DK Street Journal

For three years the fear in the music business was that streaming platforms would grind down what they pay for a song. Warner Music Group — the New York company behind Atlantic and Warner Records on one side and a publishing catalogue of roughly a million compositions on the other — reported the reverse. Acting chief financial officer Lou Dickler decomposed the June quarter for investors on the August 6 call: subscriber growth contributed six to seven percentage points of subscription-streaming growth, pricing about 3.5 points, market share about one.

That distinction matters well beyond one quarter, because the per-stream rate is the coupon on an asset class. Catalogues change hands at a multiple of the publisher's share of the royalties they produce, which makes them long-duration income streams priced against risk-free yields — an 18x multiple is defensible when the ten-year Treasury pays under 2% and indefensible at 4-5%. Owners of songs therefore live off two numbers: the royalty rate and the cost of money. The first is rising. The second rose faster.

The rate is going up, contractually

Spotify's US effective rate to rights holders reached $4.43 per 1,000 streams in January 2026, up 34% from the spring 2023 trough, as the individual premium tier climbed to $12.99 and a minimum-stream threshold reallocated money away from the long tail. Warner has been hard-wiring that in: chief executive Robert Kyncl said per-subscriber-minimum pricing clauses now cover 88% of the company's subscription streaming revenue, against none two years ago, with a renewed Apple deal completing the set. Its Spotify agreement also overrides the bundling structure that had cut US mechanical royalties to publishers.

The accounts follow. June-quarter revenue of $1,864m was up 10.4%, the third consecutive double-digit quarter, and operating income rose 80.5% to $305m for a 16.4% margin against 10.0% a year earlier — helped by a restructuring programme on track for $200m of savings this year. Publishing revenue grew 12% against recorded music's 10%, with publishing adjusted cash operating earnings up 14% at an expanding margin. Whatever is wrong here, it is not that the statutory side lags the negotiated side.

The shares disagree. Warner is down 21.7% over twelve months and trades at 14.9x forward earnings against 20.8x trailing, with enterprise value at 11.9x cash earnings, while consensus has fiscal 2026 earnings per share rising 44%. A forward multiple that far below trailing is the market declining to underwrite the inflection the company is reporting. The visible catalyst is sell-side: BofA Securities' Jessica Reif Ehrlich cut the stock to Underperform and her target to $30 from $33, against a consensus average near $38.

The cost of owning the songs

Reservoir Media, a 100-employee rights buyer that acquires publishing catalogues and signs songwriters, is where the mechanism bites. Its net publisher share grew from $42m in fiscal 2022 to $68m in fiscal 2026, a 62% gain; interest expense over the same span went from $10.9m to $26.5m, up 143%. Chief financial officer Jim Heindlmeyer said the latest increase, to $6.9m from $6.3m, was "driven primarily by increased debt balances used to fund acquisitions of music catalogs." That $6.9m consumed roughly half of the quarter's $13.7m of cash operating earnings.

So revenue up 11.6% to $41.5m — only six points of it organic — produced operating income down 1.3% and a small net loss. Debt ended fiscal 2026 at $455.7m against $98.9m of liquidity a quarter later. Meanwhile the asset itself has cheapened: catalogue prices that peaked at 18-25x net publisher's share in 2021 now clear around 12-18x. Buying cheaper assets with dearer money is a narrowing spread, and it is the whole model.

Reservoir is also in a takeover contest. Irenic Capital bid $10.00-$11.00 a share in February; Wesbild and Richmond Hill countered at $10.50 in cash. The October 2 close of $9.52 sits 9.3% below that standing proposal, which is the market assigning real odds to no deal. Any transaction needs Wesbild, holder of about 44% and controlled by the chief executive's father.

Nobody is listening less

The payer side rules out softening demand. Sirius XM, which licenses this music rather than owning it, added 22,000 self-pay subscribers in the June quarter — a 90,000 swing from a year earlier — at record-low 1.4% churn, with advertising up 5% and full-year guidance raised. Revenue grew 1.0% and operating income 29.3%; the shares trade at 0.73x book and a 17.9% trailing free-cash-flow yield. Sirius supplied most of the group's thirty-day decline all the same, giving back a run that had reached roughly 43% by late July. Spotify fell 15.6% over the same month and Live Nation 6.0%, which points to broad risk reduction across music equities rather than rights holders losing rate.

The verdict splits cleanly. Reservoir's markdown is earned — not by royalties, which are growing, but by a capital structure in which financing costs compounded at more than twice the rate of the income they bought, with a cash bid now acting as the floor. Warner's is not earned by anything in its filings: rate, mix, margin and publishing all moved the right way, and the explanation on offer is a downgrade. Rate compression was the wrong thing to watch.

The right thing arrives this week. The Copyright Royalty Board has split off interactive streaming mechanicals for 2028-2032 and told participants who cannot agree to file written direct statements by October 5, while objectors fight a proposed settlement that would reset the base mechanical rate to 12 cents from today's 13.1. The negotiated rate is finally working for the owners of songs. The statutory one is still being argued.

Where the profit comes from

Most of Ormat Technologies' first-half profit did not come from selling electricity. The company — which owns roughly 1,355 megawatts of geothermal, solar and recovered-energy plants and separately manufactures the power blocks that rival developers buy — reported $71.2m of net income for the six months to June. Inside the income-tax line of its quarterly filing sit $33.2m of income from the sale of tax benefits and $32.5m of transferable investment tax credits: $65.7m together, about 92% of the reported figure, on a comparison Ormat does not itself present.

That is what the phrase "firm clean baseload" looks like on an income statement. A geothermal plant sells output under contracts running two decades at an average realized price of $92.6 per megawatt-hour in 2025, down from $94.3 the year before. The reservoir underneath it declines whether or not demand grows. So the reported earnings arrive largely through the tax code, and the cash goes back into the ground.

The field decays on its own schedule

A Stanford Geothermal Workshop case study of Kenya's Olkaria East field — part of the complex where Ormat operates plants — records reservoir pressure falling 2–4% a year and natural steam decline of 2–3% annually, with production held flat only after make-up wells were connected. More than 50 re-injection wells have been drilled there.

