DK Street Journal

Clearway Energy, which owns roughly 5,000 net megawatts of American wind and solar alongside a gas-fired fleet, restructured the offtake contracts on its Texas wind farms in June at higher prices, with terms now running past 2040. In January it had signed nearly 1.2 GW of twenty-year power purchase agreements with Google across three states, more than $2.4bn of investment with first power due in 2027. The shares are down about a fifth over six months.

The reading on offer for the renewable yield vehicles is that the long-dated contracted revenue which is their whole earnings engine is quietly decaying: agreements recontracted below the expiring ones, tax credits phasing out, interest absorbing the distribution. Clearway's disclosures record the reverse, and so does the market for new power contracts. What moved both it and Brookfield Renewable, the owner of hydro, wind and solar across the Americas that sells most of its output under long-term contracts, was the price of money: the Federal Reserve raised its target a quarter point on 16 September, its first increase since 2023, the 30-year Treasury printed 5.42% on 20 October against 4.60% a year earlier, and utilities fell 6% in a month to 17% below their 52-week high.

The guidance cut was weather

Clearway's June quarter grew revenue 22.7% to $481m, with operating income up 36.5% to $116m and gross margin at 69.0% against 66.6% a year earlier. In August it cut the midpoint of 2026 cash available for distribution about 8%, to $430-470m from $470-510m, on below-normal wind and solar output, and left the 2027 target of $2.70 a share and the 2030 range of $2.90-3.10 where they were.

"While we are lowering our 2026 financial guidance due to factors outlined in our mid-July operational preview, our team is focused on maintaining our trademark operational excellence to uphold our historic track record of meeting our financial targets," chief executive Craig Cornelius said in the 5 August results release.

The policy channel has gone the same way. Clearway has safe-harbored tax-credit eligibility on 14.7 GW of wind and solar, with 9 GW and more qualified for service in 2031-2034, and has sourced equipment for its entire development program through 2027 and substantially all of 2028 — all of it predating the 4 July start-construction cliff that accelerated the credit's termination. Transfer discounts on those credits remain narrow, $0.92 to $0.945 on the dollar for production credits. And new contracts are getting dearer: the average North American solar agreement rose 4.9% in the third quarter to $64.23 per MWh, with wind up 13.8% on the year.

The constraint is the balance sheet, and Clearway named it. The company needs $0.5-1.0bn of external equity through 2029 to fund a $3bn capital program and has raised about $50m, which has paused third-party purchases. "When it's not where we want it to be or valuations for third-party acquisitions, we think, are not really reflective of their intrinsic value, then we step out," Cornelius told investors on the 5 August call. At $31.43 the shares sit at 0.70 times book — below the accounting value of the plants — on 13.8 times trailing EV/EBITDA and a 10.4% trailing free-cash-flow yield, or about 11.6 times the 2027 cash target, a figure derived from the reaffirmed guidance rather than one the company prints.

Brookfield's record came partly out of the sale ledger

"We delivered another record quarter, generating FFO of $421 million, or $0.62 per unit, up 13% or 11% per unit year-over-year," Connor Teskey said on Brookfield Renewable's 31 July call. Roughly $175m of that sat in the hydro segment as other income from developed-asset and non-core sales; the partnership closed two-thirds of a sale of about 2,100 MW to the Northview platform in the quarter. Strip the gains and per-unit funds from operations fall near $0.36 against a declared distribution of $0.392, coverage of about 0.92 times on that derived basis. Interest expense rose 5.4% to $658m, more than the whole quarter's funds from operations. Reported revenue fell 1.8%.

At $28.91 the $1.568 annualized distribution yields 5.42%, the same as the 30-year Treasury on 20 October, leaving nothing for equity risk. Clearway's $1.90 annualized dividend yields 6.05%, roughly 63 basis points of spread, down from about 90 in late September.

Demand is not the dispute. Brookfield's framework with Microsoft commits it to 10.5 GW delivered between 2026 and 2030, and Teskey argued in October that supply will trail demand for compute and power by three to ten times through the decade: "There's a lot of growth in the market and that can overwhelm slightly higher cost of capital."

The year's loss belongs to a share class

The twelve-month decline attributed to this corner of the market is largely an artifact of the corporate simplification. BEPC, the corporate twin exchangeable one-for-one into Brookfield Renewable units, fell 25.0% over the year as its premium collapsed from 38% to 1.6%; the partnership units themselves rose 2.1% and Clearway 1.4%. Holders of the corporate shares approved the transaction on 14 October with more than 99.7% in favor; the unitholder meeting was adjourned to 29 October.

So the selloff is mostly explained, and explained the same way for both: a long-dated, indexed stream of cash discounted at a rate that has moved almost a full point in a year, which is why NextEra fell 17.6% and Ormat 21.8% over the same six months. What the rate does not explain is why the two deserve the same multiple. Brookfield's de-rating has a business counterpart — a payout with no spread over government paper and coverage that depends on continuing to sell plants — while Clearway's contract meter is lengthening and its credits are locked.

Clearway's fuel costs nothing and its competitors' fuel is getting scarcer. What it cannot buy cheaply, at 0.70 times the value of its own turbines, is its own equity.

Fair Isaac's competitor in mortgage credit scoring now has a published price and a published expiry date. Equifax said on 8 October that it will keep selling VantageScore 4.0 to mortgage lenders at $1 through the end of 2028, a commitment it claims could save the industry about $1bn, and TransUnion has fixed its standalone price at 99 cents through December 2028, saying it will not charge separately for that score when a customer also buys a FICO score. Both bureaus distribute Fair Isaac's score for a living.

The company's fall since September has been read as the market pricing one lost customer and a weaker origination forecast. The bureaus' commitments describe a slower problem. Price, not volume, is what has been growing the Scores segment at Fair Isaac, which sells credit scores and decision-management software from Bozeman, Montana, and the price of the alternative is now set for three more years.

Nearly all of the mortgage gain was the fee

In the June quarter Scores revenue rose 41% to $458.9m, with the business-to-business half up 49%, which the company attributed "primarily" to a higher mortgage origination scores unit price. Mortgage origination revenue rose 97%; mortgage score volumes grew low single digits. Mortgage is 62% of Scores revenue.

