DK Street Journal

Digital Realty signed more new rent in the September quarter than in any quarter of its existence, and told investors on 29 October that the capacity it currently has under construction is underwritten to stabilize at a yield of 11.6%. The most recent debt it raised to pay for that construction, $1.7bn, carried a blended coupon of 4%.

That gap is the whole argument. Data center landlords have been marked down since the summer on a mechanical story: long rates rose, so the spread between the yield on a newly built megawatt and the cost of capital funding it must be closing, and hyperscale tenants would rather self-build anyway. The quarters reported on 28 and 29 October say the spread went the other way. A prior reading of Digital Realty's book had an 11.5% stabilized yield on a 1.4-gigawatt pipeline against a 5.28% 30-year Treasury; the September quarter shows 11.6% on 1.5 gigawatts, with the long bond at 5.31% on 30 October after touching a 24-year high of 5.67% on 7 October.

$14.7m buys a megawatt; the stabilized rent on it is about $1.7m a year

The arithmetic is reproducible from the company's own disclosure. Digital Realty, which owns 309 data centers and about 3.0 gigawatts of installed capacity leased to cloud, financial and manufacturing tenants, has 1.5 gigawatts under construction at an expected cost of $22bn: roughly $14.7m per megawatt. At the stated 11.6% stabilized yield, each leased megawatt should throw off about $1.70m of annual net operating income. That capacity is 63% pre-leased before it opens.

The in-place book got better on every line. Cash re-leasing spreads on renewals ran about 12%, management raised full-year renewal-spread guidance by a point to 10-12%, churn fell to a record-low 0.7%, and stabilized occupancy rose 90 basis points to 93.4%. Bookings will generate $855m of annualized base rent at 100% share, $627m at Digital Realty's share, leaving a record $2.5bn backlog of signed-but-not-commenced rent, roughly 40% of in-place rent. "Total leasing for the quarter was a blowout at $855 million or $627 million at our share, with new records set across several metrics," investor relations head Jordan Sadler said on the 29 October call.

The dilution objection fails on its own numbers. Diluted shares rose 8.8% year on year, and core funds from operations per share still grew 11%, to $2.10 — implying aggregate core FFO grew about a fifth. Full-year guidance went to $8.30-$8.35, near 13% growth. Against $19.5bn of debt, net leverage is 5.2x adjusted EBITDA.

Equinix re-lets a vacated cabinet at 7 kilowatts

Equinix, which sells colocation and interconnection inside its own campuses to more than 3,200 customers, runs a different engine on the same spread. Cabinet utilization was flat at about 78% year over year, so growth came from price and power: customers churned out at 5 kilowatts per cabinet and in at 7, with churn held at 1.7%. "Pricing on new deals was up 12% year-over-year in the quarter," chief executive Adaire Fox-Martin said on 28 October. Adjusted EBITDA margin widened 180 basis points to 51.4%, AFFO per share grew 12% on just 1.1% more shares, and guidance was raised for a third consecutive quarter. Stabilized assets yield 27% cash-on-cash against a 3.2% blended borrowing cost. Most hyperscale capital sits outside the balance sheet in a $15bn joint venture in which Equinix holds only 25% of the equity.

The segment's 20% decline is one pre-revenue developer

The four-name data center group is down about 20% over twelve months on an equal-weighted basis. Strip out Fermi and the rest average roughly flat. Fermi has reported exactly zero revenue in every quarter of its existence, lost $177.8m in 2025, and fell after its prospective anchor tenant terminated a $150m construction-funding agreement and a federal subpoena landed; its shares are down 89%. Blackstone Digital Infrastructure Trust, which listed 87.5m shares at $20.00 in May as a blind pool with no assets acquired, trades at $15.75 on a $1.59bn market value and a single reported quarter of $9.3m of revenue. Neither collects rent on a mature megawatt.

Where the de-rating actually came from

Digital Realty is flat over twelve months on revenue up 20.6% and operating margin 660 basis points wider, and sits 11.4% below its 52-week high. Equinix is up 20.8% while its trailing earnings multiple compressed from about 74x in May to 60.6x, meaning earnings, not re-rating, paid for the move. Nothing in either leasing book explains the first case.

The residual causes are the discount rate and a wholesale repricing of AI risk. More than $2trn of Nasdaq-100 value evaporated in October. Tenant budgets stopped surprising upward: Microsoft cut its fiscal-year capex forecast to $170bn from $175bn and Alphabet's third-quarter spend came in at $51.36bn against $55.09bn expected, even as combined guidance across the four largest hyperscalers reached $720-745bn for 2026. Funding did tighten, but selectively: CoreWeave's debt slid toward distressed prices and several private data center listings were pulled or paused. Over the same 30 sessions, tower landlord American Tower rose 16.6%, which is hard to square with a sector-wide REIT funding squeeze.

On its raised guidance, Digital Realty trades at 21.7x core FFO, below the 22-25x it carried in May when management was guiding FFO growth of 8-9% rather than 13%. Equinix at 26.6x trailing EV/EBITDA remains dear against any property-sector comparison; Digital Realty's 24.4x is not cheap either.

The open question is conversion, and both companies named it themselves: power availability, grid capacity, water, community sentiment and permitting. A $2.5bn backlog is only worth 11.6% once the substation is energized.

Celestica told investors on 27 October that it expects about $40bn of revenue in 2027, nearly double the roughly $21bn it now guides for this year, and in the same breath that the year would be a net use of cash with positive free cash flow not returning until 2028. The shares rose 12.1% that day and a further 8.4% the next, then gave the entire move back on 29 and 30 October.

The re-rating that took Celestica from single-digit earnings multiples rested on a belief about mix: that designing hyperscale racks pays better than screwing them together. The September quarter is the first clean test of that belief at scale, and it reads three ways across the shelf — compressing at Celestica, expanding at Flex, and barely capital-intensive at all at Jabil.

The quarter that grew 83% and gave up gross margin

Celestica, the Toronto contract manufacturer that designs and assembles switches, data-center interconnects, servers and storage for hyperscale cloud buyers, reported revenue of $5,829.7m, up 82.5%. Gross profit grew 54%, so gross margin fell to 10.99% from 13.03% a year earlier on the reported series; management's own adjusted figure, on a different basis, was 11.2%, down 50 basis points on mix.