That physics shows up as capital spending behaving like a second depreciation charge. Ormat's rest-of-2026 program runs to about $587m, including $139m of exploration, drilling and development and $36m of maintenance capital on plants already operating. Roughly $436m of it lands in the Electricity segment, against guided Electricity revenue of $710–725m for the year. The trailing free-cash-flow yield is minus 4.6%. In 2024 the consolidated fleet generated 7,450,071 megawatt-hours — about 850 megawatts of continuous output, indicatively, from a portfolio now near 1,355 megawatts including solar and recovered-energy assets.

What actually grew

Second-quarter revenue rose 10.6% to $258.8m and gross margin widened to 26.5%, but the geothermal line grew 5.8%, to $169.3m, and generation 3%. The Electricity segment "built on its growth momentum during the quarter, driven by contributions from our Blue Mountain geothermal power plant acquired in June 2025, improved performance at our Olkaria and Puna power plants, and lower curtailments in the USA," chief executive Doron Blachar told investors on the second-quarter call. Blue Mountain was bought, not drilled.

The acceleration came from merchant batteries, where revenue nearly tripled to $42.8m at a 56.2% gross margin — a margin management guided down to 30–40% in the second half. Over five years consolidated gross margin has compressed from 39.9% to 27.6%, and reported net income has sat near $124m for three straight years.

Fervo, from the drilling end

Fervo Energy, the Houston enhanced-geothermal developer that listed on Nasdaq in May, is the same equation inverted: its entire value is making the drilling cheap enough to finance. Cost across the first four horizontal wells at Cape Station in Utah fell from $9.4m to $4.8m, and spud-to-rig-release from 70 days in Nevada to an average of 21, per CNBC. The first 33-megawatt block reached commercial operation on 30 September, a day inside its contractual deadline, against a binding offtake book of about 1,054 megawatts. The June quarter carried $113,000 of revenue and a $59.5m net loss. With $2.1bn of cash against a $4.15bn market value, 1.52x book is the only anchor that means anything; management has guided 2027 revenue to a deliberately wide $60–80m because of transmission curtailment it says is unique to that year.

What the de-rating earns

Ormat fell 11.3% in the month to 2 October and sits 36% below its May high, though only 7.1% lower over twelve months; Fervo fell 19.6%. Contracted generation sold off together — XPLR Infrastructure down 12.7%, Constellation 11.2%, NextEra 7.5% — as Treasury yields kept climbing through September. Jefferies cut its target to $104 while raising its 2030 estimates, citing "mark-to-market adjustments against a higher cost of equity". JPMorgan trimmed on "a slightly higher mix of its fiscal 2027 estimated pro forma EBITDA derived from tax credits".

Most of the fall is earned. At 44.6x trailing and 36.5x forward earnings, down from 56.1x trailing earlier this year, Ormat is still priced richly against consensus that has 2027 earnings per share falling 4.2% to $2.44 and against cash consumption after capital spending. The discount rate explains the timing; profit composition explains the level. What nothing yet explains is the gap between the axes — management's 2030 target of $1.0–1.1bn of EBITDA sits a fifth above consensus and was raised into a falling share price. The test is not capacity added but whether Electricity gross margin climbs back toward the 40% that plan assumes, and whether credit buyers keep paying.

Ormat's first contract with a data-center operator — 20 years, about 13 megawatts to Switch from the Salt Wells plant in Nevada — starts delivering in the first quarter of 2030. The scarcity everyone is bidding for shows up, here, in four years.

Upstart Holdings, which runs an artificial-intelligence underwriting platform for banks and credit unions and holds no bank charter of its own, spent this year lining up people to buy the loans its models approve. It has announced roughly $7.45bn of forward purchase commitments during 2026 — up to $4bn from Castlelake alone, plus Fortress, Centerbridge and a joint Eltura Ventures and Aperture Investors facility — and in late September priced a $400m consumer-loan securitization, its 52nd. Its shares are down 56.9% over twelve months and 19.0% in September alone.

For any lender that does not take deposits, the business question is whether somebody will buy the loan and at what price. Through the third quarter the answer improved. About $255bn of consumer asset-backed paper had priced, roughly 16% ahead of 2025 at the same point, and spreads on top-rated three-year prime auto bonds stood at 37 basis points over benchmarks on 24 September, three tighter than a month earlier. Truist Securities expects issuance up another 10% next year because issuers can borrow on better terms.

What moved was the rate underneath all of it. The Federal Reserve raised its target range to 3.75%-4.00% on 16 September, its first hike since July 2023, and the ten-year Treasury rose to about 5.2% by the end of the month, roughly 45 basis points above where it sat in mid-September. On 23 September Upstart and Affirm each fell 4% while the broader financials group barely moved, with no company news at either. Affirm in fact spent that session launching instalment payments at Amazon's UK checkout — interest-free over three payments above £50, or up to 48 months at a 22% representative rate.

Who holds the paper

The base rate bites unequally. Affirm Holdings, whose button at checkout splits purchases over one to 48 months and which bills merchants for the service, chose to carry the risk: $9.56bn of loans held for investment at 30 June against $7.03bn a year earlier, funding debt doubled to $3.3bn, an average cost of funds near 5.8%. Revenue less transaction costs — the company's own gross-profit line — ran 4.2% of volume, and on the 27 August call it guided fiscal 2027 to 4.16%, crediting favorable debt capital markets. Thirty-day delinquencies on monthly instalment loans improved to 2.5%. "We are in control of the credit reality. You would see us slow down growth before you would see us have a real credit disturbance," founder and chief executive Max Levchin told investors that day.

The business behind it is not the problem. Affirm's fiscal 2026 operating income was $870.9m against a loss of $87.3m the year before; the June quarter's operating margin reached 24.3% versus 6.6%. The honest caveat is the top line: year-over-year growth slid from 33.6% to 24.3% across the four quarters, while consensus still carries fiscal 2027 revenue 30.6% above the prior year's estimate. Affirm trades at 6.34x forward gross profit against 8.31x trailing and roughly 9.6x trailing in mid-August; its trailing price-to-earnings ratio of 12.4x is meaningless, inflated by a deferred-tax valuation-allowance release of about $1.46bn.