The royalty shows the arithmetic. Fair Isaac took its wholesale mortgage fee from $4.95 to $10.00 a score for 2026, after moving it from $0.60 to $4.95 over the prior five years. A roughly 102% price step produced a 97% revenue gain, so units and mix contributed close to nothing. Group revenue grew 25.7% to $674.2m with operating margin at 53.8%, against 48.9% a year earlier.

The rival is already inside agency underwriting

Fannie Mae's Lender Letter LL-2026-06 and Freddie Mac's Bulletin 2026-H, both dated 9 September, made VantageScore 4.0 available to every approved seller without prior written approval, and 37% of conventional applications created in Fannie Mae's Desktop Underwriter in September included a VantageScore 4.0 score. Equifax says nearly 2,000 mortgage lenders and resellers now access the model, that pulls rose 230% between April and August, and that more than 165 lenders use it exclusively at the $1 price for certain loan types. Rocket Mortgage said on 28 September that it will default to VantageScore 4.0 this quarter on all eligible loans, including loans delivered to Fannie Mae and Freddie Mac, citing average closing savings of $1,600 for borrowers who benefited. Senator Josh Hawley asked the Federal Trade Commission on 23 March to investigate the price increases.

"We're not crazy about lenders choice as a policy. We think it's bad policy because it encourages gaming," chief executive William Lansing said on Fair Isaac's 29 July earnings call, adding that volumes were not falling, "which suggests that they're pulling both scores." Rocket's decision came two months later.

What the company is doing about the dollars

Fair Isaac's answer is its Mortgage Direct License Program, which lets tri-merge resellers calculate and distribute FICO scores without going through the bureaus, at $4.95 a score plus a $33 fee on funded loans, or $10 flat. Six resellers are under contract — Xactus, Cotality, Ascend Companies, CIC Credit, MeridianLink and Credit Interlink — together more than half the reseller market by origination volume. "This change eliminates unnecessary mark-ups on the FICO Score," Lansing said announcing the program on 1 October 2025. FICO Score 10T becomes available to FHA-approved lenders on 1 January 2027. On 1 October the company also committed to eliminating about 15% of its positions, with roughly $27.0m of pre-tax charges in the fiscal fourth quarter, which protects margin on the revenue that arrives and does nothing for the score count.

A 10% fall inside a software rebound

The shares closed at $823.27 on 22 October, 23.7x trailing earnings against roughly 62x in January, and 16.8x the fiscal-2027 consensus of $48.92 a share drawn from 13 analysts. Fair Isaac fell 10% over the month to that date while most of the application-software names beside it rose, in a rebound that carried the S&P 500 software and services index to its highest level since November 2025. Vertex, which sells the indirect-tax engine that sits inside customers' SAP and Oracle systems, gained 29.9% and trades at 4.84x forward gross profit against 3.70x in late September, with revenue growth decelerating to 10.5%. BlackLine, whose software automates the corporate accounting close, rose 23.4% to 3.45x forward gross profit from 2.85x, on annual recurring revenue of $719m growing 6%. Neither has reported since early August.

The fee, not the quarter, is what is being priced

Fair Isaac's reported figures account for none of its decline, and the consensus estimate is where the argument actually sits: $48.92 assumes the $10 score survives into a year in which the Mortgage Bankers Association's 11 October forecast puts origination counts at 5.0 million, 5.6% below 2026. The signed direct-license channel and 10T's entry into FHA lending are genuine defences of that fee. Against them sit a rival locked near a dollar until 2028, a large lender switching its default, and a third of agency applications already carrying the other score. If lenders keep pulling both, the price holds and the forward earnings arrive. If they stop, the price step unwinds into a shrinking loan count, and the Software segment is too small to cover the gap: Platform recurring revenue grew 62% to $413m, but total Software revenue rose 2% to $215m.

Fair Isaac reports fiscal fourth-quarter and full-year results after the close on 4 November, a quarter billed before Rocket's switch took effect. The question those results cannot answer is what a score is worth to a lender who already has a cheaper one sitting in the same credit file.

American Airlines told investors on 22 October that it now expects to lose between $1.00 and $1.60 a share for the full year, abandoning a range that had allowed for a profit, in the same hour it reported the fastest quarterly revenue growth its chief executive says he has seen. Revenue rose 19.3% to $16.3bn. Operating income fell to $115m, a margin of 0.7%, and the quarter produced a $269m net loss.

The fall across this shelf since mid-September has been read as the main cabin finally giving way: domestic seats outrunning demand, coach fares capped, premium cabins and co-brand credit-card payments left holding the earnings up. The four quarters reported between 9 and 22 October test that belief and break it. Coach is the strong half, fares are the best on record, and the seats are being taken out rather than added.

The coach cabin outperformed the front of the plane

Southwest posted unit revenue up 21.6% year on year, a company record and above its own 17.5%-19.5% guide, achieved while lapping the introduction of bag fees. United's total unit revenue rose 15.9%, and inside it non-premium revenue rose 18.4% on a 0.5% capacity increase, for coach unit revenue up 17.8% against premium unit revenue of 14.1%. Basic Economy revenue rose 16% and loyalty 17%. Delta's system unit revenue rose 15.4%, with main-cabin unit revenue up 17%, a third consecutive quarter of improvement. American's rose 14.5%, domestic 13.6%, after a deliberate choice: "American traditionally has been a leader in load factor," chief commercial officer Nathaniel Pieper said on 22 October, and the company "really identified yield improvement as a top priority."

Nor is anyone flooding the domestic market. About 3% of domestic seats were pulled from the 25 November to 31 December schedule after American, United and Southwest announced cuts around the 16 September Morgan Stanley conference. American said then that demand remained strong at higher fares and the problem was the economics of low-margin flying. Delta guides fourth-quarter revenue up roughly 20% with main-cabin seats down year on year.

Every carrier that reported posted double-digit unit revenue growth. Every one of them also paid close to twice as much for a gallon of fuel.

One input took the whole gain

The Strait of Hormuz has been closed to commercial shipping since 28 February, and because refined-product flows were disrupted along with crude, jet fuel crack spreads now run $85 to $95 a barrel against roughly $15-20 before. Spot jet fuel averaged $3.72 a gallon in August, 84% above a year earlier. Fourth-quarter assumptions sit close to double last year's: American roughly $4.70 a gallon, United $4.80, Southwest $4.55, Alaska $4.65 to $4.85 against an original $2.50 plan.

