DK Street Journal

NextEra Energy, which owns Florida Power & Light and the largest contracted clean-energy development arm in the United States, knows precisely what it may earn on money invested in Florida through the end of 2029. Florida regulators approved a four-year settlement setting an authorized return on equity of 10.95% on a 59.6% equity layer, with base rates rising $945m from January 2026 and a further $705m a year later. What the company pays for that equity is not settled, and in September it went up.

The distance between those two numbers is where a regulated utility's profit actually lives. Entergy, NextEra and Dominion Energy all run negative trailing free cash flow — Dominion the deepest at -12.4% of market value, Entergy at -6.7%, NextEra at -6.2% — which means each spends more on plant than operations generate and issues securities to cover the difference. When the Federal Reserve raised rates on September 16 and the 10-year Treasury yield returned to 5%, its highest since 2007, the cost of the next dollar of that funding moved while allowed returns near 9.8% to 11% stayed exactly where commissions put them.

What actually moved

Between September 8 and September 21 every large regulated power name fell — Entergy 6.5%, PSEG 6.7%, Dominion 5.1%, NextEra 5.0%, AEP 4.3%, Southern 3.5%, Duke 3.4% — while the S&P 500 exchange-traded fund rose 0.9%. The thirty-day group number is worse, but half of it is one broken company: PG&E dropped 23% across two sessions at the end of August after California's legislature passed a wildfire bill stripped of the liability protections utilities expected. That is a Sacramento event, not a cost-of-capital one. Utilities are now roughly flat for the year and the second-worst of the index's eleven industries.

Entergy: the load is real, the shares are new

Entergy sells power to about 3 million customers in Arkansas, Louisiana, Mississippi and Texas, and it is building into the biggest industrial order book in its history. "We continue to have seven to 12 gigawatts of hyperscale data center potential in our pipeline, as well as three to five gigawatts of interest from traditional industrial segments," chair and chief executive Drew Marsh told investors on July 29. Louisiana's rate base is expected to roughly double to near $50bn by 2030 inside a $67bn plan.

The terms are better than the skeptics assume. Entergy's "Fair Share Plus" framework carries prepayments, parent-company guarantees and early-termination penalties, and in Arkansas Alphabet agreed to pay $526m toward a solar-and-storage project plus $190m for transmission. The problem is arithmetic between the plant and the shareholder. June-quarter revenue rose 5.9% to $3.524bn and net income 3.4%, but operating margin fell to 23.7% from 25.2% and diluted earnings per share slipped to $1.03, because the share count rose 4.6%. On September 2 the company settled 11.1m more shares for about $913m while reaffirming guidance of $4.25 to $4.45. Entergy trades at 25.6x trailing and 23.0x forward earnings, against 29.9x trailing in mid-May — and at 2.52x book value against a Louisiana return of about 9.7% once affiliate preferred investments are stripped out.

NextEra: nothing in the demand book broke

"NextEra Energy delivered a strong second quarter, with adjusted earnings per share increasing by 9.5% year-over-year," chairman and chief executive John Ketchum said of results reported July 24. Revenue grew 12.4%, operating income 17.1% — faster than sales — and the share count rose 1.3%. Yet the stock has fallen 12.9% in six months and trades at 17.7x trailing earnings below its own 19.7x forward, which says trailing accounting profit sits above the adjusted run-rate, not that the shares are cheap. Consensus still models 9% adjusted earnings growth this year and next.

Dominion, which NextEra agreed in May to buy entirely in stock at 0.8138 shares apiece, is now an input to that judgment rather than a separate one: both shareholder bases approved the $66.8bn combination on September 3, and on September 21 the two closed at a ratio of 0.7907, a 2.8% discount for deal risk. What NextEra inherits is a Virginia franchise with about 12 gigawatts under executed service agreements out of 53 in some stage of contracting, a new class for customers above 25 megawatts that bills a minimum on 85% of contracted delivery demand from January 2027 — and an offshore wind project whose cost has climbed to $11.5bn from $9.8bn at 81% complete, earning a state-allowed return of only 9.8%.

The split

September hit these two the same way and they did not deserve the same verdict. Entergy's compression is rational work on an expensive starting point: the hyperscaler load is contracted and the risk terms are sound, but the equity funding it is being issued today against rates a commission will approve later, and per-share earnings have already gone backwards once. NextEra's is financing cost and nothing else — a locked allowed return, a growing signed backlog, positive operating leverage and almost no dilution. The honest bear case is that a higher discount rate is permanent; it is not that the order book softened.

The largest single block of contracted megawatts in the group still has no approved rate behind it. Louisiana regulators take up Entergy's seven gas plants for Meta's Hyperion campus on December 16.

The discount that never reaches the sticker price

KB Home reports fiscal third-quarter results today, and the line that decides the quarter is not deliveries. The Los Angeles builder, which sells single-family homes and townhomes to first-time, move-up and active-adult buyers across the West Coast, Southwest, Central and Southeast, guided in June to a housing gross margin of 16.0–16.6% on 2,600–2,800 deliveries. Consensus earnings per share is $0.90.

Margin is where the discounting lives. The industry's preferred concession is a builder-paid mortgage-rate buydown — a forward commitment bought from a mortgage bank so the buyer gets a below-market rate — and it lands in cost of sales rather than in the list price. KB Home told analysts on June 23 that roughly 60% of its orders carry some form of mortgage concession, flat against the prior two quarters. The cost of that concession rises with the rate being bought down, and the 30-year benchmark reached a one-year high of 7.05% in the week of September 14–18, with the 10-year Treasury above 5%. The Federal Reserve raised its policy rate a quarter point to 3.75–4% on September 16, its first increase since 2023. Builder sentiment fell to 32, a 12-month low, and 66% of builders used incentives in September, the highest share since December.