The composition explains the gap. Connectivity & Cloud Solutions reached $4.9bn, up 102%, with Compute and Storage up 242%. The design-led Hardware Platform Solutions line reached $2.3bn, up 70% — slower than the group — which on its own disclosed figures takes HPS from 42% of revenue in the March quarter to 39.5% in September. The pure-assembly work accelerated; the design work did not keep up.

None of which is a market-share problem. Celestica puts itself at 31% of revenue and 44% of port shipments in 800G-plus switching and 58% of custom Ethernet switches, with ten 1.6-terabit programs won and research spending now above $300m a year, three times its 2023 level. Its three largest customers were 35%, 17% and 14% of September-quarter revenue, and the top ten rose to 85% from 80%.

The bill arrives a year before the revenue

Free-cash-flow guidance for 2026 came down to $460m from $600m to fund working capital for next year's growth. The 2027 capital budget of $1.5–2.0bn is 3.75% to 5.0% of the guided revenue it supports, and Celestica raised $3.45bn of equity during the quarter. Its trailing free-cash-flow yield of 1.64% is the lowest of the six names on this shelf.

"Our $40 billion guide took into consideration the risks around the ramps, the component availability, and the operational execution, David," chief financial officer Todd Ankenmann told analysts on 27 October. "At 90% year-to-year growth, we feel high confidence that that number is supported." The confidence is in the revenue. The cash timing is the disclosure.

Component prices are part of what has to be financed: the contract price of a 64GB DDR5 server memory module reached roughly $1,500 from about $272 a year earlier, and suppliers do not forecast relief before 2028.

Flex sold the part of the rack that pays

Flex, which builds switchgear, busways and power distribution for data centers alongside its broader manufacturing book, reported its Cloud, Power & Infrastructure segment up 54% to $2.5bn with operating margin up 240 basis points to 10.7%. "Power segment operating margin increased to approximately 16%, up roughly 550-basis points year over year driven by strong volume growth and the benefits of scaling the investments we made last year," chief financial officer Kevin S. Krumm said on 28 October. Group gross margin rose to 9.56% from 9.02%, a fourth consecutive annual improvement from 7.06% in fiscal 2024. Inventory rose 39%, pushing net working capital to 60 days. The segment is being separated as Axiom Solutions International, targeted for the first quarter of 2027.

Jabil's version is the capital-light one. Intelligent Infrastructure grew 56% to $5.8bn, net capital spending ran at 1.3% of fiscal 2026 revenue — $470m on $36bn — free-cash-flow conversion exceeded 100%, and core return on invested capital is 59%. Its shares still fell about 10% on 30 September on a beat and a raise.

What the selling was actually about

The margin split does not explain the drawdowns, because the name with no AI-rack exposure fell too. Plexus — semiconductor capital equipment, aerospace, defense and medical — dropped 20.3% over the 30 days to 30 October after a record quarter of $1.43bn, up 35%, a 6.2% full-year operating margin and the best cash cycle in its history at 62 days. Sanmina now owns AMD's former rack-assembly plants and Benchmark's advanced computing line grew 50%, so neither is a control on anything.

What moved was the willingness to pay, not the order book. Hyperscale capital budgets for 2026 stand at $720–745bn, and the dated trigger was financing discipline rather than demand: Microsoft trimmed its figure to $170bn and Google declined to raise its forecast in late October, before a session that erased more than $1trn of technology market value.

So the multiples did the moving while the businesses accelerated. Celestica now trades at 24.6x trailing earnings and about 12.1x the $21.98 fiscal 2027 consensus, against the roughly 30x forward multiple of mid-May. Measured against trailing gross profit, the cheapest of the three is Jabil at 8.5x, down from 12.8x in May. The reported results account for none of the decline; the question the market is now pricing is who funds the next rack, and Celestica has answered it in writing.

Sanmina reports on 9 November and Flex details both post-spin companies at an innovation day the following day. Celestica's own next disclosure is further out, which leaves its 2027 budget standing alone for now: a year of spending it has already promised, against revenue it will book before the cash shows up.

Curtiss-Wright sells coolant pumps, control-rod drive mechanisms and valves into the US Navy's nuclear fleet and spent-fuel hardware to commercial reactor operators, and in the June quarter its revenue, margins, orders and backlog all moved up together. The shares have since fallen 38% from their June close, a slide that was already running when the company named a successor to its chief executive on 22 September and that steepened on the day long-term Treasury yields hit their highest level since 2002.

The run in the naval-nuclear supply chain was read as a bet on shipped content: submarine reactor parts already contracted, commercial reactor hardware already specified. The marked-down half is the second one. Nothing in Curtiss-Wright's submarine business got worse this year, and the catalyst its valuation leaned on — a first firm order for commercial reactor components — has not arrived.

The submarine side got bigger

Naval & Power sales reached $410m in the June quarter, up 7% on submarine-programme production timing and shipyard aftermarket work, with management guiding the segment to 10–11% sales growth in the September quarter. In the March quarter the same segment earned a 14.9% operating margin, 230 basis points better than a year earlier. Group revenue rose 5.4% to $924.0m, operating margin widened 151 basis points to 19.34% and net income rose 24.9%. Backlog finished the half at a record $4.5bn, up 10% from the end of 2025, with Naval & Power booking 1.5 times what it shipped.

The money behind that is not in doubt either. The Defense Department awarded $76.6bn in July 2026 for nine Virginia- and five Columbia-class submarines, and the Navy's fiscal 2027 request puts $65.8bn into shipbuilding, up 48%.

The order that did not come

Curtiss-Wright carries roughly $120m of content per AP1000 reactor, and the Energy Department has made a conditional $17.5bn loan commitment for up to ten of them. "Long-established and significant commercial nuclear expertise dating back to the industry's inception" was how Bamford described the position on the August earnings call, telling investors launch customers "were overwhelmingly impressed with the quality of our critical manufacturing processes" and that she expected a first production order in 2026. No such award had been identified by mid-September.