"Higher fuel added $1.9 billion to our costs in the third quarter. Nearly 70% more than we paid during the same period last year," American chief executive Robert Isom said on 22 October. Delta's Ed Bastian, on 9 October: "we expect to deliver pretax profits of roughly $4.5 billion fairly close to last year's profitability. Even with our fuel bill expected to increase by 60%, or $6 billion."

Recapture and the balance sheet set the dispersion

Delta kept revenue growth accelerating all year, to 21.1% in the September quarter, and still watched operating income fall 13.7%; it has a refinery worth about $0.40 a gallon of offset, diversified revenue at 61% of the total and American Express remuneration above $9bn tracking to $10bn, and trades at 15.7x trailing earnings against 14.6x forward. United recovered 72% of a fuel increase of $1.8bn and was the only one of the three to grow operating profit, up 3.7%; its forward earnings multiple of 13.5x sits above its 9.3x trailing precisely because consensus 2026 earnings are 28% lower, and 2027 consensus puts it on 8.3x. Southwest expanded operating margin 4.8 points to 5.4% despite a fuel headwind of $820m, on 19.9x trailing earnings against 14.9x forward, though part of its reported passenger-line gain came from a credit-card reclassification.

American has no forward earnings multiple to quote, against a consensus 2026 loss of $1.03 a share, negative book value and $36bn of debt behind an equity value of $8.3bn. On enterprise value to earnings before interest, tax, depreciation and amortization it is the dearest of the three at 13.6x, against United's 7.1x.

Alaska is the counter-example to the mix argument. Premium revenue is 36% of its mix, up three points, premium traffic rose 24.5% and loyalty cash 21% to $721m. It still lost $147m in the quarter and guides to a full-year adjusted loss of $5.00 to $5.50, and the shares sit 41% below their 52-week high.

The prices argue the same way

Over 30 days United fell 11.5% and American 7.4%; Delta, at 15.4% below its high, is the least damaged. The shelf's headline decline is mostly JetBlue, down 41.8% in a month to $2.65. The five-session slide from 12 to 19 October hit all four majors within three points of each other and reversed into the prints: every one of them rose on 22 October.

So the revenue line explains none of the de-rating and the earnings line explains most of it. Fares are not the variable in doubt; the variable is how much of a doubled fuel bill each carrier can pass through before its balance sheet decides the outcome. "To get there in 2027, we probably also need at least 1 of the 3 following things to happen: Lower fuel, lower industry capacity, or a continued willingness for the industry to trade load factor for yield," United's Scott Kirby said on 21 October. Two of those three are inside the industry's control, and both are already in motion: American says 2027 capacity will decelerate considerably, and Kirby says he will cut loss-making flights with no hesitation.

The third is not. Four ships transited the Strait of Hormuz on 18 October, against roughly 85 a day before the closure, and no one has put a date on its reopening. Every carrier's path back to profit is written against a forward curve that assumes one.

Applied Digital, which designs and operates single-tenant data centers for artificial-intelligence and high-performance computing workloads in North America, booked $341.9m of revenue in the quarter ended 31 August. The largest line inside it was fit-out construction performed for tenants, at $183.5m, against $65.8m of base rent and $37.8m of legacy bitcoin hosting. That mix pulled group gross margin to 15.3% from 42.5% two quarters earlier, even as adjusted earnings before interest, taxes, depreciation and amortization reached $64.4m against $0.5m a year earlier.

The shares have been read as a verdict on whether AI capacity is still worth what tenants agreed to pay for it. The company's own October disclosures describe the opposite happening on both sides of its economics: the leases repriced up and the debt repriced down. What moved against the equity was the price of the money that funds the gap between pouring concrete and sending an invoice.

The coupon fell 225 basis points while the stock fell by a third

"Our first Polaris Forge 1 related notes placed last year carry a 9.25% coupon. This tranche priced at 7%," chief financial officer Mohammad Saidal L. Mohmand told investors on 7 October, on $1.59bn of senior secured notes due 2031. The improvement was attributed to the investment-grade credit quality of CoreWeave, the company's anchor tenant. On the same call, chairman and chief executive Wesley Cummins said the company was targeting roughly 250MW of expansion leases by year-end at "materially higher" pricing, meaning 15%-plus rate increases and potentially longer duration, and expects to place more than 600MW into service over the next twelve months against 250MW in the prior twelve.

The contracted book is not speculative. As of 31 May the company disclosed 1,410MW of contracted critical IT load and approximately $36.2bn of take-or-pay leases with CoreWeave and investment-grade hyperscalers on initial 15-year terms.

What the equity is funding is construction

Trailing free cash flow runs at -77.2% of market value. That is the mechanism. A company placing 600MW into service spends the capital first and collects rent afterward, so its equity is a claim on the terms of the next financing rather than on the last quarter's leases. Those terms changed. The 10-year Treasury yield rose 87 basis points in the third quarter and touched its highest level since 2002 on 7 October, and the window for new data-center paper narrowed with it: listings for a Blackstone-backed AirTrunk vehicle, an Nvidia-backed developer, Firmus, and a SoftBank-backed developer, SB Energy, were pulled or paused, and a Korean data-center listing fell 30% on debut on 20 October.

The valuation has finally done work. With no earnings to anchor against, the usable measure is price to gross profit, at 33.4x trailing against 60.7x in late August; trailing price to sales is 6.8x and 6.1x forward, down from 14.3x in August and 28.7x in early May. The thin margin on fit-out revenue is why the sales multiple flatters this business and the gross-profit multiple does not.

Core Scientific's version of the problem is one payer

Core Scientific, which runs data-center facilities for digital-asset mining and colocation hosting for third parties, grew second-quarter revenue 108.8% to $164.2m with gross margin recovering to 42.7% from 6.4%, and still posted a $1.155bn net loss and negative book value. Its trailing free-cash-flow yield is -22.8%. CoreWeave supplied about 77% of first-half revenue under a 12-year colocation contract worth roughly $850m of average annualized revenue, take-or-pay, with no unilateral termination right. Durable terms, one counterparty, whose debt has traded toward distressed levels. Shareholders rejected CoreWeave's acquisition of the company in 2025, leaving it standalone. Its $13.20 close on 22 October was its low for the year, against a high of $29.16.