Three builders, three meters

KB Home sells built-to-order: 73% of net orders in the quarter ended May 31 were houses configured by a buyer, so booked revenue reflects contracts struck months earlier. That quarter's revenue fell 27.3% to $1.112bn and gross margin came in at 15.6% against 19.5% a year earlier — the fourth consecutive quarterly decline from 20.3%. Average selling price for the half fell 8% to $457,000. The order book is thinning as the store count grows: ending community count rose 11% to 280 while backlog value fell 7% to $2.14bn on 4,526 homes. Cancellations improved, to 12% of gross orders from 16%. "We produced solid second-quarter results that met or exceeded the mid-point of our key guidance ranges," executive chairman Jeffrey Mezger said on June 23.

Meritage Homes, which builds mostly move-in-ready houses for entry-level buyers across Texas, Arizona, Florida and the Southeast, gives its concessions in the quarter it closes the sale. Its home closing gross margin fell to 18.3% in the June quarter from 21.1%, with orders down 9% as a 19% slower absorption pace outran a 14% larger community count. Cancellations rose to 13% from 10% — the only worsening of the three. But its unsold finished houses are shrinking, not piling up: completed specs fell to 30% of total spec count from 50% at the end of 2025. "Our second quarter results reflect solid execution in a softer demand environment," chief executive Phillippe Lord told investors on the July 30 call, adding that the company "saw no meaningful deterioration in demand from the first quarter to the second quarter." It cut 2026 guidance to roughly 5% below 2025 at the same time — before, not during, the recent slide.

Dream Finders Homes, a Jacksonville builder that controls most of its lots through option deposits with land bankers instead of owning dirt, is the test of whether that structure helps. Orders rose 15% to a second-quarter record 2,232 and cancellations fell to 11.1% from 14.0%, yet homebuilding gross margin dropped to 14.2% from 16.5%, which the company attributed primarily to higher land and financing costs. The optioned pipeline is being released rather than extended — 54,091 controlled lots at June 30 against 63,121 at year-end — and $3.0m of lot-deposit and deal-abandonment charges appeared in selling and administrative expense.

What the shares did, and what they are priced at

Over the thirty days to September 21, KB Home fell 13.6%, Meritage 10.2% and Dream Finders 23.6%, alongside D.R. Horton at -6.1%, Lennar -10.5% and PulteGroup -7.9%. No company news dated inside that window explains it; the repricing is the rate move.

All three now sit below book value — Meritage at 0.85 times, KB Home 0.78, Dream Finders 0.66 — and each bought back stock last quarter, $100m, $75m and $15m respectively. On earnings the picture inverts. KB Home's forward price-to-earnings of 14.5x sits above its trailing 11.4x, because fiscal 2026 consensus has already been cut 48% to $3.31 a share. Meritage's forward multiple of 13.0x is barely below its trailing 13.4x on consensus of $4.94, down 26%. Dream Finders looks cheapest at 7.2x trailing, but it consumes cash — free cash flow yield of minus 15.4%, against plus 12.4% at KB Home — and analysts pencil in only $1.19 for 2027 after $1.13 this year.

The verdict

The business earns most of this de-rating. Revenue is falling at all three, reported margins are compressing at all three, and estimates have been cut by a quarter to a half — this is not multiple compression against intact earnings. What it is not is a demand collapse: cancellations improved at two of the three, and Dream Finders sold more houses than in any prior second quarter. The squeeze is arriving through the concession. Because the discount is a bought-down mortgage rate, the price line stays respectable and the margin line absorbs it, which means a 7% mortgage does not show up as falling average selling prices — it shows up as the number KB Home publishes today.

And Dream Finders has chosen this moment to get much bigger: it agreed in August to buy Beazer Homes for $33.50 a share in cash, an equity cheque of about $916m from a company worth roughly $1.05bn, on committed financing from Goldman Sachs, Bank of America and Kennedy Lewis, with closing expected in the fourth quarter. It has promised to keep a "100% land-light strategy" afterwards while buying a builder that owns its dirt.

FMC agreed in June to sell roughly a fifth of itself at $13.30 a share to Tessenderlo Group, a Belgian industrial company. The stock purchase agreement covers 30,319,166 new shares for $403.2m, and the money is not for growth: it goes to lenders. Tessenderlo's price sits 29% above FMC's September 21 close of $10.28.

That gap is the story. FMC ended the June quarter with $4.1bn of net debt, equal to 5.2 times earnings before interest, taxes, depreciation and amortization. Room under its loan terms exists because the terms were rewritten: an April 16 credit amendment leaves the maximum leverage ratio untested for the first three quarters of 2026, sets it at 6.75x through the end of 2027, and pledges collateral. S&P Global Ratings has since cut FMC to 'BB+', below investment grade.

One molecule family, repriced

FMC sells branded insecticides, herbicides and fungicides through distributors and alliance partners, and its earnings base has for a decade rested on the diamide actives — chlorantraniliprole, sold as Rynaxypyr, and cyantraniliprole. Chlorantraniliprole's Chinese compound patent expired on August 13, 2022, after which Chinese producers scaled capacity on cost advantage and turned a premium active into a fragmented multi-supplier commodity. FMC has guided 2026 price down mid-single digits, mainly on Rynaxypyr.

The second quarter shows where the damage lands. Revenue fell 17.5% to $867.1m, volumes fell 10% on lower diamide partner orders and weak legacy-product demand, and North American sales dropped 22.4%. But sales of FMC's own differentiated post-patent diamide formulations grew more than 35% year on year. The commodity active-ingredient socket is being taken; the branded formulation business is growing. Generic economics are not a one-way street either — technical-grade chlorantraniliprole rose to $40.50 a kilogram by late March from $26.98 in January as Chinese regulators pushed compliant intermediate production.