The sell side then named the move for what it was. Truist cut the shares to Hold on the view that nuclear's long-term addressable market is "adequately reflected in current valuations and stock prices". On 7 October, the day the ten-year Treasury yield touched 5.286%, Citigroup pulled its target to $622 from $793 and the stock fell 8.0% to $509.10; Stifel went to $518 from $700 before a new 52-week low on 20 October. This is a repricing of distant cash flows, not a demand problem: NuScale and Oklo, which ship nothing, are down 87% and 81% over twelve months.

BWXT's compression is partly its own

BWX Technologies, the sole-source maker of naval reactors and fuel for the Navy's propulsion programme, entered its downtrend on 8 July and has fallen 29% since. Some of that it earned. June-quarter revenue grew 18.0% to $901.6m while operating margin compressed 344 basis points to 9.97%, the fourth consecutive quarter of lower operating income, because Commercial Operations grew 121% at a 12.9% adjusted-EBITDA margin against Government Operations' 20.9%, and the commercial margin guide was cut to about 13% from about 14% to fund capacity.

The order book did the opposite. Backlog reached $8.4bn, up 40%, on 1.7 times trailing book-to-bill, and free-cash-flow guidance went up by $30m to $345m–$360m. BWXT trades at 33.7x trailing earnings against 55.0x in early May, with a forward 27.6x.

Ducommun, which builds contoured fuselage panels, winglets and rotor blades and has no nuclear content, shows what an aerospace-structures business did instead: June-quarter revenue up 11.0% to $224.5m, operating income up 59.4%, the shares up 75% over twelve months.

What the multiple gave back

For Curtiss-Wright, the results account for none of the decline. Earnings rose, margins widened, backlog set a record, and the trailing multiple fell from 56.5x in May to 32.6x. Its forward 31.0x sits only 1.6 turns below trailing against consensus 2026 earnings of $15.36 a share, up 17% — the market has stopped paying for acceleration in a business whose naval revenue is contracted and whose commercial upside is a decade out. BWXT's 27.6x forward against 33.7x trailing still embeds a 24% earnings jump, which is the harder claim of the two, and its mix is working against margins while it funds capacity.

The split has a cause: Curtiss-Wright's nuclear content is already inside expanding segment margins, so only the valuation moved; BWXT's growth is arriving through a lower-margin commercial line it is still building out, so both the valuation and the earnings line moved.

BWXT reports third-quarter results after the close on Monday, with analysts looking for $1.19 a share. It will say a good deal about mix and nothing about the reactor order neither company has yet been given.

Enphase Energy, which makes module-level microinverters, alternating-current batteries and the Enlighten monitoring service and sells through solar distributors and installers, shipped 1.42m microinverters in the September quarter and reported $127.1m of gross profit on them. Its quarterly filing for the period ended 30 September shows $58.6m of that figure was the Section 45X advanced manufacturing production credit, which pays $0.11 for every watt of alternating-current capacity built in an American microinverter plant and arrives as a reduction of cost of revenues. Take it out and a shipped watt earned the company $0.105 of gross profit, against $0.164 in the same quarter of 2025.

The first three quarters of 2026 had been read as evidence that module-level electronics could hold price through the expiry of the homeowner solar credit: reported gross margin was 43.3%, and at SolarEdge it had risen for six consecutive quarters. At the shipped unit, with the manufacturing credit removed, product economics went the other way at both. Enphase's ex-credit gross margin fell to 23.4% from 31.3%. SolarEdge's filing says that without the same credit its gross profit would have been a loss.

Nine cents inside a nineteen-cent watt

Enphase's unit disclosure for the quarter, filed with its results of 27 October, shows 655.2 MW DC shipped against 784.6 MW a year earlier, and 129.9 MWh of batteries against a then-record 195.0 MWh. Revenue fell 28.5% to $293.3m and operating income 90.2% to $6.5m. Gross profit per microinverter works out at $48.24 once the credit is stripped, against $72.66 a year earlier. The entire $23.4m of reported net income came from a roughly $27.0m income tax benefit sitting on a pretax loss of about $3.6m.

All three variables are guided lower together. Fourth-quarter revenue is set at $225–255m including roughly $51m of safe-harbor revenue, batteries at 100–120 MWh, and non-GAAP gross margin at 40–43% from 45%. Global sell-through, flat near $230m a quarter all year, is expected to fall about 10% to around $210m, and Enphase intends to ship some $20m below even that to drain distributor stock. US battery prices were cut 8–9%.

"We are not seeing any meaningful recovery in the U.S. demand," chief executive Badrinarayanan Kothandaraman told investors on 27 October.

The person claiming the subsidy changed

Section 25D, the residential credit, expired for expenditures after 31 December 2025, taking 30% of system cost away from anyone paying cash or borrowing. What survives is the Section 48 commercial credit, claimable only by a third-party owner, and third-party ownership's share of US residential volume rose from 40% to 52% during 2024 with 60–65% projected on exit from this quarter. Borrowing got no easier: the Federal Open Market Committee held at 3.75–4.00% on 28 October with the average 30-year fixed mortgage at 7.33%, up from 7.12% a month earlier. Enphase's US residential sell-through fell about 35% year on year. Its Propel lease programme now runs in 13 states with more than 350 installers and a 71% battery attach rate, but the equipment revenue from those originations is expected in the first half of 2027.

One sentence in SolarEdge's filing

SolarEdge, the Herzliya-based maker of DC-optimized inverters and power optimizers, grew June-quarter revenue 19.6% to $346.2m and lifted GAAP gross margin to 27.5% from 11.1%, its first non-GAAP operating profit since mid-2023. The composition undercuts the direction. Its 10-Q states that excluding the same manufacturing incentives would have caused gross profit "to transition from a gross profit to a gross loss," and the 28.6% non-GAAP figure included $13.3m of tariff refunds. Europe, up 36% sequentially to $154.4m, drove the quarter; the US slipped 2%. Third-quarter guidance is $310–340m on 22–26% margin, and the results land on 4 November.

A premium that closed in one session

Enphase fell 16.0% on 28 October and 22.3% over the 30 days to 30 October; SolarEdge fell 8.3% that day on a rival's numbers. First Solar, Nextracker and Array Technologies all rose the same session, which argues against a sector-wide repricing.

