DK Street Journal

DA Davidson cut BlackLine to Underperform on 28 September, lowering its price target to $23 from $30 and citing competition from emerging AI labs as a threat to the growth trajectory of the accounting-close software maker. The shares closed down 4.8%. It was the first company-specific news to reach any of three vertical software makers in weeks — and it arrived after all three had already fallen between 17% and 23% over the preceding month.

That uniformity is the finding. Appian, which sells a low-code platform for building workflows and applications to banks, insurers, drugmakers and government agencies; BlackLine, which automates the corporate accounting close; and Procore, whose platform runs construction projects for owners and contractors, are paid in three ways that share nothing. Yet September moved them as a single instrument. Over the three months from late June they had done the opposite — Appian up 61.2%, Procore up 17.1%, BlackLine down 5.9%.

Three meters, three answers

Appian is paid on cloud subscriptions and services. Second-quarter cloud subscription revenue reached $131.7m, up 23%, with total revenue of $203.3m up 19.1% and adjusted earnings before interest, taxes, depreciation and amortization doubling to $16.2m. Four verticals — financial services, insurance, life sciences and the public sector — supply 80% of annual recurring revenue, chief financial officer Serge Tanjga told a William Blair conference, and more than 70% of the business comes from customers spending over $1m a year.

BlackLine no longer sells seats. More than 90% of new customers now sign uncapped platform contracts with no limit on logins, a shift chief executive Owen Ryan told investors on the 4 August call deliberately reduces the near-term contribution seat additions used to make. Procore never sold seats at all: it charges an annual fee per product set by a customer's Annual Construction Volume — the aggregate dollar value of work across that customer's projects — with unlimited users included.

Where the businesses and the prices disagree

BlackLine's reported accounts improved. Revenue growth accelerated from 7.2% for full-year 2025 to 9.2% in the June quarter, operating income rose 45.7% to $11.0m on a 76.0% gross margin, and the company held full-year guidance at $765m-$769m. The leading meter tells the other story: annual recurring revenue of $719m grew 6% against a stated 13-16% ambition, dollar-based net revenue retention was 102.4%, and the mid-market customer count is still falling. Ryan said artificial-intelligence reviews have made large enterprise deals harder to predict, with finance, security and governance teams now assessing them together. The de-rating tracks the recurring-revenue line, not the profit-and-loss statement.

Appian runs the reverse. Growth decelerated across three quarters — 21.7%, then 21.5%, then 19.1% — gross margin compressed to 71.2% from 75.8% a year earlier, and the raised full-year revenue guide of $845m-$853m implies about 17%, below the pace just posted. Against that, its price-to-gross-profit ratio expanded from 3.07x in early May to 4.46x. The business did earn something real: a first full-year operating profit in 2025, of $0.6m on $726.9m of revenue, against a $60.9m loss the year before. "[Pulling 2027 hiring forward] might be an indication of strong pipeline that's unusually strong and also our confidence of being able to win in the current environment," chief executive Matt Calkins said on the 6 August call. Consensus does not agree: forecasts have revenue growth falling to 10.8% in 2027.

Procore's growth has been the steadiest of the three, inside a band from 15.6% to 15.8% for four straight quarters, and the June quarter produced its first positive operating margin at 1.2%. Its problem sits outside software budgets. Private nonresidential construction spending fell 4.7% year on year in June and 7.9% excluding data centers; the Architecture Billings Index read 47.2 in August, below the line separating growth from contraction. On 9 September Procore closed its largest acquisition, paying about $845m in cash for DroneDeploy, a reality-capture and robotics platform — roughly 11% of its market value, spent as its index contracts. Gross margins differ too widely for revenue multiples to compare here; measured against gross profit Procore is the dearest of the three at 6.52x, down from 7.96x in early May, while BlackLine sits at 2.82x from 3.58x.

The verdict

What the businesses earn and what the market did are separable. BlackLine's fall is earned — by 6% recurring-revenue growth, not by its accounts. Procore's is earned by the construction index its contracts are written on, not by churn. Appian's summer advance is the one nothing in the revenue line explains: growth slowed at every reading while the multiple did the work. And the September leg belongs to none of them. It came with a broader software de-rating in which roughly $2 trillion of enterprise software value has gone since early 2026, alongside the higher long yields the market has been living with since the 10 September 30-year auction. The likelier reading is rotation out of long-duration software, which does not read billing models.

The largest number attached to any of the three appears in none of their revenue lines. Appian's vacated $2.036bn trade-secrets award against Pegasystems returns to a Fairfax County jury on 11 January, and Pegasystems carries no accrual for it.

Amkor Technology, which assembles and tests other companies' chips for a fee rather than designing any, told the market on 8 September that it would raise planned investment in its Peoria, Arizona campus from about $7bn to roughly $12bn, adding some 60,000 square meters of cleanroom. The reason given was that customer commitments had already exceeded the 33,000 square meters planned for the first phase.

Part of the money is somebody else's. Nvidia has committed a $1.5bn prepayment — cash against future capacity, not an equity stake — to fund the US expansion and move packaging technology from Korea to Arizona. The campus also carries $407m of funding under the CHIPS and Science Act, lists Apple and Nvidia as lead customers, and finishes construction in mid-2027 for production starting in early 2028. The spend is now; the revenue is three years out. And Amkor is making that commitment while the market has cut its valuation in half.

One shortage, two directions

Amkor's June quarter was its best in years: revenue of $1.898bn, up 25.6%, with net income more than tripling and gross margin widening from 12.0% to 16.8%. "Amkor delivered record second quarter revenue and strong profitability, with record revenue in our Computing and Automotive & Industrial end markets," chief executive Kevin Engel said on 27 July. The shares fell more than 10% anyway, on a third-quarter guide of $1.95–2.05bn against a $2.12bn consensus.