The control that isn't falling

Corteva, FMC's closest branded competitor and now also its licensee, reported crop-protection price down 4% and volume down 2% in the June quarter, blaming Latin American competition — and still grew first-half segment operating EBITDA 9%, to $776m. The price war is industry-wide; the 10% volume drop and the 23% decline in full-year adjusted EBITDA that FMC guides to are FMC's own. Corteva's shares are up 14.1% over twelve months and its separation into two public companies targets October 1, which will hand the market a pure-play crop-protection comparable within days.

Underneath the loss line, cash turned first. Free cash flow reached $357m in the quarter, an improvement of $318m, and full-year free-cash-flow guidance was raised to $75–225m. Roughly $1bn has been assembled for debt paydown from four actions: the Tessenderlo placement, a $252m sale of the India commercial business, a $114m Newark sale-leaseback, and a $200m upfront payment from Corteva under a rimisoxafen supply and license agreement — a herbicide FMC still owns outright, not a diamide. "During the quarter, we completed several important actions that strengthened FMC's financial foundation," chairman and chief executive Pierre Brondeau said on July 29. A manufacturing restructuring approved in December targets $175m or more of annual savings by the end of 2027, at a cost of $560–635m in charges.

What the shares have done

The stock is down 71% over twelve months and 83% over two years, against a 52-week range of $9.71 to $36.18. It is not a smooth grind: it rose 13.9% on the July 30 earnings session, 13.4% on September 2 with no discoverable company news, and made a fresh 52-week low on September 18.

At 7.7 times consensus 2026 earnings of $1.33 a share and 0.79 times book, the equity screens cheap. On the whole enterprise it does not: $1.29bn of market value plus $4.1bn of net debt is about 8.3 times the midpoint of FMC's own $620–680m EBITDA guide, and the share count has 24% of dilution still to come. The business has stopped burning cash and is growing the part of the diamide franchise it can defend; the equity is a thin slice on top of a balance sheet that has to shrink before any of that reaches shareholders.

Tessenderlo's regulatory clearances were expected by the end of September or October, its finance chief said. Until that $403.2m lands, the cheapest thing about FMC is the debt the buyer is paying off.

Two identity vendors sell into the same AI-agent story and only one of them has put an invoice on it. SailPoint, which charges by the identity it governs, reported AI-driven annual recurring revenue above $70m in the July quarter — more than 30% of net new ARR — with total ARR up 25% to $1.231bn and dollar-based net revenue retention of 113%. Okta, which charges per workforce user per month, folded agent sign-on into its existing seat product as a price uplift; finance chief Brett Tighe called the revenue "still immaterial" for this fiscal year.

That difference is now the whole argument in identity software, and last week it got priced. Okta closed at a five-year high on September 21, carried there by target raises that lifted the multiple rather than the estimates: Needham went to $230 from $200 ahead of Okta's Oktane conference and BTIG to $219 from $187, while Bank of America's $200 came explicitly from raising its calendar-2027 enterprise-value-to-sales multiple from nine times to eleven. The average Street target of $186.73 now sits below the share price.

Two meters, and agents fit neither

Okta, the largest independent seller of workforce single sign-on and multi-factor authentication, runs two billing units: seats for the workforce product and monthly active end-users for Auth0, its developer-facing customer identity business. Agent single sign-on was folded into the standard edition with a consumption cap built but not switched on — scaffolding for a meter that does not yet run. Okta prices the service as an uplift on seats it already sells, on mostly one-year agreements, billing per user rather than per agent.

The scale problem that creates is one Okta itself describes. An enterprise it was courting had 50 Claude agents running when the evaluation began. "And then we came back a few weeks and there was 1,500 Claude agents in the environment," chief executive Todd McKinnon told investors on the August 26 earnings call. Neither number reaches an invoice. Brett Tighe, Okta's chief financial officer, was direct on the same call: "Still immaterial. Still very small. We're very early innings… for FY '27, we don't think it's going to be material."

What Okta discloses instead is momentum without a unit: dozens of agent-related deals, 26 software vendors supporting its Cross-App Access standard, and newer products — governance, privileged access, threat protection — at 30% of second-quarter bookings. The core meters slowed. Revenue of $805m grew 10.6%, a fourth consecutive quarterly deceleration, and net retention was 107%, one point better than a year earlier. Bookings ran ahead of that — current remaining performance obligations rose 14% — but management guided the third quarter to 11-12%.

SailPoint sells the other half of identity: governance, deciding which entitlements each human, contractor and machine account may hold, priced against identities governed. An agent is an identity whether or not it is an employee. Its Entro acquisition took discovery coverage past 1,200 non-human identity types. ARR reached $1.231bn on 25% growth, software-as-a-service ARR grew 36%, backlog accelerated to $1.9bn, and the AI-driven line passed $70m against a $100m target for this year and $800m by fiscal 2029. Co-founder and chief executive Mark McClain named the forcing function on the September 9 call: "The EU AI Act contains specific human oversight and audit logging requirements, which are scheduled to become enforceable as early as 2027."

What the shares did with that

SailPoint's reported revenue of $308.8m grew 16.8%, well under its ARR, because term licenses recognized up front are migrating to ratable subscriptions. The shares fell 2.8% on the day it published those results. Over twelve months they are down 12.1%, at $20.32 against the $20.00 at which Thoma Bravo re-listed the company in February 2025 still holding roughly 88.5% of the stock. The sponsor debt is gone — the term loans were repaid in full during 2025 — so what hangs over the equity is share supply.

Okta's month, by contrast, is two sessions. Its August beat-and-raise added a third in one day, and a sector-wide rally on September 14 that followed an essay on AI risk by Anthropic's Dario Amodei — a document Okta did not write — added another eighth; the other eighteen trading days were collectively negative.