The de-rating is doing most of the work the results ask of it. Enphase trades at 6.12x trailing gross profit, rising to 6.62x forward because the denominator is shrinking faster than the price, against 14.55x in May. SolarEdge sits at 6.26x against 16.09x then.

So the two now carry effectively the same price against gross profit, though Enphase's fell 35.2% year on year and SolarEdge's tripled. That is a judgment that neither company's margin is its own: both are priced off the durability of one federal credit rather than off the trajectory of the product. It fits the numbers better than it fits the narratives. Enphase still earns 23.4% at the gross line without help, on consensus revenue the Street sees falling a further 8.6% in 2027 to $1.02bn, which puts the stock near 18.5x the $1.39 of earnings expected that year. SolarEdge, on its own disclosure, does not clear zero without the credit at all. Oppenheimer's Colin Rusch, cutting Enphase to Perform, wrote that a recovery in the core business "is not materializing".

The credit that is holding both income statements up has a schedule. Section 45X pays in full through 2029, then steps down each year and pays nothing for product sold after 2032. Residential inverter makers have until then to make a watt worth shipping on its own.

Commercial bankruptcy filings in the United States ran 11% ahead of last year through September, and FTI Consulting, the advisory firm built to staff the largest of those cases, got less work out of each consultant than it did a year earlier. Utilization in its Corporate Finance & Restructuring segment fell to 58% from 63%, and the Restructuring practice inside that segment saw revenue decline 1%.

The 29 October print was read as the distress cycle finally reaching the invoices. FTI did cross $1bn of quarterly revenue for the first time, at $1.013bn, with adjusted EBITDA margin of 15.2% against 13.7%. But the hours behind that revenue tell a different story about where advisory money is coming from, and it is not the bankruptcy docket.

Why a filing count stopped predicting a fee

The year's insolvency growth is small-company insolvency. Subchapter V small-business elections were up 46% year on year as of October, and a small-business reorganization generates a fraction of the expert hours a large case does. Meanwhile the headline series has turned: Epiq counted 6,560 commercial Chapter 11 filings through September, but September alone produced 648 against 769 a year earlier, a 16% decrease. The wave crested in the first half, when filings ran 28% ahead.

The antitrust channel has the same shape. Merger notifications reached 2,555 in fiscal 2026, the largest total since 2022, but second requests — the investigations that pay for testifying economists — numbered 41 in fiscal 2025, about 2.1% of adjusted reported deals, and in July the Justice Department's Antitrust Division published a model timing agreement meant to compress those reviews. More deals, narrower investigations, fewer hours each.

What actually paid at FTI

Corporate Finance & Restructuring revenue rose 4.8% to $424.4m on a 3.2% higher average bill rate of $550, with Transactions up 13% on success fees and Transformation up 7%. Success fees hit a record $40.7m, including a Federal Reserve Bank of Boston loan-portfolio sale. That is event income, not a recurring run rate of billed hours.

The genuine repair is at Compass Lexecon, the Economic Consulting arm, whose dislocation management says had been costing roughly $22m of EBITDA a quarter for five quarters. Revenue there rose 9.3% to $189.3m and adjusted segment EBITDA swung to $14.0m from minus $4.6m, at a $639 average rate, the firm's richest hour. "That dislocation is behind us, and the power of everything else we have been doing is once again beginning to shine through," Gunby told investors on 29 October.

Forensic & Litigation, by contrast, was flat at $195.1m on 54% utilization, record rates offsetting lost volume as US enforcement pulled back. FTI took a $46m charge for cuts touching about 4% of staff, with roughly $80m of annualized savings expected. "We believe these actions further strengthen our business and position us for sustained growth," chief financial officer Angela Nam said the same day.

CRA re-rated before it reported

CRA International, the $1.23bn Boston economics and expert-testimony firm that competes with Compass Lexecon for antitrust mandates, rose 19.5% over October with nothing published since 6 August; it reports on 5 November. Its largest day, 22 October, came as Accenture rose 7.4%, Huron 6.5% and Exponent 5.2% on a session when the Nasdaq fell 2.6% — the likelier reading is rotation into de-rated labor businesses. Its last numbers were good: revenue up 12.8% to $210.8m at 77% utilization, with guidance raised to $805m–$820m on an unchanged 12.0%–13.0% margin target, and chief executive Paul Maleh describing "peak levels of new project opportunities" on 6 August.

Still, consensus has CRA earning $8.47 for the year ending January 2027, up 2.8%, against a forward multiple that went from 19.4x to 23.2x on no news. Its trailing free cash flow yield is minus 2.3% against FTI's 8.0%, and it is the dearer of the two on price to gross profit, 5.78x versus 3.89x.

FTI's results account for about an eighth of its 36% month: the forward multiple moved from 13.9x to 18.9x while the adjusted earnings guide rose 4.3% at the midpoint, to $9.50–$10.00. And the earnings that justified it are mostly not hours. Gunby normalized the quarter's $3.84 of adjusted earnings per share to roughly $2.90 after success fees and a tax rate near 22%, while buybacks took the diluted share count to 27.0m from 31.8m the prior quarter. The one durable gain is Compass Lexecon's return, which management frames as a five-year recovery to historical profitability.

What both firms are really defending is the named expert. FTI promoted 100 senior managing directors and hired more than 140 laterals over two years, which is why its rates can set records while its utilization falls. Paying up for experts works when the cases are big enough to keep them busy, and this year's cases are getting smaller.

GE Vernova booked $21.8bn of Power orders in the third quarter, up 178% organically, and signed fewer gigawatts of gas turbines doing it than it had three months before. The company, which sells gas, nuclear, steam and hydro generating plant to utilities, independent power producers and increasingly data-center developers, took 11 GW of new gas equipment contracts in the quarter — 8 GW of slot reservation agreements plus 3 GW of firm orders — according to its 28 October call, against roughly 19 GW of gross intake in the second quarter on the same basis. That is a derived comparison, built from contracted capacity and shipments rather than a disclosed figure.