The split sorts by who needs outside capital

Over three months Applied Digital fell 36.5% and Core Scientific 44.1%, while Equinix fell 7.4% and Digital Realty 4.2%. Both incumbent landlords raised 2026 guidance in late July and de-rated anyway: Equinix, whose interconnection-rich colocation campuses connect carriers, clouds and enterprises, to roughly 22.2x the midpoint of guided adjusted funds from operations of $42.69–$43.29, against the 25–30x that franchise has historically commanded; Digital Realty, the wholesale and hyperscale landlord, to roughly 20.9x guided core funds from operations of $8.15–$8.20 against a 22–25x band. They fund from investment-grade balance sheets. The two operators fund from the market that just tightened.

So almost none of the four-to-tenfold difference in the fall is leasing. Applied Digital's rates are going up by its own account, and Digital Realty's second-quarter renewals repriced cash rents 25.4% higher. What repriced is the cost of capital spent before a tenant's first payment, plus, at Core Scientific, the standing of a single payer. The remaining question is political: investors now see the 3 November midterms as a channel for data-center backlash, through hearings and investigations that analysts doubt could materially stall construction.

"Put simply, our view is that every new restriction elsewhere makes what we already own harder to replicate and more valuable," Cummins said on 7 October. He is describing a moat that only pays once the megawatts are energized, and the next 600 of them have to be financed before a single invoice goes out.

Insulet took type-2 diabetics past 40% of its new American customers in the June quarter, and then cut the growth forecast for its home market because too many of them stopped using the product within three months. The population it could not hold on to is the same one paying Dexcom's sensor business, and on October 14 Dexcom's CONNECT trial was published in The New England Journal of Medicine — the evidence file it has handed to Medicare in support of covering those patients.

The halving of Insulet's shares over the past year has been read as proof that the type-2 opportunity both companies sell is smaller than advertised. The company's own numbers say something narrower. Omnipod revenue rose 24.6% to $795.9m, the global customer base grew 23%, gross margin held at 70.2% against 69.7% a year earlier, and net income more than quadrupled to $95.0m.

What broke was onboarding: one company, one country, one segment.

The first 90 days did the damage

Insulet, which sells the tubeless Omnipod insulin pump through the pharmacy channel, reduced full-year US Omnipod growth guidance to 17-19% from 20-22% on weaker type-2 retention and utilisation, concentrated in the first 90 days after a patient is set up. "As CEO, I take accountability for our reduced U.S. outlook," chief executive Ashley McEvoy told investors on the second-quarter call, adding that "once Type 2 customers remain on Omnipod for the first 90 days, retention rates stabilize, a trend that we also see in our Type 1 population." The fixes are operational: more customer-care staff for insurance navigation, sales compensation re-weighted to reward retention in the first 45 days, a refined sampling programme.

Group guidance moved to roughly $3.25-3.30bn from $3.28-3.33bn, under a 1% cut, with international Omnipod growth raised to 30-32% from 26-28%. At $141.90 the shares carry 21.8x this year's consensus earnings of $6.51; the same consensus against last October's price implies about 50x. Two things sit inside that compression which the income statement does not show: a March 12 voluntary Medical Device Correction and a second on May 26 covering Omnipod 5, DASH and Eros lots, with a securities class action alleging the first affected more pods than disclosed, and the collapse of drug-delivery revenue to $5.8m, down 78%.

Dexcom gets paid for the patient Insulet lost

For a sensor, the type-2 push is clean volume. CONNECT randomised 283 adults at 22 US primary-care sites over 26 weeks and cut A1C by 1.6 percentage points from a baseline of 8.8%, 0.9 points better than routine care. Dexcom's own survey presented at the EASD meeting finds 58% of type-2 patients taking GLP-1 drugs wearing a sensor against 27% of type-2 patients not on insulin, with use nearly quadrupling among non-insulin patients after a GLP-1 starts. Weight-loss drugs recruit sensor customers; they do not have to recruit pump customers.

The gate is reimbursement. Medicare today covers a sensor for any diabetic using insulin or with documented problematic hypoglycaemia; patients on oral drugs alone do not automatically qualify. Chief executive Jake Leach said Dexcom "submitted the CONNECT data for publication and provided the evidence to CMS in support of the non-insulin coverage expansion", a segment he said has "delivered some of the largest growth of our business over the last couple years".

That is a 2027 argument. What Dexcom has delivered now is margin: non-GAAP gross margin of 64.1% against 60.1% and operating income at 25.1% of revenue against 19.2%, on revenue up 13% to $1.308bn — down from 21.6% growth in the September 2025 quarter. Full-year revenue is guided to 11-13% after 15.6% in 2025, and the shares at 30.0x this year's consensus are about 17% dearer on that basis than a year ago.

Tandem's volume is its own installed base

Tandem Diabetes Care, which sells the tubed t:slim X2 and Tandem Mobi pumps, shipped more than 33,000 pumps in the quarter for 10% growth and lifted gross margin to 57%, up 460 basis points, narrowing operating loss to -$13.8m on revenue growth of 5.8%. But new pump starts were flat year over year and a few hundred units below plan, and management needs 12-13% growth in new starts in the second half to reach $1.065-1.085bn for the year. Its answer to Omnipod's form factor, a tubeless Mobi, is at the FDA and not expected in the US until 2027. At 1.95x trailing gross profit against Dexcom's 9.76x, it is priced as the loser of a pump race it also supplies: Tandem added Dexcom G7 15-day compatibility this quarter, so its lost starts are partly revenue-neutral for Dexcom.

Why the same patient splits the two businesses

The results account for almost none of Insulet's de-rating. A 23.5% revenue quarter with margins intact does not produce a 56% fall over twelve months; the recalls, the litigation and doubt about how much of the insulin-using type-2 market actually converts are doing that work, and the guidance cut is the smallest part of it. Dexcom's 25% twelve-month gain came with decelerating revenue and a wider multiple, which makes it a position on the coverage decision more than on current volume. Tandem has the margin recovery and neither the growth nor the device.

