DK Street Journal

The units arrived; the dollars did not

Hesai Group, the Shanghai maker of three-dimensional lidar sensors sold to Chinese carmakers and robotaxi fleets, shipped more than 628,000 units in its second quarter, up 80% from a year earlier. Revenue rose 21.9% to RMB 860.8m and gross profit 14.9%. Operating income came to RMB 2.2m — about $0.3m — against RMB 22.9m a year earlier.

That gap between the unit count and the profit is the economics of advanced driver assistance in one line. Global automotive lidar crossed $1bn of revenue for the first time in 2025, on 3.7m passenger-vehicle units, up 60% year on year, with Chinese suppliers holding roughly 95% of the market. A market growing at semiconductor speed has produced a revenue pool smaller than one quarter at a mid-sized parts maker. Cars are taking more sensors and more silicon; so far nobody in this corner is being paid more for supplying them.

Price engineering as strategy

Hesai is not a victim of the deflation — it authored it. Its ATX unit is expected to sell at about $150 in 2026, and around RMB 800 for high-volume customers such as BYD and Geely, against roughly $400 for the older AT128. Chief financial officer Andrew Fan told investors that the right measure is total lidar content per vehicle rather than the price of one unit, with multi-lidar cars heading toward $500 to $1,000 of content, and defended a near-40% full-year gross margin as a matter of price minus cost — proprietary chips, in-house engineering, factory automation. Li Auto fits four Hesai lidars to its L6, L8 and L9. Capacity is going from 2m units to more than 4m this year, and management flagged the third quarter as the point where non-automotive lidar approaches half of revenue.

The other meter

Mobileye Global, which sells the EyeQ vision chip and the higher-content Super Vision stack to carmakers and remains majority-owned by Intel, bills per car rather than per sensor. It shipped 10.0m systems in the June quarter, up 3%, while its ten largest customers' own vehicle production fell 3% — genuine share gain. And the average system price fell to $48.50 from $49.70. Gross margin dropped to 46.3% from 49.8%; gross profit shrank 6.7% on revenue up 0.4%.

The mix shift meant to fix that is not yet visible. The high-content stack ran about 20,000 units in the quarter and over 40,000 in the half, with full-year guidance still under 60,000 and second-half volumes expected to fall as customers draw down inventory. A high-volume 2027 Stellantis award carries gross profit per unit more than double base driver assistance, but management put the point where Surround ADAS lifts average pricing at 2028. Meanwhile the raised full-year profit guidance — adjusted operating income of $395m against $210m before — rests on a research-and-development tax credit worth $93m in the quarter and $180m to $200m for the year, which roughly halves if Intel ceases to control the company. The improvement is fiscal and cost-side.

"Mobileye's core business continues to perform very well in 2026, providing a strong foundation as we build towards upcoming advanced product launches," founder and chief executive Amnon Shashua said on July 23 — the same day the company said he would step aside once a successor is named, after 27 years.

Pricing leverage sits with the buyers. Mobileye's four largest customers were 32%, 19%, 15% and 13% of revenue in the first half of 2025 — about four-fifths in four hands. In China, Horizon Robotics held 35.9% of driver-assistance solutions among domestic carmakers in the first half of 2025 against Mobileye's 26.9%, which is why Mobileye's China growth now comes from Geely and Chery exporting cars rather than from domestic programs. The base beneath both suppliers is contracting: August passenger-car retail sales fell 19% and electrified retail 4%, an eighth straight monthly decline, even as wholesale volumes rose 16%, after the purchase-tax exemption halved to a maximum RMB 15,000 on January 1.

What a mature supplier looks like

Autoliv, the Stockholm maker of airbags, seatbelts and steering wheels, earns an 18.2% gross margin — less than half Hesai's and well under Mobileye's — and is the only one of the three with a real operating margin, at 6.8%. Even there, operating income fell 22.3% year on year on carmaker pricing. It trades at 14.7 times trailing earnings and 12.4 times forward.