Turbine slots sold out to the end of the decade have been read as evidence that the order line measures demand. This quarter it measured price: equipment pricing was up more than 20% year on year, and full-year 2026 pricing is running 50% above 2025. Wood Mackenzie puts gas turbine prices 195% above 2019 levels. The shipped economics improved sharply, which the selloff ignored. The reservation queue drained, which it may not have.

What five gigawatts earned

GE Vernova shipped 5 GW of gas equipment in the quarter, 32 turbines including 13 aeroderivatives, a 60% increase in units. Power revenue of $6.1bn rose 25%, and segment EBITDA margin reached 21.4% from 13.4% a year earlier on pricing and volume. Divide those disclosures by shipments and the quarter carried about $1.22bn of Power revenue and roughly $260m of Power EBITDA per gigawatt out the door — a derived upper bound, since the segment also contains services and non-gas lines.

The annuity behind the hardware is the reason pricing holds. "We expect every gigawatt of HAs to generate $0.5 billion of high-margin services revenue over the first 20 years of its lifetime," chief executive Scott Strazik told investors on 28 October, with the installed or contracted fleet of those advanced-class units heading to at least 400 by the end of 2027. Full-year Power guidance went up for a third time, to 19–21% revenue growth and a 19–21% segment margin.

The deposits land years before the turbine

Equipment remaining performance obligations stood at $104.1bn on 30 September, with 33% expected in revenue within a year and 60% within two. Contract liabilities and deferred income rose $19.8bn over nine months. A roughly 20% cash deposit converts a reservation into a slot agreement, which is how a heavy-equipment maker guides to $16–17bn of free cash flow on $46.5–47.5bn of revenue, about 35% of sales.

The queue got shorter while the book got bigger

Against 8 GW of new reservations, GE Vernova converted 13 GW of existing ones into firm orders. The pool therefore shrank by about 5 GW, and total contracted capacity rose only 6 GW, to 122 GW from 116 GW, against double-digit quarterly gains earlier in the year. Strazik said he expects solid order and reservation growth in 2027, but not at the pace of the first half of 2026. Management said it had seen no financing disruptions and no slowdown in converting reservations to backlog. In the same release the company agreed to buy high-voltage equipment maker Trench Group for $9bn, half cash and half debt, closing late 2027 with about $1.9bn of 2028 revenue attached.

Caterpillar ships faster; Bloom stopped counting megawatts

The demand being announced is also being shipped. Caterpillar, whose Power & Energy arm sells reciprocating engines, generator sets and industrial gas turbines alongside construction machines, grew that segment's revenue 18% to $8.78bn with profit up 31% and margin of 24.4%. Company backlog hit a record $78bn, 96% higher, and 58% of it converts within twelve months — a far quicker clock than GE Vernova's equipment book. Caterpillar's shares fell on about $2.2bn of full-year tariff cost and margin guided near the bottom of its range, leaving 22.9x forward earnings against the roughly 38x on next-twelve-month earnings that David Giroux flagged in June.

Bloom Energy, which sells solid-oxide fuel cell servers converting gas or hydrogen to on-site power without combustion, lifted product revenue 179% to $1.07bn at a 38.4% product gross margin, up 250 basis points. It no longer discloses megawatts accepted each quarter, so margin per megawatt cannot be computed from what it publishes. At 64.1x forward earnings, down from about 119x in May, it remains the most demanding name of the three. Wärtsilä, whose Energy arm builds reciprocating-engine plants, lifted order intake 121% to €1,421m on a record €9,876m book and its shares fell too.

How much the results explain

Very little. Over the 30 days to 30 October, GE Vernova fell 9.7%, Caterpillar 23.2% and Bloom 26.7% while the S&P 500 tracker slipped 0.9%; the worst session was 22 October, when Meta's capital-spending commentary hit everything levered to data-center construction and the Nasdaq lost 2.6%. GE Vernova's 27.2x forward earnings is the honest anchor, since its 22.3x trailing is flattered by a $4.5bn one-off gain on Prolec GE; in May the shares traded at 32.1x trailing earnings. The de-rating is wider than any disclosed deterioration, and it is global.

One business fact does point the bears' way, and it is the one Matthew Smith's July thesis predicted: gigawatt intake slowing while dollars accelerate. Mitsubishi Power said in late October that demand is pushing deliveries out to 2031, so capacity remains the constraint. But a book of orders priced 50% above last year's, converting a third inside a year, is worth less if the queue behind it stops lengthening.

Pricing power this strong is usually the last thing to break. The reservation queue, where deposits are taken years before a turbine ships, is the first — and it got shorter this quarter while every other number got better.

A year of one diabetic patient's business is worth very different money depending on which device maker is billing for it. Over Insulet's and Dexcom's own disclosed user counts and margins, a year of one Omnipod wearer throws off roughly $3,440 of gross profit against about $885 for a patient wearing a Dexcom sensor — arithmetic derived from the two companies' disclosures, not reported by either. The company at the top of that ladder is also the fastest-growing of the three and the cheapest on earnings.

October looked like the turn for diabetes devices. Averaged evenly, the four listed American makers of insulin pumps and glucose sensors gained 3.9% in the 30 days to 30 October. Weighted by company size they gained 0.6%, because Dexcom alone is $30.7bn of the group's $43.2bn and Dexcom fell 4.6%. The rises came from the pumps: Tandem Diabetes Care, which sells the tubed t:slim X2 with the Control-IQ algorithm, up 17.2%, and Insulet, maker of the tubeless Omnipod, up 14.7%. Neither move came from a patient count.

What a pod collects that a sensor cannot

Insulet passed 600,000 active Omnipod users at the end of 2025 and has grown that base 23%. Against full-year revenue guidance of about $3.25–3.30bn, that is roughly $4,900 of revenue per patient-year, essentially unchanged from about $4,980 in 2025. Dexcom exited 2025 with around 3.5 million active customers and guides to $5.20–5.25bn: roughly $1,375 a patient, down from about $1,480. Tandem sits between the two, on about $8,000 of sales across a four-year renewal cycle — a pump once, supplies each year — or roughly $2,000 annually, about $1,140 of gross profit at its second-quarter margin of 56.9%.