Gross margins differ by thirteen points, at 79.6% for Okta against 66.5% for SailPoint, so sales multiples do not compare between them. Against trailing gross profit, Okta is at 13.24 times, up from 5.97 times on May 3 while the gross profit underneath grew about 15%. SailPoint is the dearer of the two at 14.93 times, and it is the one growing bookings at 25%.

The verdict

Okta has earned a re-rating on profits, not on growth: operating income of $107m more than doubled year on year and operating margin widened to 13.3% from 5.6%, on a debt-free balance sheet with $2.3bn of cash and guided free-cash-flow margin of 28-29%. What it has not earned is the doubling. Consensus carries no re-acceleration whatever — roughly 11% this fiscal year and under 10% the next — so every dollar above the old multiple is payment for a meter Okta's own finance chief says will not matter this year. SailPoint has the meter and the disclosed dollars and its stock has gone nowhere, which points to the 88.5% stake overhead rather than to the business.

McKinnon told investors agent identity "could be the biggest category of cyber." Oktane, where Okta stages its product announcements, comes next; the test is not another keynote but whether the November quarter reports agents as something the company charges for.

The Infrastructure Investment and Jobs Act, the 2021 law that has paid for American highway, bridge and transit work for five years, loses its program authority on September 30, and no successor surface transportation bill has been introduced in either chamber. More than 100 industry organizations warn that formula funding would revert to pre-IIJA levels and discretionary grant programs would stop making new awards — a $36.8bn annual shortfall beginning October 1.

For Vulcan Materials, the largest US producer of construction aggregates, and Martin Marietta Materials, which sells the same crushed stone and gravel alongside asphalt, paving and a fast-growing lime and magnesia business, that is not a 2026 problem. It is a 2027 one, and 2027 is where the entire growth case now sits. Consensus has Vulcan's EBITDA falling 3.5% this year to $2.58bn and Martin Marietta's falling 1.3% to $2.20bn, then rebounding 9.4% and 20.6% respectively — in the first year with no authorization behind it.

The market has been taking that year apart in public. Wells Fargo's Timna Tanners cut Vulcan to Underweight on September 9, target $254 from $305, citing an "increasingly cautious" 2027 of leaner government spending, constrained state budgets and "limited benefit from data-center related demand"; the same firm trimmed Martin Marietta's target to $585 on September 21, four days after the shares set a 52-week low of $491.41, 31% below their high. Both have traded with the 50-day average below the 200-day every session this month.

The rent is intact

Crushed stone is too heavy to ship far, so a quarry is effectively the only supplier inside a haul radius of roughly thirty miles and raises its price on a schedule rather than in response to a bid. That machinery is working. Vulcan's June quarter carried freight-adjusted prices up 3.9%, 4.7% adjusted for mix, on shipments up 1% to 59.9m tons. Martin Marietta's reported average selling price fell 2.0% to $22.74 — but that is arithmetic, not a price cut: the New Frontier quarries it bought sell rock at roughly $12 a ton, and mix-adjusted pricing rose 3.7%.

What is failing is the conversion. Vulcan's cash gross profit per ton went from $11.88 a year ago to $12.02, a gain of just over 1% on a price increase four times that size, because diesel created a headwind of nearly $40m in the quarter alone. US on-highway diesel hit a record $6.31 a gallon on September 16, more than 70% above a year earlier, the downstream result of Brent going from $61 to $118 in the first quarter after the Strait of Hormuz closure. Anirban Basu, chief economist of Associated Builders and Contractors, said the crude spike put upward pressure on "virtually every construction material". Operating income fell 9.7% at Vulcan and 17.2% at Martin Marietta on revenue that grew at both.

The company that did everything right

Martin Marietta raised full-year revenue guidance to $7.2bn-$7.4bn on July 30, shipped 17% more tons, and on August 21 closed a $13.5bn combination with Lhoist North America that adds $786m of adjusted EBITDA and makes it the leading US lime franchise. Its specialties segment set a record $152m quarter. Its aggregates gross profit per ton still fell 17%, to $6.78, of which $0.84 is the accounting cost of selling acquired inventory written up to fair value. The 52-week low came after all of it.

Demand itself has not rolled over. Single-family starts rose to 918,000 in August, up 5.2% year on year though still down year to date, and data-center construction spending is running 45.8% above last year. Knife River, the upper-Midwest contractor, holds a record $1.2bn backlog. But ARTBA's 2026 outlook has public highway and street work growing just under 1% this year, plateauing at record levels in the authorization's final year.

What the de-rating buys

Martin Marietta's price against trailing gross profit has fallen to 15.9x from 19.5x in early May, its market value from $36.96bn to $30.01bn; Vulcan's has gone to 14.3x from 17.7x. (Martin Marietta's 12x trailing earnings is an artifact of the $1.4bn after-tax gain on the cement assets it swapped to QUIKRETE in February and should be ignored.) Yet both still carry roughly 27x forward earnings, and Vulcan's forward multiple sits only 7% below its trailing one — the market is paying a premium for a business it expects to earn no more next year than this one.

So the de-rating is half-earned. The quarries kept their rent; an oil shock took the increment, and that part of the fall is honest and reversible with crude. The other half is a bet on a vote. Consensus has moved every dollar of growth into a year whose federal funding does not yet legally exist, which means an extension at current levels turns these into cheap cyclicals and a lapse turns 2027 into the air pocket the price already assumes.

Rock stays where it is; the money that buys it does not. Eight days from now, the country finds out which.

Hewlett Packard Enterprise spent the summer renaming itself. "We are leading with networking. We are at the core becoming a networking company," chief executive Antonio Neri told investors on the September 2 call, reporting the first full quarter with Juniper Networks consolidated into the results.