The verdict

Both perception names left a bear trend for a flat, directionless one on July 28 and have gone nowhere in the five weeks since; Mobileye is down 42.4% over twelve months and 18.8% over three, Hesai down 28.5% over twelve. Hesai jumped 11.3% on September 3 with no company announcement discoverable, alongside preliminary August industry data — the likelier reading is positioning rather than news. These declines are earned: prices have tracked gross and operating profit, not shipment counts.

What is left is two cheap-looking claims on the same car. Mobileye trades at 7.3 times trailing gross profit and 7.4 times forward — the market pricing zero gross-profit growth — and 0.85 times book after the write-down that produced a $3.8bn quarterly loss; that is a dislocation only if Stellantis in 2027 and Surround in 2028 land as described. Hesai, at 13.6 times trailing gross profit, roughly double Mobileye and triple Autoliv's 4.3 times, is priced for the multi-lidar content thesis to arrive before the price war finishes its work. Its forward multiples divide a dollar share price by renminbi estimates and cannot be compared with either.

Hesai will finish this year with capacity for more than 4m lidars. The question both meters pose is not whether cars get more sensing — they will — but whether the companies building it collect anything for the increment. On the evidence of one quarter each, the units showed up and the money stayed with the carmakers.

The pipe did not narrow the discount

EQT, the largest natural gas producer in the United States, holds roughly 2.0 million gross acres in Appalachia and, since absorbing Equitrans Midstream, is the only producer in the basin that also owns the gathering system its molecules travel through and a stake in Mountain Valley Pipeline — the one major interstate line built out of the region in more than a decade. The commercial argument for that deal was that Appalachian gas sells at a discount to the national benchmark unless it is moved on firm transportation, and that owning the pipe converts the expense into margin. In the second quarter of 2026, EQT's realized price came in $0.67 per thousand cubic feet equivalent below the benchmark. Its own release calls that better than guidance, "despite widening basis during the quarter due to benefits from marketing optimization and curtailment strategy."

Better than guidance, but wider than the neighbors. Range Resources, a pure-play Marcellus producer with no interstate pipeline of its own, realized $2.42 per thousand cubic feet on gas including basis hedges — a $0.47 discount — and tightened its full-year differential guidance to a range of $0.35 to $0.40 below the benchmark. Antero Resources, the liquids-weighted Appalachian producer that is also the largest US exporter of natural gas liquids, guides its 2026 gas price to a premium of 5 to 15 cents. The three measures are not identical — EQT's spans all products, Range's is gas including basis hedges — but no version of the arithmetic makes the integrated producer the one capturing the differential.

The reason is that Appalachian basis is a regional clearing price rather than a toll any single owner collects. Forward fixed prices at Tetco M-2, the basin delivery point, average $2.940 per million British thermal units for 2026 against Henry Hub forwards near $3.813, and they sit there because the region has fully utilized its effective egress capacity. Every producer without unused firm transport clears into the same constrained market.

What integration actually earned

Something real, just not what was advertised. EQT sold 634 billion cubic feet equivalent in the quarter, above the high end of guidance on well performance and system-pressure optimization; total per-unit operating costs fell to $1.03 per thousand cubic feet equivalent from $1.08 a year earlier; capital spending came in 9% under the low end. Compression investments on the acquired midstream system extended flat production times on new wells, and full-year output guidance went up by roughly 90 billion cubic feet equivalent while capital guidance came down $25m. The counterweight sits in the 10-Q: gathering expense per unit actually rose year on year, on higher volumes gathered by third parties.

Underneath the operational beat, the income statement collapsed with the strip. Revenue fell 29% to $1.81bn, operating income fell 60% to $454m, and diluted earnings per share came in at $0.44 against $1.30 a year earlier. Free cash flow attributable to EQT was still $330m with gas averaging $2.89 per million British thermal units, and net debt stands at $5.5bn against a $5bn target.