What broke was the phone line. Amkor guided communications revenue down high-single-digits on the shift of system-in-package work to Vietnam, "ongoing memory supply constraints" and changed customer build patterns, with Android revenue already down 20%. Memory makers have diverted wafers to high-bandwidth memory for AI servers; conventional DRAM contract prices rose about 90% in the first quarter of 2026 alone. Amkor cannot assemble modules around chips its customers cannot buy.

ChipMOS, the smallest of the three, sits on the other side of the same trade. It packages memory and display driver chips, and its August revenue rose 33.3% year over year on what the company called strong memory demand led by a persistent AI-related imbalance. Memory reached 51% of revenue in the June quarter, up 46.7%, and gross margin went from 6.6% to 18.0% — while utilization fell from 84% to 78%. "Better pricing, product mix, and utilization are now translating demand strength into meaningful operating leverage," chairman S.J. Cheng told investors on the 11 August call. Margin up on fewer loaded machines is price, not volume.

What ASE is charging

ASE Technology, the world's largest outsourced assembler, raised advanced-packaging quotes by more than 20% in July, covering the chip-on-wafer work that AI accelerators need. Its assembly-and-test arm set a record at TWD 126.1bn in the June quarter at a 27.3% gross margin, producing 94% of group operating profit on 66% of revenue; the electronic manufacturing services arm contributed a third of sales and almost nothing to profit. Chief operating officer Tien Wu described the company on the 30 July call as being in "a very awkward and peculiar position because we're capacity-constrained." ASE lifted capital spending by a further $2bn, to about $10.5bn, and its trailing free cash flow has turned negative.

The three trade as though only two are in the same business. Amkor is at 10.0x trailing enterprise value to EBITDA and 20.2x forward earnings, its trailing price-to-earnings ratio down to 23.2x from 44.4x in May even as trailing profit rose. ChipMOS is at 10.5x on the same EBITDA measure and, converting Taiwan-dollar consensus at the rate implied by its ADR, roughly 24x 2026 earnings. ASE is at 21.3x EBITDA and about 37x 2026 earnings — more than double its peers on the first measure. Over twelve months Amkor has risen 79%, against 283% for ASE and 269% for ChipMOS, and it sits 39% below its 25 June close with its 50-day average under its 200-day.

The verdict

Packaging is genuinely being repriced: ASE's quote increase and ChipMOS's eleven-point margin gain on falling utilization are both price and mix, and that is broader than AI packaging alone. Amkor earns the same tailwind in computing and gets taxed by memory scarcity in phones, which is what its discount reflects — not a verdict on its AI book. The Arizona decision converts it from a cyclical earner into a capital-spending story with an anchor customer's cash on the balance sheet and no revenue until 2028.

Micron reports on 30 September. How much DRAM it is willing to sell to anyone not building an AI server sets the sign on two of these three companies at once.

An activist investor has taken a position in the smallest listed US aggregates producer and given its board a clock. Starboard Value delivered a letter to Knife River's board on 24 September disclosing a stake and calling for a credible plan to reach at least 22% adjusted EBITDA margins by fiscal 2029, through better aggregates pricing and cost reductions — or, failing that, for directors to consider strategic alternatives including a sale. Starboard called the company "an exceptional business in a highly attractive industry" that had "failed to realize its potential," with returns since the 2023 spin-off from MDU Resources it termed "unacceptable."

The demand lands on a company whose problem is not the price of rock. Knife River, which mines aggregates and mixes concrete across the Midwest and Northwest and also bids its own paving, grading and heavy-civil work for state and municipal customers, lifted mix-adjusted aggregate pricing 8% in the June quarter — more than either larger peer — grew contracting revenue 20% and carried a record $1.2bn backlog at 30 June. Revenue rose 12.6% to $938.6m. Operating income fell 7.9%. The company blamed energy costs, delayed projects and the timing and type of contracting work, raised its 2026 revenue guide to $3.4bn–$3.6bn, and cut expected full-year aggregates margin expansion from two percentage points to about one.

Price lands, profit doesn't

That gap between the top line and the profit line is the segment's condition, not one company's. Vulcan Materials, the largest US aggregates producer, raised freight-adjusted mix-adjusted prices 5% in the same quarter and converted it into cash gross profit of $12.02 a ton against $11.88 — fourteen cents — with almost $40m of energy inflation absorbing the rest. "Price is our biggest lever when it comes to overcoming headwinds like this and inflationary pressures," chief executive Ronnie Pruitt told analysts on Vulcan's late-July call. Martin Marietta, whose quarries and cement plants concentrate in the Sun Belt, shipped 61.6m tons, up 2.3% organically, raised organic mix-adjusted price 4%, and reported gross profit per ton down 17% to $6.78 — of which $0.84 was the charge for selling acquired inventory written up to fair value. Revenue grew 7.5%; operating income fell 17.2%. Neither company cut guidance: Martin Marietta raised its revenue range and reaffirmed $2.36bn–$2.5bn of adjusted EBITDA.

The volume question sits in 2027. Federal highway funding was extended only to 11 December 2026, and the enacting resolution cut transportation funding 25% in total. Wells Fargo cut Vulcan to Underweight in early September on slowing construction and constrained state budgets while upgrading Martin Marietta, a split verdict inside one industry.

What the discount pays for

Knife River trades at 10.8x trailing enterprise value to EBITDA against Vulcan's 14.2x and Martin Marietta's 16.8x, and at 19.7x forward earnings — cheaper than both larger peers on every measure recorded here. Vulcan's price-to-gross-profit has compressed about 20% since early May, when it stood at 17.7x, while its gross profit grew. Martin Marietta's reported multiples flatter it: the $13.5bn Lhoist lime combination closed on 21 August issued 10.95m shares, roughly 18% dilution, alongside $5.5bn of notes priced between 4.85% and 6.375%.

Shares tell the sequence. Knife River is down 41% in three months and closed at $51.43 on 23 September before recovering to $54.58 after the Starboard news; Martin Marietta and Vulcan are each off roughly a fifth over the same span, with Vulcan trading below Wells Fargo's own bearish $254 target.