The mechanism underneath is the product. A sensor is a consumable any persuaded, reimbursed patient can wear, so prescribing patterns in diabetes drugs feed it directly. A pump asks the patient to keep wearing a device through a difficult first month, so the same expansion imports churn. From January 2028 CMS moves sensors and pumps to monthly rental payment with supplies bundled into the rate, a rule diabetes organisations are lobbying Congress to delay. Dexcom reports on October 29 and Tandem on November 5.

Insulet's quarter, due November 4, is the first that can show whether the type-2 patients it signed up over the summer are still wearing pods. The sensor only has to be prescribed; the pump has to be worn, and that is the harder sale to make twice.

Powell Industries, the Houston builder of custom switchgear and modular "E-House" substation packages for refineries, liquefied natural gas terminals, utilities and data centers, has given back a quarter of its value since July without reporting a single new order. What it reported in August was a record.

The October slide across power-equipment shares has been read as the moment switchgear scarcity pricing cracks. The calendar argues against that. September consumer prices rose 3.7% against 3.6% expected and producer prices jumped 6.3% annually, the largest annual gain since March 2022, with Fed officials signalling another hike; Treasuries slumped on 14 October as the S&P 500 fell 1.11%. What moved was the price of money, and it moved at the customer rather than in the supplier's price list.

The shelf sorted itself by who borrows

Over the 22 sessions to 21 October, Vertiv, which sells data-center power and cooling systems, fell 16.1%; Powell fell 12.6%; nVent, which makes electrical enclosures and connections, fell 11.6%; and Eaton, the largest listed power-management supplier, fell 11.5%. Diversified industrials went the other way, with Emerson up 5.4% and Ametek up 1.5%. Hubbell, the Shelton, Connecticut maker of utility distribution and substation products and of wiring and lighting, was flat at +0.3%.

The stress is visible where the buildout is funded. CoreWeave fell 15.5% on 14 October as its bond yields surged, and listings for a Blackstone-backed AirTrunk real-estate trust, Nvidia-backed Firmus and SoftBank-backed SB Energy were pulled or paused.

Powell's last number was two months old and the biggest it has printed

In the June quarter Powell booked $934m of new orders, up 158%, for a book-to-bill of 3.0x in the quarter and 2.2x for the fiscal year to date. Backlog reached nearly $2.4bn, against $1.4bn at the end of September 2025. Three mega orders did it: a data-center award above $400m, a $75m petrochemical order for a fertiliser plant, and roughly $60m of liquefaction work on the Gulf Coast.

"Commercial momentum across our key end markets continues to accelerate as Powell was awarded a record $934 million of new orders in the quarter," chairman and chief executive Brett Cope said on the 4 August call, where he also described the backlog passing $2bn for the first time in the company's 79-year history.

The mix inside that book has turned toward the buyers whose financing just got dearer. Commercial and other industrial work rose from 29% of backlog at the end of March to 40% at the end of June, while oil and gas excluding petrochemicals fell from 29% to 24%. Pricing has not normalized either: gross margin widened through fiscal 2026, from 28.4% to 29.6% to 30.6%. Powell is spending against the constraint, adding 335,000 square feet at Jacintoport that management said could support well above $100m of incremental annualized revenue.

It was not a clean quarter. Earnings of $1.42 a share missed the $1.49 consensus on revenue of $311.7m against $316.7m expected. But revenue growth accelerated across the year to 8.9%, and the shares now cost 24.7x forward earnings against 32.4x trailing, and against 49.2x trailing in May, on consensus fiscal 2027 earnings of $6.88 a share, up 27.8%.

Lead times that have not shortened

The supply relief that would end scarcity pricing is not in the quotes. In 2026, 15kV medium-voltage switchgear still quotes 52 to 80 weeks, 38kV equipment reaches 104 weeks, and power and substation transformers run beyond 160 weeks, two to four times pre-pandemic norms. New plants are years out: Hitachi Energy's $528m Mississippi transformer works is scheduled to start production in 2029. Eaton's June-quarter backlog was up 33% year on year with an Americas book-to-bill of 1.3.

Why Hubbell sat still

Hubbell's June quarter gives the mechanism its test. Utility Solutions sales rose 10% to $1,026m on 5.5% organic growth, while Electrical Solutions rose 25% to $686m on 18.3% organic growth led by data centers, where sales rose about 65%. On those figures, derived, utility products were roughly 60% of group sales, and utility capex is recovered through rates rather than raised in credit markets. The faster leg is the one giving back margin: Electrical Solutions' adjusted operating margin fell to 21.2% from 22.5%.

"Strong first half order trends provide visibility to our second half outlook," chairman and chief executive Gerben Bakker said on 28 July, as Hubbell raised full-year guidance to 16-18% sales growth and $20.25 to $20.55 of adjusted earnings per share. Reported net income still fell 1.6%, the interest cost of a $3.0bn acquisition closed in June rather than a demand problem. At 22.5x forward earnings against 27.2x trailing, Hubbell is cheaper than Powell on the forward line but carries roughly 12% consensus earnings growth to Powell's 27.8%.

So the business results account for none of October on the order line, and some of it on the valuation line: Powell's miss was real and its May multiple was extreme. What the month priced is the cost of capital behind a backlog that books work into fiscal 2028 and is increasingly owed by borrowers. Hubbell's position is the inverse, with slower growth bought at a steadier source of funding.

Hubbell reports on 27 October, the nearer test. Powell's next order number does not land until 8 December, which leaves seven weeks in which the only new information about the switchgear shortage will come from the bond market.

Equity Lifestyle Properties, the Chicago landlord that rents land to owners of manufactured homes and recreational vehicles across about 423 properties, told investors on October 21 what it intends to charge its manufactured-housing residents next year: 5.1% more. That is the identical figure the company put out a year earlier, and it sits below the 5.6% rate growth it actually realized in the third quarter partly because the preliminary number deliberately excludes occupancy gains and the rent reset on new residents. The shares fell 4% on the day.

The year has been read as the moment the most defensive niche in residential real estate lost its pricing engine: resident affordability cracks, lot-rent escalations decelerate toward the consumer price index, Sunbelt insurance and property taxes widen the expense line. Each leg is testable from disclosure, and the first two fail on this quarter's numbers.