An unhedged view, held on purpose

EQT entered 2026 with roughly 35% of its local basis sales hedged against a historical norm nearer 90%, a deliberate directional position that Appalachian differentials will improve. Chief financial officer Jeremy Knop put it plainly on the July 22 call: "I don't see a way for basis not to continue to strengthen materially." The 2027 commodity book, per the first-quarter filing, totals about 138 million dekatherms against guided 2026 production of 2,375 to 2,450 billion cubic feet equivalent. Antero, by contrast, has 34% of its 2027 gas hedged at $3.84 plus collars. Two neighbors, the same rock, very different exposure to the same curve — and the EIA now expects a record 3,985 billion cubic feet in storage at end-October and spot prices below $3.00 until November.

The demand contracts that are supposed to bypass basis entirely are dated later than the exposure. The 10-year agreement to supply 325 million cubic feet a day to Competitive Power Ventures' 2-gigawatt Shay Energy Center in West Virginia, priced off regional power rather than a gas index, is definitive — but the plant is not expected in service until early 2031. Management sizes it at about $100m of annual free cash flow and five cents on the corporate differential. A liquefied natural gas offtake starts in 2028. The Homer City data-center campus arrangement, the largest of them by volume, remains an agreement in principle.

The verdict

Integration bought EQT operating control — flatter declines, lower unit costs, the ability to curtail into weak prices — and it did not buy the basis capture that justified the merger's premium. That distinction matters because the market has already made it. EQT trades at 13.3 times forward earnings against 12.1 times trailing, the forward multiple sitting above the trailing one because consensus 2026 earnings of $4.14 a share are below the roughly $4.55 the company has just earned. That is the most expensive forward multiple of the Appalachian group, where Antero trades near 9x and Range near 10x. The shares, at $55.17, are flat over three months and 19% below their 52-week high, up 6.9% over twelve months against roughly 21% for both Range and Antero; those two recovered a rising trend in late August, and EQT did not.

So the next two years of EQT earnings are, in substance, an unhedged position on Henry Hub and on a basin differential narrowing before the power contracts arrive. That may prove right — Toby Rice counts more than 45 Appalachian demand and takeaway projects under construction or evaluation, nearly 20 billion cubic feet a day of potential demand. It is simply a different proposition from the one the pipeline purchase was sold as.

One more thing the filing settles. With the stock at its 52-week low, Knop told the July call EQT would look to be more aggressive on repurchases, while adding it was willing to build a few billion dollars of cash first. The quarter's filing shows no shares bought at all; Expand Energy, the former Chesapeake, retired $850m of its own stock in the same three months.

The plant arrives before the rate case

CMS Energy, the Michigan holding company whose Consumers Energy unit serves 1.9 million electric and 1.8 million gas customers, reported second-quarter operating income of $264m, down 16.7% from a year earlier on revenue that barely moved at $1.83bn. Its operating margin fell to 14.4% from 17.2%. DTE Energy, which serves 2.3 million electric customers in southeastern Michigan and 1.3 million gas customers statewide, saw operating income fall 7.3% to $396m. Both companies reaffirmed a 6-8% long-term earnings growth algorithm on their July 28 calls.

The two are in the middle of the largest construction programs they have ever run. DTE's five-year capital plan rose to $36.5bn from $30bn, with $30bn at DTE Electric; CMS's utility plan runs to $24bn through 2030 and supports 10.5% annual rate-base growth. The accounting for that plant lands before the revenue does — depreciation and financing costs hit the quarter they are incurred, while rate relief arrives on the commission's schedule. Both companies now show deeply negative trailing free-cash-flow yields, minus 8.9% at CMS and minus 6.8% at DTE, which is simply capital spending running ahead of operating cash flow. That gap has to be financed.