The quarter both supports the label and complicates it. HPE — which sells ProLiant servers, storage, Aruba campus wireless and GreenLake consumption contracts through resellers and integrators — posted revenue of $12.2bn, up about a third, with gross margin at 40.1% against 28.4% a year earlier and operating income up 182%. Networking revenue reached $2.9bn, up roughly 75%, at a 22.0% segment operating margin against 22.1% a year earlier. The growth was bought, not manufactured out of the deal's cost base.

One customer's budget

So, in a narrower sense, was the customer. On the same day as the results, HPE announced an expanded Oracle collaboration covering multi-year deployment of Juniper PTX and MX routers and QFX and EX switches across Oracle's AI data centers. It had already issued Oracle a warrant over 4,156,466 of its own shares at an exercise price of $0.01, vesting in tranches tied to Oracle's purchases and running to June 2029 — around $205m of equity handed over at a nominal strike.

That is worth the price because of what Oracle now spends. Its capital expenditure reached $55.7bn in fiscal 2026 against $21.2bn a year earlier, with roughly $70bn guided for fiscal 2027. Dell'Oro Group expects data-center switch sales into AI back-end networks to exceed $100bn by 2030. Arista, Cisco, HPE's Juniper and Nvidia are all bidding into the same pool, and increasingly into the same handful of buyers.

HPE's own AI networking line is running ahead of plan. "Cumulative networks for AI orders were $2.2 billion, surpassing our FY '26 target. As a result, we are increasing our year-end target to $2.5 billion to $3 billion," chief financial officer Marie Myers said on September 2. Total orders rose 42%, and unconditional purchase obligations — the working capital of funding backlog months ahead of shipment — stood at about $30.4bn at the end of July, from roughly $6bn. Net leverage was 1.8x, under the company's 2x target after the $1.4bn H3C sale, so the roughly $14bn Juniper purchase is not yet squeezing buybacks. Management guided fiscal 2027 to 13-17% revenue growth and earnings of $4.40 to $4.60 a share; consensus sits at the top of that range.

The same demand, priced three ways

Arista, which sells high-speed Ethernet switches running its own operating system to cloud and financial customers, is being paid very differently. Revenue grew 37.7% in the June quarter to $3.04bn, the fourth straight acceleration, and operating margin rose to 45.4% — yet gross margin fell 232 basis points to 62.9% as hyperscaler mix rose, with management saying offsetting pricing does not arrive until 2027. Deferred revenue was $6.9bn and multiyear purchase commitments tripled to $9.7bn. Microsoft and Meta were 26% and 16% of 2025 revenue. Meanwhile Nvidia took the number-one revenue position in data-center Ethernet switching in the March quarter, at 21.5% share against Arista's 20.7%, by bundling switches with accelerators. Chairperson and chief executive Jayshree Ullal was dismissive of the cheaper end of the threat on the August 4 call: "White box is certainly a tactical solution that we tend to see more in use cases that are simple, scale-up or scale-out where the actual amount of software and system requirements are low."

Cisco is the odd one out. Its revenue accelerated four quarters running to 17.6% growth in the July quarter, operating margin hit a record, and it booked $9.3bn of hyperscaler AI infrastructure orders in fiscal 2026 against $2bn the year before — though roughly 40% of that is Acacia optics rather than switching. The shares fell 9% over three months.

What the prices have and have not earned

HPE's is the move with the clearest backing: up 16.7% since the print, including an 11.3% jump on September 11, the session after Oracle detailed its spending. Its trailing enterprise value to EBITDA compressed to 16.9x from above 21x in late August because earnings arrived, not because the price fell, and at 16.2x forward with a 5.1% free-cash-flow yield it is the cheapest and highest-yielding of the three against roughly 10-13x forward in May. Arista, at 50x forward against about 32.5x a year ago, is being paid a rising multiple for growth consensus already models decelerating, by a company that no longer leads its own market on revenue. Cisco, at 21.8x forward versus roughly 18x in May, is expensive against its own pre-AI history and cheap against these peers; the two readings disagree, and both are true.

The binding constraint is not orders. HPE booked $3.1bn of AI orders in the quarter and converted roughly half of that into AI systems revenue, a gap Neri put down to memory supply — DDR5, wafers and clean rooms. The networking company HPE says it is becoming is gated, for now, by a shortage of something it does not make.

Lamb Weston reaches its next earnings date as the odd name in frozen food: the only large frozen manufacturer whose volumes are still climbing, and the only one whose shares carry a premium for it. The company said on September 8 that it will report fiscal 2027 first-quarter results on October 6.

Nothing about how it earns money resembles a grocery shelf. Lamb Weston, based in Eagle, Idaho, sells frozen potatoes — fries, commercial ingredients, appetizers — by the pound to restaurant chains and foodservice distributors, with a smaller retail business under Grown in Idaho, Alexia and customer private labels. Its prices are set in an annual contracting season rather than at a checkout, so a year's economics are largely fixed before the year begins. In fiscal 2026 that produced net sales up 2% to $6.6bn, with volume up 7% and price/mix down 6%. Operating profit fell 11%. Gross margin came in at 20.6%, against 27.3% two years earlier, and diluted earnings per share of $2.08 were less than half fiscal 2024's $4.98.

Pounds bought with price

"At the end of the day, we're winning with customers and growing and we expect some volume growth," chief executive Mike Smith told investors on the July 24 call. He is right about the pounds: North American sales rose 3% on a 9% volume increase from customer retention and contract wins. The cost of those wins arrives as price. On the same call, management traced fiscal 2027's "modest price mix investments" largely to "decisions made in the last contracting season".