The expense line narrowed instead of widening

Core property operating revenues rose 4.4% in the third quarter while core operating expenses excluding property management rose 1.3%, producing core income from property operations up 6.7%, and 6% for the year to date. Management put that expense growth roughly 135 basis points below inflation. Florida's annual tax notices arrived with smaller real-estate tax increases than budgeted, enough for the company to take $600,000 out of its full-year tax assumption. Normalized funds from operations came in at $0.79 a share, up 4.6%, and full-year guidance was left unchanged at $3.15 to $3.21.

Pricing here is largely inherited from lease structure. Roughly half the book is market-based, a quarter is linked to inflation or rent-controlled, and a quarter sits under long-term Florida agreements with escalation floors, which is why the reported rate does not track the consumer price index in either direction.

Where residents do turn over, the reset is large. "What we're seeing on mark-to-markets for new residents coming in is a 12% increase," Patrick Waite, president and chief operating officer, said on the October 21 call. Pressed by an analyst on why the 2027 rate matched last year's despite higher inflation and cost-of-living adjustments, chief executive Marguerite Nader pointed to better resident attrition than the prior year and said she expects that to continue.

The part that is genuinely shrinking is the campground

Seasonal RV revenue fell 4.4% and transient RV revenue 5.8% in the quarter, with fourth-quarter RV and marina rent growth guided near 2.8%. Sun Communities, the Michigan owner of about 455 manufactured-housing and RV properties, shows the same split more starkly: its 2026 guidance calls for manufactured-housing same-property net operating income growth of 5.9% against 0.9% for RV, and its June-quarter transient RV revenue fell $2.9m, or 4.8%. Equity Lifestyle is converting that traffic into annual leases: 14% of annual manufactured-housing sales this year came from seasonal and transient guests, at roughly $8,000 of annual revenue each.

UMH Properties owns no RV sites at all, which is why the three cannot be one story. The Northeast and Midwest operator, which supplies its own homes to entry-level renters, grew same-store rental income 8.7% in the third quarter and gross home sales 28%, to $12.8m, a quarterly record, with same-property occupancy up 110 basis points to 89.4% in the June quarter on 5% site-rent increases. The affordability squeeze that was supposed to break the model is the thing filling its pads.

What repriced was the discount rate

The ten-year Treasury yield reached 5.13% on October 21, against 4.02% a year earlier, and the FTSE Nareit All Equity REITs index fell 5.7% in September against a 0.6% decline for the Russell 1000. That gap points at the discount rate, because property-level cash flow at these landlords went up. Equity Lifestyle now trades at about 18.7 times its unchanged $3.18 guidance midpoint, against roughly 20.7 times a year ago on a smaller funds-from-operations base; the multiple gave up about a tenth while per-share cash earnings grew 4.6%. Sun Communities sits near 16.8 times its $6.87 to $7.07 core guidance and UMH near 15.1 times its $0.98 to $1.04 range, the cheapest of the three.

So the stall is a valuation event with one operating qualification. Lot rents at all three are compounding between 5% and 6%, expenses are growing slower than rents at the two that disclose the split, and the single decelerating line — nights sold to travelers who have signed no annual lease — is the one with no lease protecting it. At Sun Communities the picture is genuinely mixed: the company wrote $1.1bn off its UK business before selling it for about $1.03bn in September and has bought back roughly $425m of its own stock, so the revenue base being valued today is a different one from last year's. Part of the de-rating is rational re-anchoring, too, since ten-year mortgage terms the company described run 6.1% to 6.6% and the acquisition market Equity Lifestyle called muted is repricing to match.

The cleanest remaining threat to the escalator is legislative, not economic. New Jersey and Washington both capped lot-rent increases last year, and a Pennsylvania bill introduced in September would limit mobile-home fee increases. Those are among the states where UMH, the one landlord here whose rents and occupancy are both still climbing, does most of its business. Affordability did not break the annual rent increase this year; a statehouse is what can.

Kodiak Gas Services, which leases large-horsepower compressors to Permian producers and the midstream companies that gather their gas, rented that equipment at the best rates in its history last quarter and accelerated its revenue growth while doing it. Its shares have lost roughly a quarter of their value since July.

The fall has been read as the compression tightness cycle topping out — new large-horsepower units finally arriving in volume, capping the monthly rate per horsepower that is a contract compressor's entire earnings engine. Every input that thesis needs moved the wrong way for it. Rates rose, utilisation held near full, and the engines that would have to arrive to break the squeeze are ordered years out. What actually changed in October was the delivery schedule on Kodiak's second business: about 395MW of distributed gas generation, two-thirds of it contracted to data centers.

The rate card went up, not sideways

Kodiak's average monthly rate reached $23.80 per revenue-generating horsepower, up 4.5%, on 98.2% fleet utilisation and record quarterly adjusted earnings before interest, tax, depreciation and amortisation of $216.8m. Second-quarter revenue of $391.1m was 21.1% higher year on year, against 4.9% growth in the first quarter. "Strong demand for large horsepower compression continued to drive improvements in pricing and utilization," chief executive Mickey McKee said on the second-quarter call in August. New-unit horsepower is already fully contracted for 2026 and into 2027.

The supply side is worse, not better, for anyone hoping to add fleet: packaged gas-engine lead times have stretched from a historical 40-to-60 weeks out to 200 weeks as generator buyers compete for the same blocks. "New large horsepower lead times remain extended, and that reality is driving customers to make compression decisions further out than they historically have," USA Compression chief executive Micah Green told Gas Compression Magazine on October 12. The demand behind it is also still building: Permian dry gas supply passed 26 Bcf/d for the first time on September 6, and the Energy Information Administration projects 29.6 Bcf/d in 2027.

The markdown was aimed at the megawatts

In September Kodiak signed a six-year agreement to supply roughly 76MW of baseload power behind the meter at a West Texas data center using about 40 reciprocating gas units — its second long-term primary-power contract, taking about half the power portfolio under long-term contract. That is the part analysts repriced. On October 16 RBC Capital's Elvira Scotto cut her target to $82 from $88 while keeping her Outperform rating, citing "a reduced Power Infrastructure EBITDA multiple due to equipment deployment timing" with long-term estimates unchanged; Mizuho went to $72 on a more back-weighted deployment schedule.