Upstart's June quarter put revenue at $377.2m, up 42.3%, with operating income of $17.2m, more than tripled, and only $1.06bn of loans on its own books — 5.9% of outstandings. "We grew, our credit performed, and we expanded margins. We didn't have to trade one for another," chief executive Paul Gu said on the 4 August call. Its shares now fetch 1.61x forward gross profit, down from 2.55x a month earlier, and 2.76x book. Two things argue for caution anyway: the company's own macro index has been stuck at 1.50, implying defaults around half again what a normal economy would produce, and consensus modelled $1.671 of 2025 earnings per share against the $0.45 delivered.

The deposit-funded case

Nu Holdings, the Brazil-based digital bank with retail operations in Mexico and Colombia, pays for its loans with customer money: $45.3bn of deposits at 88% of interbank rates, three percentage points cheaper than a year earlier, with net interest margin up 180 basis points to 22.9%. Revenue grew 49.9% and net income 66.5% in the June quarter. Early-stage delinquencies improved 16 basis points to 4.8% while loans more than 90 days overdue rose 35 basis points to 6.9%, which the company tied to seasonal migration from the first quarter. Its drop has nothing to do with any of that: Sky News reported on 26 September that Monzo was in talks over a sale to Nubank at £8bn-£10bn, about $11bn-$13bn, and Nu shed roughly $6.6bn of market value on 28 September — a fall of about 9% to 10%, depending on the closing price used — on a day the broad American market slipped under 1%. It trades at 15.4x forward earnings and 4.92x book on a business earning near 30% on equity.

So the September repricing divides cleanly. Affirm's is partly earned: it holds the paper, it borrowed to do so, and the coupon on that borrowing just went up — rational repricing of a changed input, and the reason its guided gross-profit margin is the number to watch rather than volume. Nu's discount prices an acquisition nobody has quantified. Upstart's is the one that follows from nothing in its own accounts; with buyers committing billions and spreads tightening, the likelier reading is that rate-sensitive consumer lenders were sold as a group.

Levchin's promise is that Affirm can slow growth before credit breaks. The cost it cannot slow is the interest on the $3.3bn it has already borrowed to keep the loans.

Twilio told investors in early August that its revenue this year would grow several points faster than its gross profit, and said exactly why: the fees American mobile carriers charge to deliver application-to-person text messages. The company collects those fees and hands them on. Its own definition of organic revenue, in the appendix of the deck accompanying its second-quarter results on 6 August, excludes "revenue from incremental increases to application-to-person (A2P) fees imposed by major U.S. carriers on our core messaging business", adding that "we pass these fees to our messaging customers at cost."

What has happened since is the question. Twilio — a cloud communications platform of roughly 1,800 programming interfaces that lets developers embed messaging, voice and verification inside their own software, with a meter that runs on messages, minutes and verifications — has disclosed no financial figure since that August date, and its shares have added 29.9% in the month to 2 October, reaching $294.58 and a six-month gain of 125%. Braze, which sells the orchestration layer that decides which message goes out and when, reported a beat and a raised full-year guide on 8 September and fell 18.4% the next session. The two meter the same thing. The stake is whether a dollar of this revenue is message volume from artificial-intelligence agents, as the repricing assumes, or a carrier price increase passing through.

The bill inside the top line

US operators raised those pass-through charges three times this year — T-Mobile and US Cellular in January, AT&T in April, Verizon in May. Twilio absorbed $71m of incremental carrier fees in the June quarter, which cut non-GAAP gross margin to 49.1%, and its full-year outlook assumes about $250m of such revenue, a drag of roughly 210 basis points on gross margin against 2025, according to the August call. It shows up most plainly in the messaging line: reported growth of 28% becomes about 18% once the fees come out.

Underneath it, the business is accelerating. Revenue growth went 14.3%, then 20.0%, then 22.0% across the last three quarters, to $1,499.1m; gross profit growth went 10.4%, 17.7%, 20.4%, trailing revenue each time by between 1.6 and 3.9 points. GAAP operating income more than doubled to $84.5m, free cash flow reached $353m, and the dollar-based net expansion rate climbed to 116% from around 108% a year earlier. "We are in a powerful new chapter at Twilio, marked by another quarter of organic growth acceleration as well as record profitability and free cash flow," chief executive Khozema Shipchandler said with the results. Two tests a skeptic would want are unavailable: Twilio stopped publishing active customer accounts from the first quarter of this fiscal year, and it does not split net expansion into volume and price.

The company's own guide is the caution. Third-quarter organic growth is set at 11% to 12%, down from 17% posted in June, and full-year gross profit growth at 13% to 13.5% against reported revenue growth of 18% to 18.5%.

A 27% re-rating since August, on no numbers

Twilio's 50-day average has sat above its 200-day since 17 April. On price to trailing gross profit — the only comparable measure here, since a one-off tax item put $1,067.2m of net income against $84.5m of operating income last quarter — the shares are at 16.53x, and 15.35x forward, against 12.97x in late August. Forward earnings are 49.7x against consensus revenue growth of 11.4% in 2027. The 27-analyst average target of $249.68 sits about 15% below the price, and HSBC's Sameer Lam cut the stock to Reduce on 25 September with a $211 target, arguing that rising agent activity "does not necessarily translate into a larger share of the resulting revenue for Twilio". The likelier reading of the September move is narrative: Meta launched its consumer agent Muse on 8 September, built so that talking to it feels like messaging a person.

The same bill, one layer up

Braze, founded in 2011 and run by chief executive William Magnuson, is not insulated from any of this. Its fiscal 2026 annual report states that pricing "is based in large part on the number of consumers that customers reach and the volume of messages customers send", and premium messaging volumes are the named reason its non-GAAP gross margin fell 70 basis points year on year to 68.6%. Gross profit grew 24.6% against revenue up 26.2% to $227.2m — almost exactly Twilio's shape. Everything else improved: non-GAAP operating margin reached 9.7% from 3.4%, free cash flow hit a second-quarter record $22m, customers above $500,000 of annual recurring revenue rose 28% to 361 with net retention among those accounts at 112%, and full-year guidance went up to $910m–913m. What broke the stock was a third-quarter revenue guide implying about 20% growth and adjusted earnings of $0.13–$0.14 against roughly $0.16 expected. At 5.18x trailing gross profit, down from 7.43x in late August, Braze has surrendered its entire summer re-rating.