Dexcom adds patients and collects less from each

Dexcom's reported growth has decelerated three quarters running, from 15.0% to 13.1% to 11.2%, while the active base grew in the upper teens. Part of the cause is engineering: the G7 15 Day extends wear from roughly ten days, so fewer sensors ship per patient per year. That lifted non-GAAP gross margin to 65.1%, the highest since the fourth quarter of 2022, and lowered what each patient pays over twelve months.

The company still raised its outlook. "We are updating our revenue guidance to be in the range of $5.20 billion to $5.25 billion, representing growth of 12% to 13% for the year," chief operating and financial officer Jereme Sylvain told investors on 29 October, and the call described no pricing headwinds. That was eight days after Abbott, whose FreeStyle Libre is Dexcom's largest sensor rival, cut full-year diabetes-care growth guidance to low-to-mid single digits from high single digits, citing international pricing pressure and type-2 coverage that has not landed. Dexcom fell 8.7% over the two sessions after Abbott spoke and recovered most of it on its own numbers. At 30.5x forward earnings against 31.8x trailing, the shares price in no re-rating against 12–13% guided growth.

Which benefit pays the bill is now the variable

Tandem rebuilt pharmacy formulary coverage from zero to roughly 40% in a single off-cycle quarter and expects 40–50% for the year. Wells Fargo upgraded it to overweight on 26 October and raised its target to $27 from $21, calling the pharmacy model "transformational and de-risking faster than we expected"; the shares rose 8.4% that session. That is the whole of Tandem's month, on revenue growing 5.8% and a quarterly operating loss still at -$13.8m, priced at 1.24x forward sales with no forward earnings multiple because consensus has it losing $0.67 a share this year.

Beta Bionics, which sells the iLet automated insulin delivery system, says pharmacy reimbursement for its Mint patch pump — cleared by the Food and Drug Administration in September for a first-quarter 2027 launch into Omnipod's tubeless niche — runs about six times iLet's medical-benefit rate. Dexcom is travelling the other way: CVS Caremark and Aetna commercial formularies switch to Libre-exclusive coverage on 1 January 2027, business Dexcom puts at under 4% of US revenue, with those patients moving to the medical benefit at a 10–20% monthly copay.

Insulet's discount and its 90-day problem

Insulet is the cheapest of the three on earnings, at 23.0x forward against 27.9x trailing and roughly 50x the same consensus a year ago. What de-rated it was reported, not rumored: US Omnipod growth guidance cut to 17–19% from 20–22% on type-2 patients dropping off in the first 90 days after setup, after type-2 passed 40% of new American customers. International guidance went up, to 30–32% from 26–28%. "As CEO, I take accountability for our reduced U.S. outlook," chief executive Ashley McEvoy said on the second-quarter call, adding that "once Type 2 customers remain on Omnipod for the first 90 days, retention rates stabilize." The fixes are operational: more staff for insurance navigation, sales pay re-weighted toward early retention.

Meanwhile revenue grew 23.5% in that quarter, gross margin held at 70.2%, and the shares are down 52% over the year.

The patients are arriving; the pricing is the argument

None of October's gain is traceable to fresh patient data, and the one company that did report into it — Beta Bionics, with revenue up 30.3% and full-year guidance raised to $124–126m — fell 11.8% over the month, including 16.6% on 28 October after disclosing it had withdrawn a software clearance application at the FDA's request, pushing that launch to mid-2027. At about 6x forward sales with negative free cash flow, that is de-rating from a demanding base.

The demand side of the hypothesis holds up. Dexcom's own type-2 report, published around the European diabetes meeting, found 58% of type-2 patients taking GLP-1 drugs pair a sensor with them against 27% of type-2 non-insulin patients overall, and 90% of US clinicians named coverage as the obstacle. What splits the group is not patient count but what each patient is worth: a pod and its supplies bill like a prescription every month, a sensor bills less each year as it lasts longer, and a pump bills once every four years and waits.

Insulet reports before the market opens on 4 November. Its pods earn more from a patient-year than anything else in diabetes care; whether the type-2 patients starting on them this autumn are still wearing one in the spring is the number that decides what that is worth.

Walmart signed an Enterprise Master Automation Agreement with Symbotic on 29 October, committing roughly $6bn of spend over several years across 25 US regional distribution centers and selected Americas markets, plus $125m for next-generation racking and inventory technology. Symbotic, which sells an end-to-end automated warehousing system of robots and software to wholesalers and retailers, expects the commitments to add up to $6.92bn to its future backlog — about 31% on top of the $22.5bn it carried at the end of June. The shares rose 3.7% premarket on 30 October, then closed at $37.93, below the $38.97 close of the last session before the agreement was signed.

The month's selling across robotics names has been read as a verdict on robot demand. What a deployed Symbotic system actually earns its maker says the opposite.

Fifty-six systems running, and each one earns more

In the quarter ended 27 June, revenue reached $720.8m, up 21.7%, while gross profit grew 49.5% to $160.9m — more than twice the pace of sales. Gross margin went to 22.3% from 18.2%, and the company swung to a $32.9m operating profit from a $20.2m loss. It started 11 new system deployments, brought four more into operation to reach 56 operational of 77 in deployment, and recorded adjusted EBITDA of $95m against $45m a year earlier.

The layer that compounds behind the hardware is growing faster than the hardware. Systems revenue was $671m, up 20%; software revenue was $13m, up 57%; operation services $37m, up 49%. Founder and chief executive Rick Cohen told investors the margin move was not a structural step: "it just has, once again, to do with the mix of deployments," he said on the third-quarter call, "it really then comes down, when you're tracking it, just really comes down to the 11 deployments we are putting in the quarter, just what those pricing adjustments were." The multi-year line is steadier than that caveat implies — full-year gross margin was 18.8% in fiscal 2025 on $2.247bn of revenue, against 15.6% on $1.788bn the year before.

The shares fell from $77.26 a year ago to $37.93, a decline of 50.9%, with 9.0% of that in the past month, and the trend has been bearish since early August rather than breaking in one session. On trailing earnings the stock is at 396x, which measures nothing at a 1.6% net margin. Against gross profit, the measure that fits a hardware deployer, it is 29.6x forward against 40.2x trailing — still 92x forward earnings.