The occasion those pounds serve is shrinking. US quick-service restaurant traffic fell 1.2% year on year in the second quarter of 2026, and visits shorter than ten minutes — the drive-thru trip where fries attach — dropped 6.8% in May. Lamb Weston is answering by removing capacity, not filling it: its board approved closing the Broekhuizenvorst plant in the Netherlands on June 1, affecting about 110 employees, at a pre-tax cost of $80–110m falling mostly in the year to May 2027 — two years after it opened a plant at Kruiningen, roughly 195km away. Nor is the raw material reliably getting cheaper: north-west European growers cut plantings about 11% and Idaho acreage is down some 15,000 acres, which argues against another round of contract-price relief.

Guidance for fiscal 2027 is sales flat to up 1%, adjusted EBITDA of $1.1–1.2bn against $1,147m delivered, adjusted EPS of $2.95–3.25 and capital spending cut to $380–410m. Net debt ended the year at $3.8bn, 3.4x adjusted EBITDA. Consensus of $3.02 sits inside that guidance rather than above it.

Priced as the survivor

Lamb Weston trades at 15.6x forward earnings against 22.5x trailing, and 10.6x trailing enterprise value to EBITDA, the fitter comparison given the leverage. That makes it the most expensive of the frozen names: Nomad Foods is at 7.4x forward earnings, Conagra 10.3x, Campbell's 11.8x. Sell-side targets have converged on the quote — Bernstein cut to $46 from $65, BNP Paribas Exane to $48 with a downgrade, Barclays to $55.

Someone else's dividend

The September leg down began elsewhere. On September 3, Campbell's — soup, Rao's, Pepperidge Farm, Goldfish — cut its quarterly dividend 36% to $0.25 and guided fiscal 2027 adjusted EPS to $1.65–1.80 against the $1.86 analysts expected. "Reducing the dividend is obviously a difficult decision, but it is unfortunately a necessary decision," chief executive Mick Beekhuizen said on that day's call. Campbell's fell 7.0%; Lamb Weston fell 3.8% and Conagra 3.5% with no news of their own. Campbell's is barely a frozen business — in the six months to February 1, Meals & Beverages earned $549m of segment operating profit against Snacks' $190m — and the selling was not frozen-specific either: over the past 30 days General Mills fell 10.4%, Hormel 14.3%, Tyson 10.8%, Domino's 13.6% and Wendy's 25.6%, while the S&P 500 rose 1.1%.

Conagra, which sells Birds Eye, Healthy Choice and Marie Callender's at retail shelf prices, is the mirror of Lamb Weston's contracted economics. Its fiscal 2026 revenue fell 2.9% to $11.28bn, gross margin slipped to 23.9% from 27.7% two years earlier, and a roughly $2bn impairment produced a $1.92bn net loss. It halved its dividend in July, ending 51 straight years of maintained payments, and guides fiscal 2027 volume down mid-single digits while still taking inflation-justified frozen pricing at elasticities above historical levels. Private label is the reason that hurts: US store-brand unit share hit a record 23.8%. Nomad Foods, the European owner of Birds Eye and Findus, shows the same split from the other side — its second-quarter volume fell 5.9% even as chief executive Dominic Brisby told investors on August 13 that "the frozen category in our markets year-to-date, it's up 3.4% in value terms and up 1.6% in volume terms." Frozen is not breaking; brands are losing share within it.

The shared shock has since worn off. Since September 11 Conagra has risen 4.0% while Lamb Weston has fallen 2.2% and Campbell's 3.9%, leaving Campbell's at $20.18, within 1% of its two-year low.

What the business earns and what it borrowed

Over twelve months, Lamb Weston's de-rating is earned at the profit line: the same $6.6bn of revenue now carries less than half the earnings of two years ago, and the cheaper crop has yet to reach gross margin. September added a further leg that no Lamb Weston fact explains — it came from a soup company's payout decision spreading across every food name a screen groups with it. What the remaining premium over Conagra and Nomad prices is a single proposition: that pounds bought with margin turn back into margin once the contracting cycle turns. Fiscal 2027 guidance promises no such thing this year.

October 6 is the first read on whether the pounds keep coming and the price stops falling. Lamb Weston has spent two years proving it can buy volume; it has yet to show what the volume is worth.

The session that did a year's work

A single session on February 3 took roughly a seventh off the market value of RELX, the London-headquartered owner of Elsevier's research journals, the LexisNexis legal and risk businesses and a large exhibitions arm. Thomson Reuters — which sells Westlaw, Practical Law and the CoCounsel legal assistant, largely per attorney seat — fell 15.7% the same day. Neither company had announced anything. Anthropic had shipped Claude Cowork plug-ins automating legal, sales, marketing and data-analysis work, and S&P Global separately guided 2026 earnings to $19.40–$19.65 a share against consensus near $19.96. Within days, roughly $285bn of market value had been erased across software, legal-tech and data companies on three continents, and S&P Global's outlook had dragged Moody's, FactSet, MSCI, Verisk and Nasdaq lower on the read-across.

That one day accounts for close to a third of each company's twelve-month decline — RELX is down 29.0% over the year and Thomson Reuters 41.3%. The stake is whether generative agents substitute for platforms built on proprietary content or get sold by them, and two reporting seasons have since produced an answer that the share prices have only partly absorbed.

What the companies reported afterward

RELX's first-half revenue of £4,871m rose 7% underlying, with adjusted operating margin improving to 35.5% from 34.8%. All four divisions grew: Risk 8%, Scientific, Technical & Medical 6%, Exhibitions 6% and Legal 10% — the LexisNexis division the disintermediation thesis targets most directly was the fastest of the four. "RELX delivered strong underlying revenue and profit growth and strong new sales in the first half of 2026," chief executive Erik Engstrom said on July 23, citing "a further step up in growth in Legal." He also told investors that roughly 90% of new sales value and about three-quarters of renewal value now runs through the AI-enabled platform.