"For every one example that I have where people are chasing their vibe coding dreams, we've got 10 more for whom Braze is their AI bet," Magnuson told analysts on 8 September. At its Forge conference at the end of the month the company launched Decisioning Studio Go, an agent handed an audience and guardrails that then runs its own experiments, with general availability expected 14 October.

Nothing resembling a category turn is visible around either name. Over the month to 2 October, HubSpot — the seat-priced comparison, selling marketing, sales and service software to mid-market businesses, and cheaper per gross-profit dollar than either at 3.83x trailing — fell 12.6% after adding 7,000 net customers against an expected 9,000 to 10,000 and cutting its second-half target. Salesforce, at 5.66x trailing gross profit and growing 10.8%, fell 8.7% even as Agentforce reached $1.5bn of annual recurring revenue. Klaviyo fell 18.6%, Freshworks 3.1%. The front office has been software's worst non-recessionary drawdown in more than 30 years, and the recovery has sorted companies by how they bill.

The verdict splits cleanly. Twilio's business earns the first leg of this move — three quarters of accelerating growth, doubled operating income, record cash — but nothing it has disclosed explains the multiple expansion since August, and its own guide says gross profit grows five points slower than the revenue line being extrapolated. Braze's de-rating is explained by neither its gross profit, its retention, nor its largest customers; it is explained by three cents of quarterly earnings guidance. Both own the same exposure, which is the part the agent story misses: the carrier fee sits in both revenue lines, so buying the message meter means buying part of Verizon's price list.

Shipchandler told investors in August that fewer than 6% of voice interactions are machine-driven. If that changes, the traffic arrives on rails whose owners have already shown they will raise the toll three times in a year.

M/I Homes builds single-family houses and townhomes across nine states from Ohio to Texas, and it owns the lender that finances them. In the June quarter that lender wrote the loan for 96% of the company's buyers, a record. It also earned less money than it did a year earlier.

That is the finding, and it kills a convenient story about how production builders are surviving 7% mortgages. The discount a buyer needs today is not taken off the sticker; it is spent on forward-commitment points that buy the mortgage rate down, and that cost is booked in cost of sales. The captive lender is supposed to recapture some of it. At M/I Homes, financial services pre-tax income fell to $28.5m in the first half from $30.6m, with segment revenue up 1% to $63.6m as the higher capture rate and extra originations were offset by smaller average loans. Consolidated operating income over the same half fell $106.6m, to $187.3m. The subsidy covered about 2% of the damage.

The margin stopped falling anyway

Strip the lender out and M/I Homes' homebuilding line is the most encouraging in the group. Gross margin was 22.0% in the first quarter and 22.1% in the second, after a trough of 18.1% in the fourth quarter of 2025 and a reading above 25% in early 2025. Second-quarter revenue fell 8.5% to $1.06bn while gross profit fell 17.8% — the squeeze is in price realization, not units. Orders were a second-quarter record, up 15% to 2,387 homes on a flat 234 communities, with monthly pace at 3.4 homes per community from 3.0 and cancellations at 8% against 13%. Backlog carried an average sales price of $538,000. The list price is holding; the concession is hidden in the financing.

"We continue to use mortgage rate buydowns as our primary incentive, and given the current rate environment, will continue to promote with such buydowns for the foreseeable future," chairman and chief executive Robert Schottenstein said on the July 29 call. About 78% of second-quarter orders were specs — finished or in-progress houses sold off the shelf — a bet the rest of the industry is backing away from. Schottenstein put a price on it: in nearly every one of the company's 17 markets a to-be-built home carries a better margin than a spec, "in some, just slightly; in others, it could be 100 or 200 basis points." The buydowns are priced to a payment, advertised slightly below 5% on government programs, against a 30-year survey rate Freddie Mac put at 7.28% on October 1.

The stress case, and the control

Dream Finders Homes, a Jacksonville builder of starter and move-up houses that brokers its mortgages to PrimeLending rather than consolidating a lender, shows what the same bill does to a thinner balance sheet: homebuilding gross margin of 14.2% against 16.5%, lots controlled down to 54,091 from 63,121 at year-end, and $3.0m of lot-deposit and abandonment charges — options walked away from rather than exercised. It trades at 6.8x trailing earnings and 10.1x forward, because consensus for 2026 has halved to $1.07 a share and 2027 has been cut to $0.79; free cash flow is negative. Into that it is closing a $2.2bn all-cash purchase of Beazer Homes, with Beazer holders voting October 15 and $350m of notes being redeemed October 20.

KB Home, which builds to order rather than on spec, is the test of whether any of this is margin-specific. Its third quarter showed revenue down 20% to $1.30bn and housing gross margin of 16.5% against 18.2%, with backlog rising for the first time in four years — and full-year margin guidance still cut to 16.0-16.2%. Across the industry, 66% of builders reported using incentives in September, up from 63%. The bill is still growing.

What the shares are paying for

M/I Homes closed at $134.76 on October 2, down 8.2% over thirty days and 14.1% over three months, but up 10.8% over six — the best six-month return of the ten listed builders checked. Every builder fell over that month, from Dream Finders at -19.7% to Lennar at -4.5%, and no company announcement dated inside the window was discoverable for M/I Homes, so the likelier reading is rates and the September slide in builder confidence rather than anything at the company. It trades at 11.1x trailing earnings and 11.2x forward — the market paying for no growth at all — and at 1.07x book, dearer than Meritage at 0.85x, KB Home at 0.78x and Dream Finders at 0.62x.

So the trough is real and the subsidy is not. M/I Homes has earned two flat margin quarters with better orders and fewer cancellations, which is more than anyone else on this shelf can show; what it has not earned is the idea that owning the mortgage company pays for the buydown. That makes the next print a question about one number — incentive load — and not about demand, which is arriving. Dream Finders' de-rating, by contrast, is its own arithmetic, and it is about to add Beazer's leverage to it.