So the compounder story is only half broken. Annual price increases on quarries nobody can permit a competitor into still work — all three proved it in the same quarter. What has stopped working is the translation into EBITDA, and Starboard's wager is that at Knife River the failure is self-inflicted rather than cyclical, fixable with cost discipline whatever Congress does about highways. If it is right, the segment's problem is management. If it is wrong, eight points of margin is a very long way to ask a diesel bill to travel.

Sunrun sold a slice of its seasoned rooftop solar and battery leases to bond investors in August, and they charged it less than at any point in roughly a year and a half. The $267m securitization priced its publicly placed A- rated senior notes at a 6.33% yield — a 200 basis-point spread over the benchmark, 20 basis points tighter than the company's April deal and, per chief financial officer Danny Abajian, the tightest in about 18 months. Three weeks later the shares made a 52-week low.

That gap is the whole question in residential solar. Sunrun, the largest US installer and financier of home solar-and-battery systems, builds a system, keeps it, and collects a 20-to-25-year lease or power-purchase payment — funding the up-front cost with nonrecourse asset-level debt and by selling federal tax credits. It has raised roughly $1.5bn of such debt so far this year and booked a $98.4m tax benefit from investment-tax-credit transfer agreements in the first half. Credit investors are underwriting the cash flows of systems already on roofs, and they have grown more comfortable. Equity investors are underwriting the next system, and they have not.

The next customer got much more expensive

Sunrun's June quarter made the distinction plain. Revenue rose 53% to $870.0m, but net subscriber value — the company's own measure of value created per new customer, net of the cost to create it — fell to $9,444 from $17,004 a year earlier. Strip out assumed contract renewals and contracted net subscriber value fell 61% to $5,100, even as contracted subscriber value rose 10% to $55,033: the price held, the cost did not. Additions fell 31% to 19,793, and full-year cash generation guidance came down to $200–375m from $250–450m. Sunrun also stopped publishing aggregate creation costs and aggregate net value creation, replacing them with two new measures — a disclosure change that retires the most comparable unit-economics series just as it deteriorated.

The causes are dated and mostly external. Section 25D, the 30% federal credit for homeowner-owned systems, expired on 31 December 2025 with no phase-down; leases and power-purchase agreements kept a 30% credit under Section 48E. Customer-owned share of the market duly fell from 54% to 43% while third-party ownership rose to 52%, and SEIA and Wood Mackenzie expect US residential installations to fall around a fifth this year. Wood Mackenzie also expects customer-acquisition costs to spike about 40% in 2026 — the mechanism behind the creation-cost jump. Add the failure of Freedom Forever, the second-largest installer at roughly 6.1% share against Sunrun's 12.7%, whose Chapter 11 became a Chapter 7 liquidation this month; Sunrun financed its systems, so a rival's collapse cost Sunrun volume. "Customer demand for our offering remains strong, and as our expanded sales force reaches full productivity, we believe that we will exit the year at a robust growth rate and higher unit margins," Abajian told investors on the second-quarter call. Storage attachment, at a record 74%, is the one metric moving his way.

The last month was the discount rate

None of that explains September. The Federal Open Market Committee raised its target range to 3.75–4.00% on 16 September, a first increase in more than three years, with 16 of 18 participants penciling in another; the 30-year Treasury reached its highest since June 2007. On 24 September solar equities fell together on financing costs rather than demand: First Solar about 8%, SolarEdge about 5%.

First Solar sells cadmium-telluride modules to utility-scale developers and has no rooftop business. Its June quarter carried a 57.3% gross margin, consensus has 2027 earnings of $23.36 a share against $17.55 this year, and it trades near ten times forward earnings — yet it is down 28.5% in three months and only slipped into a downtrend on 1 September, the last of the group to break. XPLR Infrastructure, the externally managed owner of contracted wind, solar and Texas gas infrastructure, is the counter-example: up 8.65% over twelve months, the only name here whose trend is not falling. It reaffirmed 2026 free cash flow before growth of $600–700m after $257m in the quarter, bought out a $150m slice of a convertible-equity portfolio financing and repaid $500m of converts while paying no distribution at all. It trades at 8.8x trailing earnings before interest, tax, depreciation and amortization and 0.31x book, against 14.3x and 0.68x for Clearway Energy, which owns a similar contracted fleet and trimmed its distributable cash guide to $430–470m on weak wind and solar resource.

What each part earns

The twelve-month verdict belongs to Sunrun alone: returns across these four names span 60.7 percentage points, from Sunrun's 52% decline to XPLR's gain, so there is no single renewables judgment to read. Sunrun's de-rating is earned — halved unit economics, a cut cash guide, and a repealed credit are a business event, not a mood. The thirty-day leg is a rate event, and it fell hardest on the businesses least exposed to rooftops, which is the signature of a discount-rate move rather than a demand one. The awkward fact for the bear case is the bond: the market that actually prices Sunrun's contracted cash flows tightened while the market that prices its growth made a new low.

The clock is the thing to watch. The 30% credit that survives only for leased and third-party-owned systems runs through the end of 2027, and only for projects whose construction began by 4 July 2026. Sunrun's financing structure has a known expiry date; its customer acquisition costs do not.

Fiserv's revenue is going backwards. The Brookfield, Wisconsin group, which sells card acceptance to small merchants through its Clover point-of-sale system and deposit and loan processing to banks, reported June-quarter revenue of $5.29bn, down 4.1% from a year earlier — the end of a four-quarter walk from growth into decline. Operating margin came in at 19.2%, against 30.7% in the same quarter last year.

That matters past one company because Fiserv, Jack Henry and Fidelity National Information Services, which trades as FIS, between them run 72% of the American bank core-processing market — Fiserv 42%, Jack Henry 21%, FIS 9% — and are paid in three incompatible ways for what gets described as the same trend. Over the past month the market priced them as one animal: seven of the eight companies in this corner of technology fell, averaging 11.1%, while the broad US market rose 0.7%.