Teradyne's 31% fall is about chip testers

Teradyne carries the robotics label on this shelf and supplied roughly two-thirds of its monthly drawdown, falling 31.2% in October. Its Robotics segment — Universal Robots collaborative arms and MiR autonomous mobile robots — did $96m in the September quarter, up 28% and profitable for the first time, with the Gen7 cobot launched in the period. That is 7.3% of the $1.306bn the company billed. Semiconductor Test passed $1bn, up 77%, and total operating margin went to 30.8% from 18.9%.

The shares fell after a beat: adjusted earnings of $2.29 against $2.06 expected, spoiled by a December-quarter guide whose $1.30bn midpoint sits fractionally below the quarter just reported and gross margin guided down to 57.5–58.5%. Then came 30 October, when the semiconductor index slid 7.2% and Intel fell nearly 29%. Chief executive Gregory Smith framed robots as the next thing, not the current thing: "with the growth of physical AI, the capabilities of physical AI, we think that that is going to unlock significant growth in the robotics group." Management puts broad physical-AI deployment at 2028 and later. At 28.9x this year's consensus earnings and 22.2x next year's, against the 30–35x the stock carried in May, the de-rating arrived while estimates rose.

The one that never fell

Rockwell Automation, which sells control hardware, industrial software and lifecycle services to factory operators, is flat over the 30-day window and up 19.3% over the year. Its June quarter had organic sales up 10% and Software & Control revenue of $751m, up 19%, at a 34.8% segment margin against 31.6%, with full-year guidance raised. Software & Control book-to-bill came in at 0.97, the one number pointing the other way, and on the call management said "the book-to-bill is in good shape." No deferral of discrete-automation spending was flagged. At 40.8x trailing and 29.3x forward earnings, Rockwell is below the 45.9x it carried in early May.

So the drawdown is a repricing, and most of it is not about robots at all. Teradyne's fall traces to a flat sequential guide inside a sector that lost a fifth of its value, and its robot line is too small to move the share price either way. Rockwell has no drawdown to explain. That leaves Symbotic, where the gap is real and one-directional: deployments started, systems brought live, margin per system and recurring revenue all moved up through fiscal 2026 while the equity lost half its value, and the single largest customer has now underwritten the next several years of installs.

What the agreement does not fix is the thing that makes the stock fragile. Walmart is getting a board nomination recommendation and non-voting observers in the boardroom of the supplier it has just committed $6bn to, and Symbotic accepted certain restrictions on whom it sells to. The customer concentration investors call Symbotic's chief risk is now also the reason its order book is growing.

Getty Images filed a prepackaged Chapter 11 petition on 21 October to put into force a recapitalisation its lenders had already signed, and the plan it implements cancels all of the company's Class A common stock with no recovery for current holders. What is odd about the filing is the business sitting underneath it. In the quarter ended 30 June, Getty's revenue fell 2.5% to $229.1m, its gross margin improved almost five percentage points to 70.2%, and the company earned $32.4m of operating income.

The visual-content libraries have been priced for two years as the clearest casualties of image generators, and the late-October record splits them on exactly that question. Getty's accounts describe a licensing operation that still turns pictures into profit, carrying about $2.1bn of debt that consumed every dollar of it. Shutterstock's accounts describe the substitution itself.

What killed Getty sat below the operating line

For the full year 2025 Getty posted $231.5m of operating income, a 23.6% margin and up 28% on the prior year, and still reported a $206.1m net loss. The June quarter repeated the shape: positive operating income, a net loss of $86.3m, roughly $119m of distance between the two.

The restructuring support agreement is dated 12 October, backed by holders of all revolving loans and more than two-thirds of the term loans and notes. The Getty family and Koch Equity Development committed up to $533m of new equity and will own 100% of a privately held reorganised company; unsecured bondholders agreed to retire $270m of notes. "Today's filing is a procedural step to implement the financial restructuring previously announced," chief executive Craig Peters said on 21 October. "This process addresses our capital structure and certain non-operating liabilities."

The listing had already ended. The New York Stock Exchange notified Getty on 29 September that it would begin delisting proceedings for an abnormally low selling price and suspended trading immediately; the shares moved to the over-the-counter Pink market the next day. On 30 October the stub fell 36.7% to $0.019, leaving $8.3m of equity value against $981.3m of 2025 revenue. No Getty disclosure carries that date, and the likelier reading is continued liquidation of an interest the plan has already written to zero.

Getty's catalogue is weakening, but in customers rather than usage: 240,000 active annual subscribers at 30 June against 321,000 a year earlier, while paid downloads over the same twelve months fell only 2.8% to 90m and subscriber revenue retention slipped to 88.4% from 93.4%. "Our second quarter results reflected continued pressure in Agency and iStock e-commerce, while the larger parts of our business serving enterprise customers continued to demonstrate resilience and growth," Peters said on 10 August.

Charging model developers through the courts is not the alternative. On 4 November 2025 the UK High Court rejected Getty's secondary copyright claim against Stability AI, holding that model weights are not a copy of the training images because the model does not store them.

Shutterstock's downloads are the substitution case

Shutterstock's June-quarter revenue fell 16.9% to $221.8m, the third consecutive double-digit decline, and gross margin fell to 57.7%. The per-unit record is unambiguous: subscribers of 951,000 against 1,073,000 and paid downloads of 98.7m against 112.6m, with average revenue per customer rising to $292 from $266. A smaller base, each remaining account spending more.

The hedge is shrinking faster than the thing it hedges. Data, Distribution and Services revenue, 25% of the total, fell 16% to $56.1m, dragged by the data offering, down 11% in the June quarter after a 63% drop in the March quarter. The company also took a $173.7m goodwill impairment after its fair value fell on the collapse of the Getty merger, and attributed lower content revenue chiefly to weak new-customer acquisition.

At $4.14 the equity is worth $153m, below a single quarter's revenue, and forward price-to-sales of 0.195x sits above the trailing 0.170x. That inversion is the market marking the denominator down, with consensus revenue falling 23% this year and a further 8% next.