The load is signed; the earnings are not

The demand side is not in doubt. DTE has 2.4 gigawatts of signed data-center agreements — 1.4 GW with Oracle under construction and 1.0 GW with Google awaiting Michigan Public Service Commission approval expected in September — plus a further 2 GW in advanced discussion. Management said on the July 28 call that the signed contracts get it to "solidly 8%" earnings growth. CMS has contracted 135 megawatts of manufacturing and industrial load year to date and signed a large data-center tariff agreement whose customer is still evaluating multiple Michigan sites, with zoning approval pending.

What that build costs is the newer part of the story. DTE told investors it is "targeting $500-600M annual equity issuances 2026-2028 with similar levels through 2030" while "maintaining FFO to debt ratio ~15% and investment-grade rating." CMS filed a $3.0bn at-the-market common equity program with forward sales on May 13 and has completed roughly $495m of a planned $700m of 2026 equity. Common shareholders are being asked to fund this directly, not spared.

The hybrid layer is a supplement. On June 18 DTE sold $1.0bn of 2026 Series C 6.200% reset-rate junior subordinated debentures due 2058, deferrable for up to ten years; Fitch rated them BB+ and gave them 50% equity credit. That is the price of half-equity today: 6.200%, against a Michigan allowed return on equity of 9.9%.

What that repricing did to the retail notes

Six $25-par junior subordinated notes from these two issuers trade on the exchange — CMSA at 5.625%, CMSC and CMSD at 5.875%, DTW at 5.25%, DTB and DTG at 4.375%, all struck between 2017 and 2021. Their prices fell between 11.4% and 14.7% over the twelve months to September 3, DTW to $19.43 and DTB to $15.72. Those price series exclude coupons; including them, holders lost roughly 5.2% to 9.0%.

The cause is arithmetic. To yield what the market now demands of utility hybrids, a note paying a fixed $1.09375 a year on $25 of face has to trade near $15.72. Current yields on the six now run 6.76% to 7.15%, up 71 to 100 basis points in a year, against about 40 basis points on the 30-year Treasury, which touched 5.31% on August 17, its highest since 2007. That leaves 30 to 60 basis points of spread widening — modest, and uniform across issuers. Duke Energy's comparable 5.625% debentures fell 11.1% over the same period. Credit is not the explanation: S&P affirmed DTE at BBB+ with a stable outlook after the Oracle downgrade, and Consumers Energy's senior secured debt is rated A, A1 and A+.

The verdict

The notes have de-rated for a reason that has nothing to do with these companies: long-end rates and a higher clearing cost for hybrid capital. Nobody will redeem 4.375% money to replace it at 6.200%, so the market is treating them as very long paper and pricing them accordingly. The common shares are the harder case. CMS fell 4.1% over twelve months and DTE rose 1.2%, while Entergy gained 23.6%, NextEra 17.4%, Dominion 13.4% and American Electric Power 13.3%. That underperformance is earned — the earnings actually declined. DTE's trailing price-to-earnings ratio has risen since May, from 20.91x to 21.36x, because profits fell faster than the price; the stock went nowhere and got more expensive on what it has already reported. Both trade near 17.6x forward earnings, below Duke's 17.9x and American Electric Power's 19.5x, but that discount only exists if the consensus 6-8% growth survives the margin compression now visible in two consecutive quarters.

The Google contract's commission decision is expected this month. It is the first dated test of whether Michigan's regulators will let the load growth reach the income statement before the next round of financing prices.

Booking Holdings spent $2.37bn on marketing in the June quarter, 11% more than a year earlier, and got gross bookings growth of 9% for it. Expedia Group, reporting a day later, raised the direct marketing behind its consumer brands by 1%, to $1.10bn, while bookings on that side of the business grew 8%.

An online travel agency keeps only a slice of the travel it sells. Revenue as a share of gross bookings ran from 14.4% at Booking — the owner of Booking.com, Agoda, Kayak, Priceline and OpenTable — down to roughly 5.4% at China's Trip.com. A large, separately disclosed piece of that slice goes straight back out to Google, metasearch and affiliates for the traffic that produced the sale. Marketing measured against bookings is therefore the line that decides whether one of these companies is a marketplace or a reseller of purchased attention. Last quarter it moved the wrong way at the company that was supposed to have escaped: Booking's marketing reached 4.7% of gross bookings, up from 4.6% a year earlier, while Expedia's consumer marketing fell about a third of a percentage point to the same 4.7% of its consumer bookings.