The company Kodiak fell with says the same thing. Over three months its 25.6% decline sat beside Vertiv's 27.1% and GE Vernova's 11.0%, while Schlumberger rose 10.0% and Baker Hughes 2.1%. Mid-October brought a financing shock to the AI buildout, with hotter inflation data pushing AI-linked issuers to delay fundraising — which reprices anyone whose growth depends on data-center capital spending. Kodiak now trades at 8.13x trailing EV/EBITDA, against roughly 9.8x for both large lessors in late August, and at 24.7x forward earnings versus 55.7x trailing.

Archrock is where the business genuinely weakened

One name on this shelf earns its de-rating. Archrock, the largest US contract compression provider, has seen revenue fall sequentially for three straight quarters, with second-quarter revenue down 3.1% year on year and operating income down 53.7%. The lease book is fine — contract operations margin held at 71%, a seventh consecutive quarter above 70%. The damage is in aftermarket services, the parts-and-overhaul book, where revenue fell 35% to $42.0m from $64.8m. Consensus 2026 earnings of $1.72 a share sit below the $1.84 Archrock earned in 2025, which is why its 17.3x forward multiple is above its 16.0x trailing. Chief executive Brad Childers still called the market "structurally tight," and guided $1.4bn to $1.6bn of growth capital for 2027 through 2030.

So the shelf splits on cause rather than on sentiment. Kodiak, Natural Gas Services and USA Compression all grew revenue at double digits or better into the drawdown — USA Compression's average fleet rose to 4.45m revenue-generating horsepower from 3.55m, with distributable cash flow covering its distribution 1.65x and leverage at 3.72x, and it fell only 4.9%. The gap between that and Kodiak's quarter is not operating: it is the share of enterprise value sitting in power megawatts whose in-service dates moved. Archrock's is operating, and concentrated in one line. Front-month gas slipping below $3/MMBtu on October 20 thins sentiment across all four, though compression revenue is contracted and rate-based.

Archrock reports third-quarter results after the close on October 27 against consensus of $385.95m, essentially flat on a year ago; Kodiak follows on November 3. Those prints will say whether the rate per horsepower is still climbing. They will say nothing about when a gas turbine arrives at a West Texas data center, which is the question the market has actually been answering.

Manhattan landlords spent 2026 clawing back months of free rent, and on the brokers' arithmetic the cash value of a new lease rose faster than its asking price: average net effective rent on new direct leases reached $99 a square foot on a 12-month moving average in September, up 28% year on year, against base rents up 23% to $120. The saving did not stay with the landlord. Tenant-improvement allowances rose 7.9% over the same year to a record $120.80 a square foot, up 59.2% since 2019, because construction costs and tenant fit-out standards rose.

That swap is the whole test of whether the Manhattan recovery is arriving in cash. Net effective rent deducts forgone rent from face rent; it does not deduct the check a landlord writes to a tenant's contractor. The summer advance in SL Green, Vornado and BXP was read as proof that the economics of a Manhattan lease had turned. They have turned, and the measure most often quoted to prove it captures only the half that improved.

The concession changed shape rather than shrinking

Free rent is down on every broker's count, from a different base each time. Avison Young has it falling 8.6% in 2026 to just over 10 months per lease; Colliers' weighted abatement for new deals and expansions ran 12.4 months year to date against 13.8 months for full-year 2025, the lowest since 2019; a third dataset puts third-quarter signings at 8.2 months with allowances of $85.20 a foot. The samples differ, so the levels do not reconcile; the direction does. Meanwhile the average work-value package crested above $90 a square foot in the first half, the highest ever recorded, and accelerating.

Space that is not coming back

The landlord's leverage is physical. Avison Young measures availability down 290 basis points over the year to 13.4%, the lowest since 2020, with new leases and expansions at 78.2% of third-quarter activity; Colliers, counting inventory differently, reads 12.4% and a tenth straight quarter of tightening, the longest run since 2007. Leasing of 32.86m square feet in the nine months was the strongest start to a year since 2000. And roughly 8.8m square feet has been permanently removed by residential conversion since early 2021, with another 9.5m square feet of conversion starts planned for 2026.

SL Green reached its occupancy target a quarter early

"A growing scarcity of premier space in desirable Midtown districts has turned the tables in our favor," chairman and chief executive Marc Holliday told investors on the second-quarter call. The 21 October print backs him: Manhattan's self-described premier office landlord posted same-store occupancy of 95.0% including signed-but-uncommenced leases, against 92.9% at the end of 2025, same-store cash net operating income up 5.0%, and funds from operations of $1.59 a share. Its 28 replacement leases were signed at $110.88 a foot, 13.6% above prior fully escalated rents, a step down from 15.7% year to date.

The guidance raise, to $6.50-$6.70 from $5.60-$5.90, is mostly not operating: it carries $1.10 a share from selling a 20% stake in the food and beverage business and absorbs $0.25 a share of higher interest cost. At $47.40 that is 7.2x the midpoint, or 8.6x without the one-off, against 9.3x forward funds from operations for the office REIT sector, and 1.03x book, with trailing free cash flow at minus 1.1% because development spending outruns operating cash.

Vornado's cash spread is thinner than its headline one

"We had another strong quarter with comparable FFO of $0.67, beating analyst consensus by $0.10 or 17.5%," chairman and chief executive Steven Roth said on Vornado's 4 August call, up from $0.56, on rent commencements at Penn 1 and Penn 2. New York occupancy was 92.2%, from an 84.4% trough in early 2025. But the 659,000 square feet it signed in the first half at $105 a foot carried a 9.5% mark-to-market on an accounting basis and 7.1% in cash, and revenue grew 4.7%. Annualizing $0.67, a crude derivation since the company publishes no full-year guidance, puts it near 12.4x, the dearest of the three on cash earnings, at 1.06x book. On 21 October it closed the 350 Park Avenue venture with Ken Griffin and Rudin, a 1.05m square foot Citadel anchor lease and a $3.3bn construction loan, leaving roughly 850,000 square feet of upper floors to let.

BXP, the largest listed owner of premier Class A office space in the US, is a contrast case here: New York is a minority of six markets, occupancy was 88.4% with 91.3% leased, revenue grew 3.1%, and guidance rose $0.05 to $6.99-$7.05. It trades at 8.4x that midpoint and 1.81x book, the richest of the three on book value. Its mechanism is the cost of debt: $700m of 6.050% notes due 2036 redeemed $1.0bn of 2.750% paper, more than three points more for the same money.