Founder Jon Oringer, chief executive until 2020, returned to the job effective 28 October. "AI is fundamentally reshaping the creative industry, and the pace of change is accelerating," he said in the 29 October announcement. The next day the company said it would not hold its 5 November earnings call while it "continues to evaluate its strategic direction and priorities"; the shares fell 6.1% on the heaviest volume of the month.

Two collapses, two causes

This corner of software looks like one story in October and is two. Getty's 78% monthly loss is a documented capital-structure event with filed terms, and its operating margin went the right way throughout. Shutterstock's much smaller monthly decline sits on top of a year in which the equity lost 83%, and there the income statement and the download count agree: the catalogue is being replaced, and the fees model developers pay for data are falling faster than the licensing revenue they displace. The software vendors, Adobe and Figma, both finished the month higher, with revenue growing 12.9% and 48.2% respectively and gross margins down roughly one and five percentage points as inference costs landed in cost of revenue.

Shutterstock will publish third-quarter figures on 5 November with nobody on the phone to explain them.

Smithfield Foods, which raises hogs and sells Smithfield bacon, Nathan's Famous hot dogs and Eckrich deli meat, told investors on 27 October that its branded Packaged Meats segment should earn adjusted operating income of $1,075m–$1,150m this fiscal year. The company as a whole is guided to $900m–$1.0bn. The branded arm is expected to out-earn the business that contains it, because the farms and slaughter plants feeding it are losing money.

Four of the five big US meat and poultry names have been marked down between about 7% and 29% over the past year, Tyson Foods excepted, on a premise that cheap grain and cheap hogs would widen processing spreads. The quarters reported in June and September show something narrower. Inside a vertically integrated protein company the cheap raw material is a transfer, not a windfall, and it settles wherever the brand sits.

Smithfield's margin came out of its own barns

Packaged Meats earned about $230m in the September quarter, with margin widening to 11.6% from 10.8%, on sales down 5.2% to $2bn. Average selling price fell 5.3% and volume rose 0.2%: the segment made more money per pound because hog and pork cost fell faster than the price it charged. Upstream, Fresh Pork swung to a loss of $89m from a $10m profit as the processing spread fell 35% and volume dropped 6.3%, while Hog Production earned $34m against $89m. Group adjusted operating profit fell 55% to $145m.

The share gains downstream rest on distribution and brand. "Across our top 25 categories, industry volume declined 3.3%, while we grew volume by nearly 1%. We were the only company in our core competitive set to grow volume in this quarter, driving 0.4 points increase in volume share," Shane Smith, president and chief executive, told investors on 27 October. Smithfield gained share in all five of its largest categories, led by 2.3 points in smoked ham, and its Nathan's Famous purchase eliminates $33m of annual licence costs on rights previously held only through 2032. A new supply agreement with a joint-venture partner cuts internally produced hogs by roughly 900,000.

The shares closed 27 October at $18.33, a few cents above a 52-week low, at 1.03x book with a 12.8% trailing free-cash-flow yield and net debt at 0.5x EBITDA. The forward earnings multiple of 9.3x sits above 8.0x trailing, so consensus is still falling.

Pilgrim's had nothing downstream to catch it

Pilgrim's Pride, the Greeley, Colorado chicken and pork processor controlled by JBS, is the same geometry with the brand missing. US net revenues slipped to $2.65bn from $2.82bn in the June quarter while US adjusted EBITDA fell to $231.5m from $482.7m, a margin of 8.7% against 17.1%, which the company attributed primarily to a 27% drop in the jumbo cutout value. Consolidated operating margin has run 10.4%, 4.5%, 3.6% and 1.4% over the last four quarters.

Feed, the premise the widening-spread case rested on, moved against the processor. Corn ran roughly $4.25–$4.50 a bushel for the year but averaged $5.18 in the first week of August, and October soybean meal traded at $350–$360 a short ton in mid-September against a first-half range of $275–$325. Corn and soymeal are 60–70% of a broiler's diet.

The company is also adding pounds into the falling cutout: about 2.5% US supply growth in the second half. "What we want is to grow our branded business in the U.S. That will compensate the volatility in the big bird commodity market. Once again, we don't want to reduce our commodity operation," chief executive Fabio Sandri said on the 30 July call. Consensus has fiscal 2026 earnings at $2.27 a share against $5.26 for 2025; at $27.09 the forward multiple of 11.9x is wider than the roughly 8.4x the stock carried last October on $4.54 of delivered earnings.

Hormel divested the turkeys and kept the margin

Hormel Foods, the SPAM, Skippy and Jennie-O owner, is the counter-case. Gross margin moved 15 hundredths of a percentage point in the July quarter, against a 7.7-point loss at Pilgrim's; adjusted operating margin expanded to 9.0% and the company raised fiscal 2026 adjusted earnings guidance to $1.45–$1.51. Roughly half its segment sales decline came from divesting the whole-bird turkey business and exiting private-label snack nuts. "With our strong year-to-date performance and continued opportunities ahead, we are raising and narrowing our adjusted earnings outlook for fiscal 2026," interim chief executive Jeff Ettinger said on 27 August. The shares still fell 19.4% over three months, and at 13.3x forward earnings Hormel is marginally cheaper than Tyson's 14.1x — Tyson being the name whose beef unit is guided to a fiscal 2026 operating loss of $625m–$775m while its chicken segment is guided to $1.85bn–$1.95bn.

Arithmetic at Pilgrim's, an open question at Smithfield

Three companies, one cheap pound of meat, three outcomes decided by who owns the label on the package.

For Pilgrim's the market has done the arithmetic: a 28.5% decline over the year against consensus earnings more than halved, and a forward multiple that widened as the price fell. Smithfield is the awkward one, sitting at a 52-week low on the day it reaffirmed the segment guided to carry all of group profit; management tied the duration of the upstream loss to the fourth-quarter holiday season and spring 2027. Hormel's de-rating tests something else entirely: organic sales growth guided to 1–2%.

Pilgrim's Pride reports its third quarter after today's close, with consensus at 48 cents a share. Smithfield has shown where the money in a cheap pound of meat ends up; nobody can yet price when the cutout stops falling for the company that only sells pounds.