The mix held; the bill did not

Booking's own disclosures show the direct-channel story intact and static. Its 10-Q for the June quarter puts consumer direct traffic in the mid-60% range, the mobile app at a high-50% share of room nights, and Genius level 2 and 3 members — more than 30% of active customers — at a high-50% share of Booking.com room nights. Those figures have held for four quarters. What changed is that revenue growth halved, to 8.1% from 16.2% in the March quarter, while room nights grew 5%.

The earnings still grew. Adjusted earnings per share rose 15% against adjusted EBITDA up 9%, the gap supplied by a 6% reduction in share count after a record $3.7bn of buybacks in the quarter, and the transformation program's run-rate savings target was raised to $650m from $550m, mostly procurement, with the benefit expected in 2027. "We are at a high single-digit level for gross bookings and revenues and at mid-teens level for EPS at the high end," chief financial officer Ewout Steenbergen told investors on August 4, noting guidance assumes seven of twelve months of Middle East disruption. Chief executive Glenn Fogel acknowledged that Google's AI Overviews are pressuring organic search; Steenbergen said referral traffic from large language models is under 1% of room nights and "hasn't moved so much recently".

Where the leverage actually showed up

Expedia's operating margin reached 23.9% against 14.0% a year earlier, on revenue up 14%, with business-to-business bookings up 21% in a twentieth consecutive double-digit quarter. "We exceeded the high end of both our top and bottom-line expectations for the fifth quarter in a row, growing bookings 12%, revenue 14%, and adjusted EBITDA 23%," chief executive Ariane Gorin said on the August 5 call; finance chief Derek Andersen raised full-year adjusted EBITDA margin guidance to expansion of 1.5 to 1.75 percentage points. Expedia's take rate, 12.7%, sits below Booking's because its fastest-growing channel is wholesale. But marketing consumes 49% of Expedia's revenue against 32% of Booking's — and it is the wider ratio that is compressing.

The two mirror images sit outside the United States. Trip.com — Ctrip, Qunar and Skyscanner — was fined RMB5.2bn (about $770m) by China's market regulator on July 27 for forcing hotels into exclusivity, and its corrective plan requires redesigning merchant classifications, commissions, traffic allocation and pricing tools. Its March-quarter sales and marketing rose 25% against revenue up 17%, operating margin fell to 24.3% from 25.8%, and June-quarter revenue growth was guided to 3-8%. MakeMyTrip, India's largest platform, grew gross bookings 19.9% in constant currency but 9.4% as reported on a weaker rupee, and net income fell 68% to $8.35m; its international air business shrank 13% as jet fuel headed for an IATA-projected average near $152 a barrel this year against roughly $90 in 2025.

What the prices have done since

Booking, Expedia and MakeMyTrip have all held uptrends since August, their 50-day averages above their 200-day; Trip.com has not, and never joined. Yet all four have faded since the results: Booking down 6.6% from its August 5 close, Expedia 6.0%, MakeMyTrip 8.9% and Trip.com 10.8%. Over twelve months the spread is 82 points — Expedia up 37.7%, MakeMyTrip down 44.4%.

Expedia trades at 14.3 times forward earnings against 17.9 times trailing, and 8.4 times trailing EV/EBITDA. Booking is at 18.5 times forward, above the 17.5 times that Scott Black argued in Barron's Midyear Roundtable on July 14 was not warranted by the size of its advantages, with consensus revenue growth of 9.6% this year. MakeMyTrip's 30.7 times trailing EV/EBITDA is the group's most expensive; Trip.com's reported multiples are unusable, its trailing earnings flattered by a one-off gain that pushed one quarter's net income above its revenue.