Higher yields reach these landlords through their coupons

All three peaked around 21 July and have given back part of a spring advance: SL Green down 9.8% over 30 days, BXP 7.7%, Vornado 5.9%, each still higher than in April. The operating meters do not explain that; occupancy, cash net operating income and comparable earnings all improved. Rates do explain some of it, and not only through the discount rate. The 10-year Treasury yield stood at 5.13% on 21 October after touching a level last seen in 2002, and it arrives as cash cost: SL Green's $0.25 a share, BXP's extra points, and SL Green and Mori Trust's $2.0bn refinancing of 245 Park at a 6.52% fixed coupon.

The remainder is a measurement gap. Rent is genuinely being won back in Midtown, and a record share of what the landlord gives up now sits in construction allowances that the quoted rent figure never touches, and that BXP's own management expects to push its leasing capital spending toward $500m while cash same-store income lags reported income.

Two-thirds of the quarter is still unreported: BXP on 27 October, Vornado on 2 November. The figure to find in them is not the rent per foot but what it cost to build the tenant who pays it.

The architectural change that optical-module makers have been warned would eventually strand them is now in production, and the first design to ship keeps the laser as a separately purchased part. Broadcom's Tomahawk 6 "Davisson" switch packages the optics alongside the switch silicon, cutting power per 1.6-terabit link from roughly 30W to about 9W and the laser count about fourfold across its 102.4 terabits per second of capacity. It also draws its light from pluggable external laser modules, and the two conventional pluggable-optics versions of the same switch enter production a quarter ahead of it.

That matters because October's selloff in optics has been read as the market starting to discount exactly this migration, with merchant module and component makers on the losing side. The record does not support the split. The companies that would collect the benefit of optics moving inside the box fell as hard as the ones it would displace, and the datable events in the window were about the cost of money.

The socket does not disappear, it gets dearer

Lumentum, which sells laser chips, transceivers and optical subsystems to telecom and data-center customers, booked its first external-laser-source module order in the fiscal fourth quarter. "Yes, we've got our first ELS module order right? Which from an ASP standpoint is meaningfully higher than the set of lasers that we ship in," chief executive Michael Hurlston told investors on August 11. On the same call he said lead co-packaged customers' production plans "remain very much on track, and their demand signal has increased since our last update." His one complaint was supply: "We are way behind. Our shipments, unfortunately, on high-powered lasers."

Coherent, the merged II-VI business selling lasers, transceivers and indium phosphide chips, said the same thing from the other end of the product line. "We've seen absolutely no pushout of CPO demand. In fact, it's been the opposite," chief executive Jim Anderson said on August 12, adding that data-center transceiver demand "is absorbing every bit of capacity that we have and then some" and that indium phosphide capacity remains the primary constraint. Scale-up co-packaged revenue, he said, arrives in the second half of calendar 2027.

The fall sorted by prior gains

Measured over 30 trading days, the numbers flatter the substitution story: Coherent is down 26.8% and Applied Optoelectronics 27.7%, against 11.1% at Arista. Reset the window to the actual drawdown and it collapses. From Arista's record close on October 9 through October 16, Ciena fell 26.0%, Coherent 25.5%, Lumentum 25.3%, Fabrinet 25.2% and Arista itself 18.3% — a record set days after it launched an open Ethernet rack-scale portfolio with AMD, Broadcom, Meta and Microsoft, with no miss and no guidance cut in between. Only Broadcom, at 9.1%, was materially insulated.

The dated catalysts were financing events. Broadcom chief executive Hock Tan told a Singapore conference on October 12 that the company is working through wafer and memory shortages but that "capital will be the biggest constraint," likening some investor behavior to "casino gambling" — while describing compute demand as exceeding supply into 2029. Days later, a strong US inflation report revived fears that higher rates would make AI buildouts costlier to fund, and the semiconductor index fell for eight straight sessions, its longest streak since 2008.

What the businesses did instead

Lumentum's revenue growth accelerated through four consecutive quarters, from 58% year on year to 109% in the June quarter, with gross margin widening from 34.0% to 47.4% and fiscal 2026 revenue of $3.01bn. At 39.2x forward earnings the shares are anchored to consensus revenue of $6.29bn next year, a doubling: the risk lives in the estimate rather than in the reported margin. Coherent's growth accelerated to 33.7% with gross margin up every quarter to 38.5%, and its forward multiple of 25.2x compares with roughly 35x in May. Its strain is cash — June-quarter capital spending of $555.7m ran 88% above consensus as the capacity build consumed the margin gain.

Applied Optoelectronics is the one name where the de-rating is earned, and not by architecture. Revenue rose 86% to $191.9m in the June quarter while gross margin fell from 31.2% to 27.7% over three quarters and the operating loss widened to $24.7m: negative operating leverage at the point in the cycle when scale should help. Its share problem is Chinese competition, with InnoLight at about 34% of datacom optical component revenue against Applied Optoelectronics' $81m quarterly datacom components. The company sold 5.7m shares at an average $105.36 in early October and has traded below that price since; at 104.5x forward earnings it is still the dearest of the four, even after its price-to-gross-profit ratio fell to 38.1x from 104.4x in May. Fabrinet, which assembles both pluggable modules and co-packaged units, grew revenue 44.6% on a 12.3% gross margin and trades at 19.6x forward earnings, the cheapest here — the market pricing an assembler, not an owner of components.

So the month explains itself without the technology. Three of the four protagonists reported accelerating revenue, widening margins and order books their managements described as booked through fiscal 2027; one reported accelerating revenue with eroding margins and is down 65% from its high. The first group's fall is multiple compression on rising estimates after a year in which Lumentum gained 424%; the second's is a margin verdict. A policy question cuts the other way entirely: the Federal Communications Commission has proposed barring new Chinese optical modules, whose two largest suppliers hold about 38.7% of the market between them.

The substitution does arrive, on a 2027 timetable and with its own bill of materials. What it has not yet done is remove anybody's socket — and the one supplier whose economics are already deteriorating is losing them to competitors building the same pluggable modules it does.