The merchant's spread

Fiserv takes a spread on volume it re-wins quarter by quarter, and that meter is still running. Clover gross payment volume rose 9% in the June quarter, 11% excluding a gateway conversion, with value-added services revenue up 10% and penetration at 25% of Clover revenue from 24%. Reported Clover revenue rose only 2%, because weaker hardware and data revenue cost roughly nine points of growth. Merchant Solutions organic revenue fell 1%.

So the damage is not, on management's account, merchants walking. The 6 August reset — organic revenue to -1% to 0% from 1% to 3%, adjusted earnings to $7.20-7.40 a share from $8.00-8.30 — was attributed in sized pieces: about two points from delayed contracted revenue and slower enterprise launches, a point from product and hardware, roughly a point each from Argentina and planned business sales. Takis Georgakopoulos, who became chief executive on 15 June after Mike Lyons left for Truist Securities, says he has seen no material change in new merchant acquisition and treats the hardware slowdown as temporary.

It was the second reset in ten months. The first, on 29 October 2025, took 44% out of the shares in a session, the worst day in the company's history. "Our current performance is not where we want it to be nor where our stakeholders expect it to be," Lyons told investors that day, five months into the job. Fiserv now trades at 6.4x consensus 2026 earnings of $7.25 a share against roughly 12.6x the guidance it carried a year ago: the multiple has halved while expectations came down about 28%. Clover is still the leader in small-restaurant card processing, at about 20% of that market with 175,000 locations, against Toast's 17%.

The renewal cycle

Jack Henry, founded in Monett, Missouri in 1976, is paid under multi-year core contracts that community banks and credit unions renew on long cycles whatever cards do. Its year to 30 June brought revenue of $2.50bn, up 5.1%, with operating income up 6.6% to $606m — and 58 competitive core wins, the most in more than two decades, 59% of them bundling core, digital and card against 39% a year earlier. "We signed the largest new bank client in our company's history in Q4, Woodforest National Bank with $9.2 billion in assets," Gregory Adelson, president and chief executive, said on the 19 August call. Fiscal 2027 guidance was raised: non-GAAP revenue growth of 6.3-7.3%, earnings of $7.33-7.38 a share, 58 to 65 core wins.

What fell was the pace. The shares dropped 6.4% on 15 September, its Investor Day, when management put organic growth at 7.5% at the midpoint in fiscal 2029 against 6.8% in fiscal 2027 and flagged margin pressure from cyber, infrastructure and artificial-intelligence spending. Its June quarter on reported measures also decelerated, to 2.9% growth with operating income down 14.8%. At 20.1x forward earnings the stock sits near last year's 21.7x — but two years ago that multiple was about 28x, so the compounder premium was taken back well before this month.

The issuer's book

FIS is the reason this is one story. On 9 January it closed the $13.5bn purchase of Global Payments' Issuer Solutions business, the former TSYS, and sold its remaining 45% of Worldpay to the same buyer, guiding the swap to add about $500m of adjusted free cash flow this year. Jacksonville is now levered to card issuance and bank accounts rather than to merchant volume at all. June-quarter revenue of $3.38bn grew 5.3% pro forma, adjusted earnings per share 8.8%, recurring revenue 5% and recurring sales 14%, with free cash flow of $525m. "Our first half reflects the strength of the business we have built defined by durable recurring growth, expanding margins, and accelerating cash generation," chief executive Stephanie Ferris said on 4 August. FIS trimmed pro forma revenue growth guidance to 4.5-5.0% from 5.1-5.7%, kept adjusted earnings growth at 7.0-8.5% and raised its free-cash-flow growth guide to 33-39%.

The shares are down 45% over twelve months while consensus 2026 earnings rose 7.4% to $6.20 a share. That is 5.7x forward, against roughly 11.1x a year ago. The trailing 5.4x is unusable: March-quarter net income includes the gain on the Worldpay sale.

What each part earns

The twelve-month damage was never shared — Fiserv lost 64%, FIS 45%, Jack Henry 2.4%, and Flywire, in the same heading, rose 32%. September was the month they moved as a bloc, and it was a grind: none of the three recorded a 7.5% session in the 26 trading days to 25 September. Fiserv and FIS now have their 50-day averages below their 200-day; Jack Henry's is still above.

Fiserv's de-rating is earned, though the price has travelled further than the estimates. Jack Henry's is a repricing of how fast a working franchise compounds, with volumes, wins and guidance all pointing up. FIS's is the one nothing in the reported numbers explains. What the three genuinely share is a renewal table: the Office of the Comptroller of the Currency asked community banks about their core providers and heard of "lopsided commercial negotiating power", which is a pricing risk to all three at once — and the only fact so far that argues for treating them as one security.

Fiserv's third-quarter results are expected on or about 27-28 October, a year on from the print that made its name a warning to the whole heading. FIS reports its third quarter and Jack Henry its fiscal first in early November, close enough behind to settle whether one company's contraction was ever the industry's.

A rally the wellhead did not get

On 24 September TC Energy's Columbia Gas Transmission declared force majeure on the Mountaineer XPress system after a leak at a West Virginia regulator station, cutting as much as 1.8 Bcf/d of firm service on a line built to carry 2.7 Bcf/d of Appalachian gas out of the basin. Henry Hub futures posted their biggest single-day gain since January. The producers stranded behind the blockage did not rally with it — an outage on takeaway lifts the national benchmark and worsens the price at the wellhead at the same time, a pattern S&P Global recorded on an earlier Mountaineer failure.

The whole month ran that way. Between 15 and 25 September the front-month contract rose 9.7% to $3.225 per million British thermal units while EQT fell 4.3%, Expand Energy 7.8% and Antero Resources 9.9%. These shares are not valued on the prompt; they are valued on next winter, and the winter 2026/27 strip hit its lowest of the year in early September on a strong El Niño and stout supply. In August the Energy Information Administration had already cut its 2026 Henry Hub forecast more than 6% to $3.44, citing record production and a projected 3,985 Bcf in storage at the end of October.