Thomson Reuters reported 8% organic revenue growth in the second quarter, Legal Professionals up 10% organically and 11% excluding government, adjusted earnings before interest, tax, depreciation and amortization at a 38.1% margin, and free cash flow up 29% to $727m. "We saw strong momentum continue in the second quarter, underscored by 10% organic revenue growth in our 'Big 3' segments," chief executive Steve Hasker said on August 5. The company raised full-year organic growth guidance to about 8%, the top of its prior range, and management attributed the legal acceleration to agentic products, saying on the call that "a large part of growth is coming from the agentic offerings, both Westlaw and CoCounsel".

The meter that is genuinely slowing

One part of RELX is not a subscription at all. Inside LexisNexis Risk Solutions, subscription revenue is 39% of the total and transactional revenue, including volumetric contracts, is 61%, so the division is paid per query on insurance underwriting and identity checks rather than per customer budget. That volume is cooling: LexisNexis's own US Insurance Demand Meter put year-on-year auto shopping growth at 3.2% in the first quarter and 1.4% in the second. Part of the de-rating is earned.

The legal exposure is narrower than the selloff implied. Wells Fargo, cutting Thomson Reuters to Equal Weight and its target to $95 on March 23, noted that AI-native startups such as Midpage are trying to replicate the Westlaw library, which is roughly a quarter of company revenue. Yet Anthropic itself launched Claude for Legal on May 12 with connectors into Westlaw, making the library a layer inside the agent; RELX answered on August 24 with a "Legal Intelligence Engine" rebuilding its Protégé assistant inside Lexis+ around agentic orchestration.

Where the group trades

The market has discriminated rather than marked down everything sold by subscription. Over twelve months MSCI fell 1.8%, Moody's 4.0% and FactSet 4.5% — franchises with benchmark or regulatory lock-in — against Clarivate's 52.1%, Thomson Reuters' 41.3% and Equifax's 38.2%. Holding today's consensus 2026 earnings of $4.43 a share constant, Thomson Reuters has gone from 36.8 times forward earnings last September to 21.5 times, and from 48.4 times at its 2025 peak, a 41% compression into raised guidance. RELX trades at 19.7 times trailing earnings on a 6.7% trailing free-cash-flow yield; because consensus has 2026 earnings of £1.41 a share against roughly £1.24 trailing, its forward multiple sits below that.

The last month carries no company news at either name. Both jumped on September 14 — Thomson Reuters 8.7%, RELX 5.7%, while the S&P 500 fell — after Dario Amodei published an essay on September 12 urging frontier labs to slow capability gains, then gave it back by September 18, the day the Federal Reserve raised rates a quarter point to a 3.75%–4.00% range. Both have traded above their long-term moving averages since late August.

The verdict

The business earns almost none of the twelve-month decline. Organic growth accelerated at both companies, margins expanded at both, and Thomson Reuters raised guidance; the one honestly deteriorating input is US insurance shopping volume feeding RELX's per-query Risk revenue, and that division still grew 8%. What the fundamentals cannot explain is the February gap, which priced a substitution that the subsequent six months of disclosure contradicted — JPMorgan analysts wrote during the selloff that "current market pricing is expressing the most bearish possible outcome". The recovery has been uneven: RELX now sits 5.9% below its pre-gap close and Thomson Reuters 12.1% below, which is the market saying that per-seat legal research is the more substitutable meter, even as it is the one growing fastest at both.

The test is arithmetic and close at hand. Thomson Reuters guided to about 8% organic growth for the full year, and it has already printed 8% with legal at 10%. A third-quarter print that holds that line leaves the February discount without a fact behind it.

Viasat switched on the second of its three ViaSat-3 satellites over the Americas on September 17, and with it declared the constellation complete — more than a terabit per second of throughput, with steerable beams pointed at shipping lanes and flight routes, and global fleet bandwidth tripled. The first of the three satellites suffered an antenna failure after launch in 2023. The program it completes was conceived to sell satellite internet to houses.

That business is going away. Fixed broadband revenue fell 27% year on year in the June quarter, leaving roughly 115,000 subscribers, which is what happens when Starlink sells the same product from lower orbit. So the interesting question about Viasat is not whether it lost the home; it is what the tripled capacity now gets sold into, and whether the cash it throws off arrives before the debt taken on to buy Inmarsat comes due.

Tails and appropriations

What pays Viasat today is not subscribers. It is aircraft and governments. Aviation revenue rose 11% on a 10% increase to 4,530 commercial aircraft in service, billed per installed tail to airlines that own the passenger. Government satcom grew 10%. Total awards in the quarter were $1.3bn and backlog reached $4.2bn, up 19%, led by the Defense and Advanced Technologies unit. "It is going to be the fastest-growing part of our business," chairman and chief executive Mark Dankberg told investors on the August 4 call, where he also framed the strategic review of that unit around growing it rapidly. Activist Carronade Capital has pressed for separating the defence business outright, and two of its nominees sat on a new Strategic Review Committee formed in May.

The income statement has not caught up. Revenue of $1.157bn was down 1.2% and the quarter carried a $51.7m net loss. Free cash flow rose 19% to $72m, net debt of $4.8bn equals 3.2 times trailing adjusted earnings against 3.6 a year earlier, and Inmarsat's $568m settlement with Ligado Networks begins arriving in roughly $16m quarterly instalments from September 30. Management guides fiscal 2027 to mid-single-digit revenue growth, capital expenditure of $950m to $1bn and about $180m of free cash flow.