Two of this shelf's names stopped trading this year because strategic buyers wanted the cash flow at these margins: Berkshire Hathaway closed its $72.50-a-share purchase of Taylor Morrison on July 24, and Sumitomo Forestry took out Tri Pointe at $47 in May. The six still listed are being marked down for selling the same houses.

Sea Limited's lending business earned more last quarter than the marketplace it was built to finance. Monee — the Singapore group's consumer and small-business credit arm, which includes the SPayLater installment product and SeaBank — turned in $288.0m of adjusted earnings before interest, taxes, depreciation and amortization in the three months to June, against $255.4m from Shopee, the mobile marketplace that is the company's public face. Garena, the games studio, contributed $429.8m of the $917.2m group total.

That ordering is worth pausing on, because it exposes how little the phrase "e-commerce revenue" settles. Shopee is paid a take rate: it moved $38.3bn of merchandise on 4.2bn orders in the quarter, an average ticket of about $9, and booked $4.3bn of core marketplace revenue from commissions and advertising on it — roughly 11% of what crossed the platform, against merchandise value up 28%. The gap between those two growth rates is the whole story of the quarter: advertising revenue grew more than 70%, the ad take rate widened by over 90 basis points, and the number of sellers buying ads rose about 45%.

Six cents an order

What Shopee keeps after costs is far thinner than what it captures. Its $255.4m of segment profit equals 0.7% of merchandise value — about six US cents per order. "With this solid momentum, we are optimistic that Shopee will achieve the milestone of $1 billion in adjusted EBITDA for the full year," chairman and chief executive Forrest Li told investors on August 11. Group revenue grew 48.1% in the quarter and gross margin held at 45.6%, but operating income rose only 28.4% — the reinvestment is visible.

Monee's arithmetic is the opposite shape. Principal outstanding grew 62% while segment profit grew 13%; the 90-day non-performing loan ratio held flat at 1.0%, so the absorbed money is provisioning and acquisition cost on unseasoned vintages rather than realized losses. An increasing share of Sea's incremental profit is a credit function, not a commerce one.

Coupang owns the goods

Coupang, the Korean retailer and marketplace run from Seattle by founder Bom Kim, books nearly the entire price of the inventory it owns, and the cost of that inventory beneath it. Hence a 28.2% gross margin, and hence no gross merchandise value and no take rate anywhere in its quarterly disclosure. What it publishes instead is customers and spend: 24.7m Product Commerce active customers, up 3%, generating $301 of net revenue each against $307 a year earlier.

The June quarter was ugly underneath. Product Commerce segment margin fell 390 basis points to 5.1%; the Developing Offerings arm (Eats, Play, fintech and Farfetch) lost $219m, only $16m better than a year ago; gross profit shrank 2.6% on revenue up 3.9%. A KRW 624.68bn ($410m) privacy fine over a breach of 33.7m customer records drove the $556m operating loss, and even excluding it the loss was $146m. "The vast majority of our customer spend never moved," Kim said on the August 4 call; management guided core margins back to pre-incident levels only by the middle of 2027. Korea's Fair Trade Commission sent roughly 30 investigators into its Seoul headquarters on September 1, and separately seeks a KRW 140bn penalty over search rankings.

MercadoLibre shows the mature version of the same combination. Revenue grew 49.8% to $10.2bn last quarter, but gross margin fell 467 basis points to 40.9% and operating income fell 17.2% as it subsidized shipping against Shopee in Brazil. Its credit book, at $14.6bn, carries over-90-day non-performing loans near 17.6% — a different animal from Monee's 1.0%.

What the prices say

All three fell 13-16% over the month to October 2, against an emerging-markets index down 0.7% in September. Coupang closed at $13.78, within a fraction of a percent of its 52-week low; no company-specific news explains Sea's slide, and the likelier reading there is profit-taking after August. Because gross margins run from 28% to 46%, the comparable yardstick is price against gross profit: Sea trades at 4.08x forward, below MercadoLibre's 4.84x, with Coupang cheapest at 2.36x. Sea's forward earnings multiple is 26.0x against 35.1x trailing; MercadoLibre's 44.5x forward barely undercuts its trailing 46.2x, because consensus models 2026 earnings per share down 5.7%.

The verdict the numbers support is narrow. Coupang's cheapness is a description of damage — consensus has it losing money at the EBITDA line this year — not of a mispricing. Sea's commerce business is genuinely compounding its take rate, but its profit is six cents an order, and the group's earnings momentum has migrated to a loan book growing five times faster than the income it throws off.

Shopee's promised $1bn profit year is now the cleanest test available: if it arrives, it arrives alongside a credit book that will be larger still, and the question of what Sea actually is will be harder to postpone than it was in August.

Every one of Aon's four business lines grew 5% organically in the June quarter, the Dublin-headquartered broker's adjusted operating margin widened, and the shares have spent the three months since falling to their lowest level in a year. Aon closed October 2 at $269.45 — exactly its 52-week low, and 29.3% below the $381.26 it traded at within the past twelve months.

The selling deepened rather than settled after mid-September, when Aon placed $13.5bn of bonds to fund its purchase of USI Insurance Services into roughly $65bn of orders and the equity kept sliding regardless. Over the thirty days to October 2, Aon fell 18.6% and Arthur J. Gallagher — the Rolling Meadows brokerage and claims administrator that places retail and wholesale property-casualty and employee-benefit cover and runs the Gallagher Bassett claims business — fell 14.5%. The next hard tests are dated: Marsh & McLennan reports on October 15, Gallagher on October 29 and Aon on October 30.

The premium is softening; the commission is not collapsing

A broker is paid a share of a premium it does not set, so the premium cycle is the revenue cycle. Marsh's own quarterly index put global commercial insurance rates down 6% in the June quarter, the eighth consecutive quarterly decline, on abundant capacity and strong insurer profitability. But the softness is concentrated: within that composite, property rates fell 12% while casualty rates rose 2% on US claims severity and litigation. Gallagher's own client renewals split the same way, property premiums down 6% and casualty up 5%.