The drawdown is six months old, not four quarters. EQT, Expand and Comstock Resources peaked on 25 March, Antero and Range Resources on 27 March; from those highs EQT is down 25.2%, Expand 23.8% and Antero 22.5%, though Antero and Range remain slightly higher than a year ago. A third of the September decline came in four sessions from 16 to 21 September, and 16 September was a broad energy sell-off — the XLE energy fund fell 1.9% and West Texas crude 3.2% while gas itself fell 1%. Tourmaline Oil, the Canadian producer selling into AECO rather than a US hub, lost 1.3% over the month. The problem has a postcode.

One curve, three different businesses

Antero, the Denver producer whose roughly 502,000 net Appalachian acres yield an unusually wet barrel of ethane, propane and heavier liquids, is having its best year. June-quarter revenue rose 11.1% to $1.34bn and net income 78% to $279m on record output above 4.1 Bcfe/d, up 21%, with full-year guidance raised to 4.15–4.2 Bcfe/d. Its C3+ liquids barrel realized $44.33, the highest since 2022, and it carries no liquids hedges. "Ten to 15 years back, we signed up for all the firm transport arrangements just to get our gas out. Now we're at the end of that. We can select the best paths," chief executive Michael Kennedy told investors on the second-quarter call. It trades at 8.22x forward earnings against 9.97x trailing — the only one of the three the market prices for growth — on a 13.1% trailing free-cash-flow yield.

EQT, the Pittsburgh gas extractor that owns its gathering system and a stake in Mountain Valley Pipeline, shows the strip in its accounts: revenue down 29.2% to $1.81bn and operating margin down to 25.1% from 44.3%. Operationally it improved — 634 Bcfe sold above the top of guidance, unit operating costs of $1.03 per Mcfe against $1.08, and net debt cut to $5.5bn from $7.7bn against a $5bn target. It is also the most expensive of the three, at 12.49x forward earnings against 11.16x trailing, and the least protected: roughly 35% of local basis hedged against a historical norm near 90%. "I don't see a way for basis not to continue to strengthen materially," chief financial officer Jeremy Knop said on the 22 July call.

Expand, the former Chesapeake and the largest US gas producer, straddles both problems: about 7.48 Bcfe/d, 42.6% of it Haynesville gas next to the Gulf Coast liquefaction trains. Revenue fell 19.7%, realized gas came in at $2.42 per Mcf, and 66% of 2026 volumes and 41% of 2027 are already hedged — protection on the way down and a ceiling on any recovery. It has retired $850m of stock this year and cut net debt to $3.1bn, trades at 3.48x trailing enterprise value to EBITDA against EQT's 6.02x and Antero's 6.44x, and has had no permanent chief executive since February.

What the fall earns

Part of this is earned. EQT's and Expand's reported revenue and margins genuinely fell with the gas price, and a market that paid a scarcity premium for gas in March is taking some of it back. But the forward numbers did not follow the shares down: consensus 2027 EBITDA for the three is flat to 5% lower than 2026, and through the exact window of the September slide EQT's 2026 estimate moved from $4.14 to $4.07 against a 7.9% share fall while Antero's held near $4.26 against an 8.9% one. At Antero, where revenue, volumes and guidance all rose, nothing in the reported business explains a fifth off the price. The mechanism is a curve, and the curve is not the company.

The tightness these shares were bought for is dated. The same EIA outlook that cut this year's price has supply outrunning demand by 0.5 Bcf/d in 2026 and falling 1.6 Bcf/d behind it in 2027 as liquefaction demand arrives. EQT's ten-year supply contract to a West Virginia power plant and Expand's Delfin cargoes both begin delivering around 2031. Owning the gas is the easy part; being paid for it requires holding the position until the buyers turn up.

Newmark Group, the New York firm that brokers property sales and leases, originates government-agency multifamily loans and then services them, ended June with a record primary servicing portfolio of $81.2bn, up 16.8% from a year earlier, according to its quarterly filing. That book bills monthly on loans outstanding, whether or not a single building changes hands. Weeks later the long end of the bond market did the thing that stops buildings from changing hands.

Newmark is the cleanest place in commercial real estate services to ask which of those two forces is winning, because it is built deliberately the opposite way from the industry's scale names: capital-markets advisory and agency origination on one side, a servicing asset that marks on rates rather than on deal count on the other. The total servicing and asset-management portfolio stood at $219.3bn, of which Fannie Mae, Freddie Mac and Federal Housing Administration loans are 30.6%, and mortgage servicing rights carried a $514.0m balance after $28.2m of additions in the quarter.

The broken rate assumption

The ten-year Treasury yield closed at 5.17% on 25 September, its highest since June 2007, after a three-session, 23-basis-point selloff — a further leg beyond the 5.04% reading this page cited a week earlier. The consequence for brokerage fees runs through a forecast, not through observed weakness. The Mortgage Bankers Association projected commercial mortgage originations rising 27% to $805.5bn in 2026, multifamily at $399.2bn; that forecast assumed a 4.2% average ten-year yield, roughly 100 basis points below what actually printed. Tenant demand is not the problem: national office net absorption ran a ninth consecutive positive quarter, and first-half absorption of 40.7m sq ft was the strongest first half since 2019, with vacancy at 15.8%.

A servicing book behaves opposite to a brokerage book when yields rise. Higher rates slow expected prepayments, lengthening the life of cash flows carried at fair value and lifting escrow earnings. With 4.8 years of weighted-average maturity on a record portfolio, the same bond move squeezing Newmark's transaction fees should be marking its servicing asset up — an inference from the June disclosure, which Newmark has not quantified for the September quarter.