The other satellite winner is not a satellite story

Iridium, which runs 66 crosslinked low-orbit satellites and sells airtime per billable subscriber, matched Viasat's twelve-month gain for an entirely different reason. On June 29 Rocket Lab agreed to buy it for $54.00 a share, about $8bn including debt, and the stock gapped 25% in a session. The buyback was terminated the day before signing; earnings calls and guidance are suspended until a mid-2027 close. Holders get $27.00 in cash plus Rocket Lab shares inside a collar that stops protecting them if Rocket Lab's average price falls below $67.50. Rocket Lab closed at $69.67 on September 21. Iridium closed at $48.02, an 11% discount to the offer.

The underlying business is fine and unexciting: 2.63m billable subscribers, up 6%, revenue up 4%, operating income down 32%. Peter Beck, Rocket Lab's chief executive, was explicit about what he bought: "L-band is important because if you want to do safety-critical comms, you need that band to penetrate through weather and harsh conditions." Spectrum, not subscribers.

What the business earns

Viasat's re-rating from roughly 1.4x forward sales in May to 2.17x now is paid for by the measurable half: backlog, tails, four-tenths of a turn of deleveraging. It is not paid for by earnings, which consensus does not expect to turn positive before fiscal 2028, and the 27% decline in fixed broadband is a real hole that defence growth is covering rather than filling. Iridium's move is not a judgement on satellite economics at all — price it against Rocket Lab's shares, not its own subscriber base. And in the same neighbourhood, Gogo, which sells connectivity to business jets, is down 72% over twelve months. There is no satellite trade here, only three separate ones.

The capacity is finally in orbit. The customer it was built for has left, and the airlines and defence ministries that must absorb it were never the plan.

On September 3, the three big publicly traded less-than-truckload carriers each put out August operating data within hours of one another. They serve overlapping shippers on overlapping lanes, and they reported three businesses that barely resemble each other.

That matters because of how this industry actually earns a dollar. A less-than-truckload carrier consolidates pallet-sized freight from many customers into one trailer, moves it across a fixed network of owned terminals and dock doors, and bills by hundredweight — per hundred pounds, not per shipment. Every driver, dock and door is a fixed cost that runs whether the trailer is full or not. So the only two levers are what you charge per pound and how much freight you can pour over a given set of docks. The three carriers are pulling different ones.

Old Dominion prices; Saia fills

Old Dominion, the largest US less-than-truckload carrier with 251 service centers, is holding price and letting freight walk. August shipments per day fell 2.4% and tons per day 0.9%, while revenue per hundredweight excluding fuel rose 4.8% quarter to date. It then announced a 4.9% general rate increase effective October 5 — a pricing action taken into a falling shipment count. "Old Dominion produced solid revenue growth for July and August, with underlying demand trends remaining relatively consistent as the quarter has progressed," chief executive Marty Freeman said in the September 3 update.

It is working on the income statement. June-quarter revenue rose 10.4% to $1.554bn after falling 2.9% in the March quarter, operating income grew 30.0%, and the operating ratio — operating costs as a share of revenue — improved to 70.1% from 74.6%.

Saia, which absorbed a large share of the terminals surrendered when Yellow collapsed in 2023, is doing the opposite. Its August tonnage per workday rose 8.7% on shipments per workday up just 1.1%, with weight per shipment up 7.5%. Since the meter is the hundredweight, heavier freight fills the dock while diluting yield: Saia's June-quarter revenue per hundredweight ex-fuel fell 2.2%, on a 3.9% rise in average shipment weight.

Saia's revenue still grew 17.1% to a record $956.5m and operating income 26.0%. But its operating ratio improved only to 86.9% from 87.8%, so the gap behind Old Dominion widened over the year from 13.3 points to 16.9. The reason is arithmetic the company does not hide: after roughly $1bn of real estate spending since 2022 and 33 new terminals, its 2023 and 2024 vintage facilities still run an operating ratio in the low 90s against a company average of 86.9%.

XPO answered the same August a third way, taking 5.7% more shipments that were 1.8% lighter — buying shipment count where Saia bought weight.

The fuel line, and the September break

Diesel is distorting all of it. On-highway prices hit a record $6.285 a gallon in the week of September 14, roughly 68% above a year earlier. Surcharge tables inflate reported revenue — Old Dominion's revenue per hundredweight is up 11.3% as reported against 4.8% without fuel — but they reset on a lag, so the cost arrives before the recovery. J.B. Hunt's mid-September warning on exactly that point took the whole freight complex down with it.

The shares had already broken. Old Dominion fell 15.7% in the month to September 21 and 20.7% over three months, leaving it 30% below its twelve-month high; Saia is down 21.3% over three months. Neither move came from guidance: Citi's June downgrade to Sell cited valuation rather than the business, and the underlying demand data is stabilizing — the American Trucking Associations' tonnage index rose 0.5% in August, its first monthly gain since March.

What each de-rating earns

Old Dominion now trades at 33.7 times trailing earnings and 30.1 times forward, against 46.1x in May, on 19.9 times enterprise value to EBITDA. Most of that compression is the market declining to pay a growth price for a carrier whose volumes are still shrinking in its fourth soft year — a de-rating the business largely earns, even as earnings improve. Saia's trailing multiple fell from 49.3x in May to 33.4x, and on enterprise value to EBITDA — the measure that matters when one company has borrowed to build terminals and the other self-funds — it is the cheapest of the three at 14.9x, against consensus earnings growth of 21.9% this year and 25.3% next. XPO sits between them at 18.8x.

The honest split is this: Old Dominion's problem is a price that got ahead of a network it is deliberately shrinking into, and Saia's is a network that got ahead of the freight. Only one of those is fixable by the freight cycle arriving.

On October 5 Old Dominion tries to charge 4.9% more for moving less. Whether shippers pay it — with a record fuel surcharge already on the same invoice — is the cleanest test of pricing discipline this cycle has offered.