That is why the commission meter has kept running. Aon's reported revenue rose only 2.0% to $4.2bn in the June quarter, but the company reconciles that to 5% organic growth, a point of favorable currency and four points of drag from selling the NFP Wealth business and Stroz Friedberg. Commercial Risk, Reinsurance, Health and Wealth each grew 5%; Reinsurance Solutions alone turned over $711m. Adjusted operating margin reached 28.9% against 28.2% a year earlier. On reported figures the margin went the other way, to 19.4% from 20.7%, which is amortization, restructuring and transaction costs rather than a thinner commission.

One line did shrink. Fiduciary investment income — the yield Aon earns on client premium it holds in trust between the insured and the underwriter — fell 12% to $58m, which the company attributes to "lower interest rates and average balances". The yield half of that explanation is now running backwards: the Federal Reserve raised its target to 3.75–4.00% in September, with futures pointing to about 4.1% by January. What remains is the balance half — a smaller premium pool to hold.

The self-inflicted part

Aon agreed on August 31 to buy USI, the tenth-largest US broker with roughly $3bn of revenue, from KKR and other holders for $17.0bn in cash, targeting $395m of annual run-rate synergies. Chief executive Gregory Case said the deal would "position Aon to accelerate organic growth" in the US middle market. It is funded entirely with debt, lifts pro-forma leverage to 4.3–4.5 times earnings before interest, tax, depreciation and amortization from 2.7 times, has suspended buybacks until that unwinds, and adds to adjusted earnings per share only in 2028. Aon paid 14.5 times earnings for USI; Gallagher paid 11.3 times for AssuredPartners.

Where the four trade

Aon's $269.45 is 14.5 times the $18.64 consensus for this year — a figure trimmed less than 1% since mid-September — against the 19.4 times its price a year ago implied on effectively the same estimate. It is also the cheapest of the four on enterprise value against cash earnings, at 11.4 times, a reading that predates the USI debt. Gallagher has fallen from 23.4 times forward earnings a year ago, and 26.7 times two years ago, to 17.2 times — and is still the group's dearest name on forward earnings, at 14.8 times on enterprise value and with a 3.94% free-cash-flow yield, the lowest of the four. Keefe, Bruyette & Woods cut its target to $250 from $271 on September 24, expecting "slowly-decelerating middle-market property and casualty pricing to constrain overall organic growth in the near term." Marsh & McLennan, with the largest consulting offset, de-rated least, to 17.7 times forward earnings; Willis Towers Watson, the smallest, is the only one of the four whose reported revenue is accelerating, up 9.1% in the June quarter.

What earns the fall and what does not

The twelve-month de-rating has a datable origin the operating numbers never touched. On February 9, OpenAI approved Insurify's app letting ChatGPT users pull personalized homeowners quotes, and Bank of America sized $15bn of low-complexity commissions as exposed to disintermediation. Willis fell 12.1% that session, Gallagher 9.9%, Aon 9.3%. Goldman Sachs called the move "disconnected from near-term fundamentals" and "overdone"; Barclays upgraded Gallagher in March with a $262 target, 13.7% above where the stock now sits.

So the fall divides cleanly. Gallagher's is a premium being normalized — it bought $13.45bn of revenue, diluted its share count 15.5% since 2024, and still commands the group's richest valuation on every measure. Aon's is harder to pin on the business: 5% organic growth across all four lines and a wider adjusted margin do not produce a 25% fall in the share price. Part of it Aon did to itself, trading a 2.7-times balance sheet and its buyback for an asset that pays back in 2028. The rest is a market pricing the terminal value of a commission rather than this year's. Nothing in the brokers' reported organic growth has yet confirmed it.

The first number comes from inside the group: Marsh reports on October 15, and with it the index that tells its three rivals what their commissions are a percentage of.

GitLab, which sells a single software-development platform — source control, build automation and security scanning — licensed per paid seat per tier, added paying accounts last quarter at barely a third of the rate its revenue grew. Customers worth more than $5,000 a year reached 11,114, up 8%, while revenue in the quarter ended July 31 rose 21.3% to $286.3m. Accounts above $100,000 grew 17%.

That gap is the whole question hanging over per-seat software: when a customer's engineering headcount stops growing, where does the next dollar come from? GitLab's answer has two parts, and only one of them is free. The first is tier uplift — Ultimate annual recurring revenue grew 35% and now accounts for 59% of the total, and deals above $500,000 jumped 150% year on year. The second is a meter. Paid consumption on the Duo Agent Platform grew about 50% sequentially to more than $40m of run rate, from $15m three months earlier, against a target above $100m by the end of this fiscal year.

"We started the year with a business monetized entirely through seat subscriptions," chief executive Bill Staples told investors on September 1. "And we now expect to exit it with seats continuing to grow alongside a meaningful and scaling consumption business." Bain, writing on the 2026 software selloff, describes the same pattern across the industry: companies moving past seats are "layering new meters on top of seats rather than replacing them outright".

The second meter has a cost of goods

Inference is a cash cost. GitLab's gross margin fell 3.9 percentage points to 84.1% from 87.9%, so gross profit grew 16.0% while revenue grew 21.3% — negative leverage at the gross line, in a quarter the sales force called its largest ever for gross bookings. Management attributes its 86.5% non-GAAP gross margin and its 85-87% full-year guide to cloud mix rather than AI adoption, and says Duo is deployed model-agnostically so token cost is not written into customer contracts. Both readings can be true; what is not in dispute is the direction.

Underneath, usage is running far ahead of seats: secure repositories up 60%, code pushes 50%, pipelines 40%. GitLab also launched Flex, a single dollar commitment a customer can spread across seats, credits and usage products without renegotiating — more than 130 customers committed over $20m in six weeks, which shifts some revenue recognition forward and sits outside current remaining performance obligations. Total obligations were about $1.2bn, with 65% due within twelve months. Dollar-based net retention of 117% accelerated for the first time since early fiscal 2024 — but GitLab's own definition folds seats, price and upgrades into one ratio, so the split the number would settle is not disclosed.