What the June quarter actually cost

Revenue grew 17.0% to $888.4m, a fourth straight double-digit quarter, with management and servicing up 18% in its fourth consecutive record quarter. Operating income nonetheless fell 5.5% to $40.4m and the margin compressed to 4.55% from 5.63%. Newmark held, rather than raised, its full-year guide of roughly 16% revenue and 19% adjusted earnings-per-share growth, management citing tougher second-half comparisons and uncertain timing on large transactions. S&P Global Ratings then raised the company to investment grade at BBB- on adjusted leverage of 1.2x. Set against that, the succession is open: Newmark said on 7 August that Barry Gosin will step down as chief executive on 31 December 2026 after holding the job since 1979, with a successor expected by year end.

CBRE, the larger rival that ranks first or second globally alongside JLL, spent the drawdown winning contractual work rather than losing it: Fermi disclosed on 24 September that CBRE had been signed as exclusive operations-and-maintenance provider for the first building at Project Matador in the Texas Panhandle, a five-year deal with five-year renewal options. Its contracted revenue grew 15% in the June quarter against 19% transactional, and it raised full-year core earnings guidance to $7.80–$7.90 a share on 29 July, about 23% growth. Nobody has revisited that, for the simple reason that no company in this group has spoken since early August; the next prints land in late October. JLL, which grew June-quarter revenue 10.8% to $6.93bn with operating income up 33% and trades at 13.0x forward earnings, is the only one of the five higher over twelve months, up 7%.

Where the group trades

On trailing enterprise value to operating earnings before depreciation, CBRE at 17.8x is the most expensive of the five, against Cushman & Wakefield at 12.7x, Colliers at 11.2x, JLL at 11.1x and Newmark at 7.9x. Part of CBRE's fall is a premium being taken back from the group's premium name, and its 17.2x forward core earnings against consensus growth near 24% is the anchor that has not moved. Newmark's case is genuinely contested rather than clean: cheapest asset, record recurring book, investment grade — and a margin that went backwards while revenue grew.

The verdict the evidence supports is narrow. Falling transaction volume is a forecast that has been invalidated, not a result that has been reported; what the filings show is recurring revenue compounding and pay costs eating the increment. The de-rating is being applied to the least cyclical parts of both businesses, and for Newmark the rate move that threatens the fee line is the same one that should lengthen the servicing asset.

Which leaves one thing genuinely unhedged. Newmark will be the first of the group to be asked what a 5%-plus long end does to its origination pipeline, and it will be asked by a chief executive who has already announced he is leaving.

On Wednesday the contract that pays for the only high-assay low-enriched uranium cascade in the United States runs out. The Department of Energy amended it on 30 June to run only through 30 September, exercising a three-month option worth a fixed $15.0m for cascade maintenance and storage of already-produced material — with no enrichment at all — and its proposed fiscal 2027 budget contains no funding to operate the Piketon cascade, with the department having told Centrus Energy it does not currently intend to exercise further options.

That is the hardest single fact behind a year in which enrichers and miners fell while the commodity they exist to supply made highs. Uranium spot sat near $90 a pound in mid-September, up roughly a fifth year on year, with the long-term contract price at a nominal record in the mid-$90s. The two can move apart because nobody in this business is paid for "demand." Centrus is paid per separative work unit out of an order book priced in earlier years; Cameco is paid for pounds it committed years ago; BWX Technologies is mostly paid cost-plus by the US Navy. Three meters, and they do not agree.

Centrus: volume down, unit cost up

Centrus Energy, the Bethesda enricher that sells separative work units and natural uranium to utilities and runs the American Centrifuge program, grew June-quarter revenue 14% to $176.1m and lost most of the profit anyway. Gross margin fell to 28.3% from 34.9% a year earlier and operating income dropped 69% to $10.4m. The mechanism is on its own face: total costs for units sold fell 23% on lower volumes while the average cost per unit sold rose 13%. Annual gross margin has now compressed four years running, from 40.1% in 2022 to 26.2% in 2025.

The forward book is not the problem. Backlog grew to $4.5bn extending to 2040, including roughly $3.0bn of contingent commitments tied to a Piketon build, and DOE signed a roughly $1bn fixed-price commercial award on 30 June requiring one metric ton of HALEU by March 2032 and an initial 12 tons of annual capacity. "This was another strong quarter of financial and operational progress for Centrus," chief executive Amir Vexler said on 5 August. Paying for it is the issue: guided capital spending of $350m–$500m against a $2.79bn market value, negative free cash flow, and a $500m offering priced 10 September creating up to about 9.5m potential new shares against roughly 22m diluted — after which the stock fell 17.5% in two sessions. At 58.6x trailing and 57.8x forward earnings on consensus 2026 earnings per share set to fall 43%, the multiple has barely moved from the 53x recorded in May; falling earnings did the work the price fall would otherwise have done.

Cameco: pounds already sold

Cameco, the Saskatoon miner that also owns 49% of Westinghouse, is up 2.8% over twelve months. Its slide dates from late August. On the uranium leg it raised 2026 realized-price guidance to $91–$96 a pound from $85–$89 and reaffirmed 19.5–21.5 million pounds of production, against commitments averaging about 28 million pounds a year through 2030. "Those old contracts are showing that upward leverage to the market that we said they would," chief executive Tim Gitzel told analysts on the second-quarter call.

The 92% fall in second-quarter net income came from the reactor side: its share of Westinghouse adjusted earnings before interest, tax, depreciation and amortization dropped to US$163m from US$352m, against a year-earlier quarter carrying the Dukovany construction project, and Cameco scrapped its five-year Westinghouse outlook in favor of current-year guidance only. Price to gross profit has risen to 56.6x from about 38x in May because gross profit fell faster than the shares. The live counter-anchor is a listing: Westinghouse is reported to be seeking above $50bn and could file as soon as October, which would mark Cameco's stake above $24.5bn against its own $38.4bn.

