short of the $7.53 Vail earned in fiscal 2025. Meanwhile the $8.88 annual dividend exceeds the roughly $6.66 per share of trailing free cash flow, net debt stood at 3.2 times reported EBITDA at the last fiscal year-end, and fiscal 2025 buybacks retired 4.5% of the shares at around $156.

So the verdict splits cleanly. Weather earns the collapse in visits, in ski school and in this year's guidance — that part is cyclical and a strong El Niño, which forecasters put at better than 90% odds for the coming winter, would reverse much of it. What weather does not explain is a unit slide that has widened every year for four years while the price went up anyway. Snow fixes ancillary spend per visit. It does not, by itself, sell a pass.

On September 28 Vail reports fiscal 2026 and, with it, the autumn pass update — the first count taken after the September 7 price increase, and the only number that says whether the base is stabilizing or still going.