Atlassian defers the same bill

Atlassian — Jira, Confluence, Bitbucket, also sold per seat per edition — bundles its Rovo AI into higher editions and reports engagement for it, not revenue. Its June-quarter cloud revenue rose 31% and gross margin went the other way from GitLab's, up 3.8 percentage points to 86.9%. Chief financial officer James Chuong told the August 6 call the beat came from cross-sell and seat expansion rather than Data Center migrations — notable for a company whose shares fell 35% earlier in 2026 on its first reported decline in enterprise seats. The cost appears in guidance instead: fiscal 2027 revenue growth of about 13%, half the year just closed, non-GAAP operating margin down to 25% from 36%, and gross margin guided lower on rising Rovo usage. "Organizations can hire intelligence by the token," co-founder Mike Cannon-Brookes wrote on August 6. "Context is much harder for organizations to build."

What the prices now assume

Both shares have already made the move: Atlassian is up 175% over six months, GitLab 121%, and both were flat over the last thirty days while the September gain in this corner of software came from the chip-design tool vendors Synopsys and Cadence. GitLab trades at 7.97 times trailing sales against roughly 5.5 times in May, and 57.5 times forward earnings against consensus revenue growth of 19.5% this year slowing to 15.5% next. Atlassian is at 7.51 times trailing sales against roughly 3.3 times in May, and 34.2 times forward earnings on that 13% guide. Figma, growing revenue 48.2% with net dollar retention of 136%, fell 17% in thirty days and trades for less on forward sales than either — its gross margin dropped 5.2 points to 83.7% as inference landed in cost of revenue.

The verdict the numbers support is narrow. The seat meter is not being retired — the metered AI line is additive, as Bain says and as GitLab's bookings show — but it has stopped being the growth engine, and the replacement is sold at a lower gross margin or given away inside a more expensive edition. What re-rated this summer was pricing power, not developer demand. At 7.5 times sales on guided growth of 13% to 19%, both companies are now priced as though the upgrade cycle runs for years and the inference bill stays small.

Figma is the warning on the second point: fastest growth in the group, strongest expansion, and the only one whose multiple is still falling. The seat disclosure is what broke Atlassian in May. The line to watch now is the one labeled cost of revenue.

Liberty Energy, which rents hydraulic fracturing horsepower to North American oil and gas producers, ran its pumps harder in the June quarter than at any point in its history. It set records for pump hours, horsepower hours and pump utilization, earned $151m of adjusted earnings before interest, taxes, depreciation and amortization on $1.19bn of revenue, and reported operating income of $12.7m.

The gap between those two figures — roughly $138m for the quarter — is depreciation, and in pressure pumping it is not an abstraction. The record utilization that produces the cash is the same thing that consumes the fleet, and the charge is a standing bill for replacing iron. Add ProPetro's implied $50m and ProFrac's $107m, and the three biggest independent American pumpers generated about $265m of cash earnings in the quarter against roughly $295m of wear. That arithmetic, rather than a slowing market, is what the past three months of share prices have been reckoning with: Liberty is down 19% and ProPetro 23% since early July, while the diversified service majors SLB and Baker Hughes rose 8% and 6%.

Demand got better, not worse

The completions market these companies serve expanded through the period. Primary Vision's count of working frac spreads rose by eight to 195 on October 1, against roughly 169 a year earlier, with the US rig count at 599. The Dallas Fed's third-quarter energy survey showed its index of prices received for oilfield services swing from -30.0 to 9.3 — services pricing turned up in the quarter.

Managements describe a tightening market. "We're beginning to see positive pricing momentum," ProPetro chief executive Sam Sledge told investors after the second quarter, as customers recognize the limited availability of next-generation gas-burning fleets; the industry, he said, has "consolidated through attrition," with much of the excess capacity that once weighed on pricing now gone. The one-month slide across the whole complex is better explained by crude: Brent fell from above $105 a barrel in mid-September to $97.44 on September 30 as Saudi exports recovered, and the majors fell with the pure plays.

The second claim on the same cash

What separates the pumpers is a second demand on cash that depreciation has already laid claim to. Liberty expects roughly $1.5bn of capital expenditure in 2026, mostly deposits on long-lead power-generation equipment; net debt rose $157m in the quarter to $736m. Its filings show equipment supply contracts worth about $1.3bn, with $1.1bn still committed at June 30, plus a further $801m order signed with Caterpillar on July 22. Shares fell 13.7% on a quarter that beat both revenue and earnings estimates.

That spending buys a different business. "We have line of sight to that inbound 3 gigawatts, and even beyond that, we have line of sight to our growth headed out into late 2029 and early 2030," chief executive Ron Gusek said on the July 23 call, referring to the company's power platform. UBS, which cut its price target to $31 from $35 while keeping a Buy rating, reckons the market is paying only $2 to $4 a share for that platform against its own $14 estimate, and models third-quarter EBITDA of $143m against a Street figure of $156m. Liberty trades at 6.8 times trailing cash earnings and 1.60 times book, yet 83 times forward earnings — the same depreciation fact, seen through the income statement.

ProPetro, a Permian-focused pumper, is the cleaner case. It raised its active fleet count from 11 to 12 and grew adjusted EBITDA 23% sequentially to $45m, 15% of revenue, and its shares still fell a third over six months. Annualized, its depreciation of about $199m exceeds the $186m of 2026 EBITDA analysts expect — the pumping fleet does not earn its own replacement. Its power unit, PROPWR, added roughly 230 megawatts of committed capacity through long-term contracts with a Targa Resources subsidiary signed September 22, taking contracted capacity to about 510 megawatts. The shares sit at 1.19 times book and 7.2 times cash earnings.

ProFrac, which pumps with captive proppant and its own pump manufacturing, earns its decline. Revenue of $498m and adjusted EBITDA of $69m came with a $79.7m net loss, and net debt of $1.08bn against $72m of liquidity exceeds its $829m market value. At more than 15 times cash earnings it is the most expensive of the three, with consensus 2026 earnings at -$1.34 a share.