BWXT: the one nothing happened to

BWX Technologies, sole supplier of naval reactors and fuel to the US Navy's propulsion program and a maker of steam generators and medical isotopes, raised all four 2026 guidance lines on 3 August — adjusted earnings per share to $4.70–$4.80 — with backlog at $8.4bn, up 40%. Revenue grew 18% in the June quarter. Its one blemish is mix: operating income has fallen year on year for four straight quarters and operating margin slipped to 10.0% from 13.4% as cost-reimbursable work grew. "The industry is in the early stages of a multi-decade super cycle of growth," chief executive Rex Geveden said on the same call. The shares are down 21% over twelve months and 32% from March, and the forward multiple is 29.2x against roughly 46–47x in May, on consensus earnings growth of 24%.

What the decline earns

Centrus earns its de-rating: unit costs rising into falling volumes, the government leg reduced to caretaking, and dilution priced at a discount are contract economics and financing, not a demand failure. Cameco's is mostly arithmetic on a comparable quarter plus the removal of a five-year forecast, with its mined product selling for more. BWXT's is unexplained by anything it has disclosed. All three fell together between 8 and 25 September, alongside the Global X Uranium ETF's 14.7% loss, in a window when the 10-year Treasury yield crossed 5% for the first time since 2023 and investors repriced AI capital intensity — which is why the smaller reactor developers, NuScale down 78% and Oklo down 68% over the year, led and the fuel names followed.

The useful conclusion is that "nuclear fuel" is not one exposure. Kazatomprom is holding output below even its own reduced ceiling and says no spot price would make it accelerate, which keeps the pound scarce and Cameco's escalators working regardless of what a cascade in Ohio does. On Wednesday that cascade stops being paid to run at all, and the commercial contract meant to replace the government one does not owe a delivery until 2032.

A totalled car only reaches an auction lane if somebody crashes it first, and Americans are crashing fewer of them. Copart, which auctions written-off vehicles on consignment for insurers and collects a fee from both the insurer and the buyer, closed its fiscal year on 31 July having sold 5.5% fewer vehicles worldwide than a year earlier — revenue of $4.666bn, up 0.4%, after three consecutive years of roughly 10% growth. Over the same stretch the owner of its only scaled competitor sold more cars, not fewer.

That is the fresh ground since Copart's $1.9bn purchase of ACV Auctions was announced on 10 September. Copart and Insurance Auto Auctions, owned by RB Global, together handle over 80% of the North American salvage market, split roughly 65/35 in Copart's favor after decades nearer an even share. Both are paid per damaged car, but on different meters: Copart bills a fee schedule per unit, RB Global a percentage of the gross value transacted. The pool they divide is set by two things neither controls — how many crashes happen, and how often an insurer writes the car off instead of repairing it.

The pool is shrinking for the unglamorous reason

The write-off decision has never been more favorable. CCC Intelligent Solutions put industry total-loss frequency at a record 23.1% of claims for 2025, and Copart cited 23.3% for the second calendar quarter of 2026, its highest second quarter on record, with collision severity up 8.8% year on year and repair costs half again above 2019 levels. Cars that once got fixed now get auctioned.

What vanished is the accidents. CCC found repairable claim volume fell 9.7% in 2025. Copart told investors miles travelled rose 0.27% while the US vehicle fleet grew 1.6%. A record share of a smaller number is still a smaller number.

Copart lost units and raised price

Domestic insurance volumes fell 7.5% in Copart's July quarter and 8% for the year, a fourth straight quarterly decline. Management's explanation was one account: excluding the loss of a single large insurance customer, domestic insurance assignments would have risen 2.3%. Price did the rest of the work — revenue per unit rose 5.7% for the year and average selling prices 5.5%.

Underneath, the mix moved the wrong way. High-margin service revenue grew $13m, or 1.4%, while lower-margin purchased-vehicle sales grew 8.3%. July-quarter gross margin came in at 41.8% against 45.3%, operating income fell 2.6% for the year, and operating expense per car rose 12.7% — spending management ties to a long-haul delivery service, title processing and 25 dedicated wholesale facilities. Copart did not retrench: capital expenditure on land runs near $500m a year, it repurchased more than 43m shares for over $1.6bn, and it paid cash for ACV with no debt. "Every car that is not damaged or lightly damaged is gonna be put in front of thousands of dealers," chief executive A. Jayson Adair — back in the job since 31 July — told investors on 10 September. "That alone gets me excited." International units rose 10% in the quarter and dealer and fleet channels inflected upward.

At $27.59 the shares trade at 17.7x trailing and 16.9x forward earnings, against roughly 28x a year ago on an earnings base that fell 2.5%. Consensus has fiscal 2027 earnings per share at $1.63, up 2.9%. HSBC's Wesley Brooks cut the stock to hold on 15 September, trimming his target to $36, expressly contrasting Copart's volumes with RB Global's.

RB Global bought the units, and got marked down for it

RB Global sold 658,800 automotive lots in the June quarter, up 11%, a sixth consecutive quarter of growing faster than the market, after extending its relationship with its largest insurance partner to all 50 states. Trade coverage describes that partner as Progressive and puts IAA's share of its salvage volume at roughly 90%, up from about 75%, won on cycle times and commercial terms. Group revenue rose 11.1% to $1.317bn, operating income 19.2%, net income 30.7%, and guidance went up.

The shares fell 13.2% the next session and are down 28.6% in three months, to a 52-week low on 9 September. The objection is visible in the filing: service revenue grew only 5% to $933.4m while the value transacted grew 11%, because the take rate fell 110 basis points to 20.0% on acquisition mix and automotive pricing incentives. "Trajectory is what matters," chief executive Jim Kessler said on 4 August. Its heavy-equipment book grew only through acquisitions, organic volumes down on cautious customers. Forward earnings of 19.3x and 13.75x trailing enterprise value to earnings before interest, tax, depreciation and amortisation sit against consensus growth above 10% in each of the next two years; the trailing price-to-earnings figure of 35x is inflated by IAA purchase accounting.

What the two prices now assert

Copart's de-rating is almost entirely multiple, and defensible: a business that grows units is worth more than one whose revenue line depends on fee increases and a customer it no longer serves. RB Global's is not a reaction to results — it is the market deciding that units bought with discounts are worth less than units held at full price, which is an argument about the next contract renewal rather than the last quarter. Both cannot be a share-shift story alone; the shrinking accident pool takes something from each, and the record write-off rate that has been offsetting it has limited room left to rise.

The other prop just slipped. The Manheim used-vehicle index fell to 206.2 in early September, down 0.4% from a year earlier — the first negative annual comparison of 2026. Selling price is the multiplier under both companies' fees, and for the first time this year it is working against them.

Dell's infrastructure business shipped a record $31.8bn of servers, storage and networking in the July quarter, and the margin on it went up. Operating margin in the Infrastructure Solutions Group reached 15%, 620 basis points better than a year earlier, while revenue from AI-optimized servers doubled to $16.4bn.

That combination matters beyond one quarter. The working assumption about AI servers is that the assembler is a toll booth — it buys somebody else's accelerator, integrates it into a chassis, and hands the economics to the silicon and memory makers. Dell's raised guidance argues the opposite. The company, which sells branded PowerEdge systems, storage and services through an enterprise channel and takes allocation priority on accelerator shipments, lifted its fiscal year by $25bn to $192bn of revenue and $25.50 of earnings per share, with AI server revenue of about $74bn — roughly three times the prior year — and infrastructure margin still expanding more than a point on that mix.

"We exited the quarter with a record $95 billion of AI backlog, and our pipeline continued to grow sequentially and remains multiples of our backlog even after converting $131.7 billion into orders over the past 12 months," chief operating officer Jeff Clarke told investors on the September 1 call. Orders in the quarter were $60.9bn. The AI customer base passed 6,500 accounts, 3,300 of them added in three quarters, which is a broader book than the handful of accelerator-rental startups the bear case assumes.

The bought-in component went up and got repriced

Management named memory as the binding constraint — dynamic random-access memory and NAND flash first, then processors, disk and optical — and attributed the guidance raise to redirecting component supply from PCs into infrastructure rather than to any change in demand. Memory is the largest bought-in item in an AI server, and its cost roughly doubled through the first half of 2026; Micron's data-center revenue rose more than sevenfold to $11.5bn in its May quarter. Dell absorbed that and widened gross margin anyway. The differentiator among assemblers this year has been quoting terms and repricing speed, not end demand.

The cash has lagged the profit. Dell's 10-Q for the quarter shows receivables of $22.9bn against $17.6bn in January, growth slower than first-half sales; inventories rose $4.7bn over six months while payables rose $11.7bn, so suppliers financed the build twice over. The operating cash figures disagree — the filing indicates $4.1bn for the half, management cited $6.3bn on the call — and only the direction, cash trailing earnings, is agreed.

Dell trades at 32.2x trailing and 21.7x forward earnings, and its price against trailing gross profit has more than doubled this year, to 12.5x from 5.85x on May 3. Consensus has revenue up about 74% this fiscal year and then 21% the next: the forward figure is paid against one exceptional year. At $562.90 the shares sit essentially on the average analyst target near $564, with Citi and Bank of America at $600.

Paid a conversion fee for the same rack

Celestica, the Toronto builder that assembles and increasingly designs racks, switches and interconnect for hyperscalers, is the awkward case. Its growth has accelerated for four straight quarters, to 62% in the June quarter, and it raised 2026 guidance for the third time in six months, to $20.5bn of revenue and $11.30 of adjusted earnings per share. Gross margin still slipped, to 12.29%, because its own-design Hardware Platform Solutions line grew 58% while the assembly-heavy Connectivity and Cloud segment around it grew 84%. Three customers are 63% of revenue. The ramp was funded with equity: about 11.1m shares priced at $310 in August, near a tenth of the diluted count. The shares are flat over three months and cheaper than in spring against gross profit, 22.5x trailing versus 26.1x on May 3, after a 31.6% climb off a September 2 low with no earnings inside the window. The investor day is October 27, with third-quarter results.

Sanmina is the cleanest test of who owns the accelerator dollar. Having bought ZT Systems' manufacturing business from AMD for $2.25bn plus consideration, the contract manufacturer saw June-quarter revenue fall 13.7% sequentially to $3.464bn — and gross profit rise 2.7%, with margin up 167 basis points to 10.49%. The dollars that left were the silicon, not the labor. Guidance puts ZT revenue at $0.8–1.0bn in the September quarter on timing shifts in legacy programs, with next-generation accelerated compute deferred to fiscal 2027. It is the cheapest of the three at 16.1x forward earnings against 39.4x trailing.

What the money is actually buying

So the AI assembly dollar is being paid better, not thinner, and the evidence runs in both directions across the group. Where an assembler resells accelerators it earns little on them, which is why Sanmina's margin rose as that content shrank; where it converts, integrates and reprices, the margin holds. Dell earns its advance on accelerating revenue, expanding segment margin and a backlog with a date on it. Celestica's business earns an advance its price has not delivered for three months; the likelier reading of the September round trip is positioning, given that industrial-side AI names decoupled from semiconductors without any change in order books.

The contrast that settles it sits at the edges. NetApp, in storage, has re-rated to 8.2x trailing gross profit from 4.6x in May while guiding fiscal 2027 gross margin down to 68.5–69.5% from 71.3% on component costs — a higher price for a thinner margin. Jabil, the most diversified and least rack-concentrated of the large contract manufacturers, has fallen 15.5% in three months and reports its fiscal fourth quarter on September 30. Being near the rack is doing the work; being near the enclosure is not.

Dell's own forecast is the claim to hold management to: AI revenue roughly tripling and infrastructure margin wider at the end of it. One of those two has never happened before at this scale.