The layers beneath split on contract tenor

The service layers around the wellbore did not move together, which rules out a simple read on completions demand. Atlas Energy Solutions, the Permian sand and conveyor-logistics supplier, sold a record 5.6m tons of proppant at $17.70 a ton against $12.39 of plant cost and has fallen 17% over three months, with third-quarter EBITDA guided to $30m-$45m. Select Water Solutions, which gathers, recycles and disposes of produced water through owned pipelines, is up 11% over the same span on 12-year dedications across more than 875,000 Northern Delaware acres and infrastructure revenue up 26% at a 58.3% gross margin. Both consume more cash than they make; what divides them is the length of the contract, not whether it is priced per ton or per hour.

What the market is actually marking down

Nothing in these businesses supports the thesis of a thinning completions market, and the pricing data points the other way. What is being repriced at Liberty and ProPetro is the claim on cash: a fleet that depreciates faster than it earns operating income, now sharing the balance sheet with turbine and engine deposits that pay back over five or six years. ProFrac has no such second story, and its leverage is doing the work there. The uncomfortable implication is that for a pumper, depreciation functions as a cash commitment, so the pricing uplift managements have promised since July only matters at the margin above the replacement bill — which is precisely why both Liberty and ProPetro are pointing capital at contracted power, where an asset earns for 15 years instead of consuming itself in three.

Liberty's third-quarter report, due around October 14, will say whether that uplift arrives in cash or only in commentary. In this business the quarter that earns the most is also the quarter that wears out the most.

Argentina's four New York-listed banks each reported a better June quarter than the one before it, and each has since lost close to a third of its market value. What changed between the results and October was the price of lending to their government.

JP Morgan's Argentina country-risk spread added almost 100 basis points over September, touching 655 basis points intraday on 2 October before closing at 646 — the highest reading of the year, and above the 600 level at which Argentina stays locked out of international debt markets. The cause starts in New York, where the 10-year Treasury yield touched 5.34%, its highest since 2002 in a global bond sell-off before retreating to 5.175%. For a lender holding roughly a quarter of its assets in Argentine government paper, that spread is simultaneously an asset yield and a cost of equity. It is the second of the two that re-rated.

How the quarter was earned

All four banks made the same trade in the June quarter: they let deposits reprice downward faster than loans and bonds did. At Banco Macro, a retail and corporate lender now shrinking its network to about 370 branches from 402, asset yields fell 280 basis points sequentially while funding costs fell 550, so net interest margin excluding currency effects widened 30 basis points to 23.5% and net income rose 36.8% from a year earlier. At Grupo Supervielle, the smallest of the group at a $601m market value, margin expanded 250 basis points to 20.3%, above its own full-year guidance, for the identical reason. Grupo Financiero Galicia — the holding company behind Banco Galicia and the Naranja X consumer-credit platform, and the largest of the four at $5.8bn — held its margin at 17.0%, helped by inflation-indexed bond yields and gains on securities sales. Its net interest income fell 3% sequentially; income from financial instruments rose 275% on derivatives and securities disposals.

That is carry income, and carry has a floor. Banco Macro's income from government securities fell 18% from the previous quarter. BBVA Argentina, the local subsidiary of Spain's BBVA, expects its headline margin to fall about 200 basis points by year-end and is building roughly two-thirds of its securities book in bonds indexed to the greater of inflation or the Tamar money-market rate, explicitly to defend net interest income over one to three years. None of this is legible in reported revenue: under the hyperinflation restatement these issuers apply, Galicia's 2025 revenue rose 16.9% while its net income fell 86.9%.

What the business does explain

The private-credit growth that was the bull case is being withdrawn by the banks themselves. Macro cut 2026 real loan growth to 2-5% from 15-20%. Supervielle cut its target to 10-15% from above 20% as total loans slipped 1% sequentially. Galicia guides to 10-15% with the majority in dollars — its peso loan book shrank 4% in the quarter while dollar lending to oil, gas and exporters grew 19%, even as its loan market share reached 15.1%, up 69 basis points, helped by the HSBC Argentina integration.

Credit is seasoning underneath. Delinquency on household loans across the Argentine system hit 12.9% in July, the highest since 2004 and more than double a year earlier, against 3.6% for companies. Galicia's group non-performing loan ratio reached 10.6% — 8.3% at the bank, 19.7% at Naranja X. "for the year-end, we are expecting to be around 17% from almost 20% levels that we have during the second quarter," Hernán García, Naranja X's chief financial officer, said on the 26 August call. Macro's ratio rose to 6.25%, with consumer loans at 8.4%; Supervielle's improved to 5.5%, some 210 basis points below the system.

The dollar earners sat it out

YPF, the state-controlled integrated oil company, and Vista Energy, the Vaca Muerta shale producer, carry the same election and sovereign risk and went the other way. Vaca Muerta output set a record near 400,000 barrels a day in the third quarter, up 35% year on year, with Brent at $102.70 a barrel on 2 October, up 59% from a year earlier after the renewed blockade on Iranian exports. YPF set a 52-week high on 15 September and trades 13.8% below it; over three months YPF rose 11.9% and Vista 6.5% while the banks fell 28% to 36%. The oil shock that widened Argentina's spread paid its exporters. The difference between the two groups is the currency the revenue arrives in.

The verdict

Macro now trades at 1.07x book value, down from 1.23x in mid-September; Galicia at 1.02x, BBVA Argentina at 0.99x, Supervielle at 0.83x. "we see our ROE around 10% for the year... we see that that will continue improving, I would say, around 12%, try to end the year with something around 12%," Gonzalo Covaro, Galicia's chief operating and financial officer, told investors on 26 August. A dollar buyer of Argentine sovereign paper collects roughly 11.7% today, adding the 646-basis-point spread to the Treasury yield. The comparison is approximate — the banks' guided returns are inflation-adjusted peso returns, not dollar ones — but at book value, a mid-teens-at-best return on a seasoning consumer book is thin compensation for owning the bond's risk through a levered balance sheet.

So the June results earn none of this fall, the loan-growth downgrades and the household delinquency earn part of it, and the remainder is the discount rate. The likelier reading is that the market has stopped paying for disinflation and started pricing how the Treasury funds itself next year.

October alone carries about ARS 27 trillion, roughly $17.7bn, of peso maturities — all of it due before any of these banks reports again. Their deposit books are the other side of that trade.