The verdict is an inversion of the received story. The direct-mix advantage is real at Booking and it is visible in the disclosures — but it is not showing up in the cost line, and this year's earnings growth is being bought with share retirement and procurement savings rather than cheaper customers. The company where customer acquisition genuinely got cheaper is the one assumed to be renting its demand. Neither is being disintermediated yet: when Google's agentic hotel-booking tool arrived in AI Mode in August, Booking.com, Expedia, Priceline and Trip.com were all launch partners.

The sharpest test is not a channel-mix number at all. Trip.com reports on September 15, the first results since a regulator ordered it to rewrite the commissions and traffic allocation that produce its take rate — the closest thing to a controlled experiment on what an online travel agency earns when it can no longer set the terms.

The cost shock is over

Brazil is finishing a record coffee harvest, and the input-cost shock that defined the industry for two years has broken with it. Arabica set a record at $4.41 a pound in February 2025, held above $4 again last October, and traded near $3.10 in early September, with more than 90% of a Brazilian crop of roughly 75 million bags already picked. The 40% U.S. tariff on Brazilian coffee was eliminated by executive order on November 20, retroactive to the 13th.

The two listed American coffee companies book that relief in different places. Westrock Coffee owns no consumer brand: it procures green coffee and roasts, extracts and cans it for retailers, restaurants, convenience chains and packaged-goods customers, on contracts that pass the commodity through with a conversion fee attached. The bean inflates and deflates its revenue without ever touching its gross profit. Black Rifle Coffee buys the bean and sells the bag — through grocery, mass and convenience retail, its own cafés and e-commerce — at shelf prices that move slowly. For Black Rifle, the coffee price is the margin.

The split that looks like a rally

Black Rifle's shares appear to have gained 718% in a month. They did not. The company executed a 1-for-10 reverse stock split effective August 21, with Class A stock trading on a split-adjusted basis from August 24 under the same ticker. The New York Stock Exchange had notified the company in February that it sat below the $1.00 minimum price; compliance was regained in late May, and the split went ahead regardless. Market value is unchanged at $278m. Split-adjusted, the stock is down about 15% over a month and 42% over a year.

The business went the other way. Second-quarter revenue rose 12.8% to $107.0m and gross margin expanded to 34.1% from 33.9% on pricing and lower shipping and fulfillment costs. Operating income was positive at $1.1m against a $12.6m loss a year earlier, and adjusted EBITDA more than doubled to $6.3m. The compression coffee was supposed to cause already happened — annual gross margin fell from 41.2% in fiscal 2024 to 34.6% in fiscal 2025 — and it is reversing, with all of 2026's green coffee secured at $2.95 a pound and more than half of 2027 at $2.65. Growth is coming from distribution rather than price: wholesale revenue grew 15.2% and retail presence reached 56.5% of all-commodity volume, while cafés shrank. "We believe in building a better business before we build an aggressively bigger business," president and chief executive Chris Mondzelewski said with the results.

The equity prices none of it. Black Rifle trades at 0.65 times trailing sales and 0.64 times forward — no spread between the business behind it and the one consensus models ahead. On cash earnings it is not cheap, at 87.5 times trailing EBITDA. And the reverse split sits on top of genuine dilution: the diluted share count has grown roughly 24% since fiscal 2025.

Westrock is the pair's actual winner

Westrock's pass-through mechanics show up in one line of its annual accounts: fiscal 2025 revenue rose 39.8% to $1.19bn while gross profit fell 20.2%, dragging gross margin from 18.1% to 10.3%. That optic is now unwinding with the coffee price — second-quarter revenue grew only 8.8%, to $305.7m. Gross margin fell to 12.3% from 14.7%, but $4.1m of the shortfall is incremental depreciation on newly commissioned assets running through cost of sales, plus a $2m mark-to-market loss in the green-coffee sourcing segment. First-half gross profit was $83.5m, up 19%.

Those assets are the $315m Conway, Arkansas roast-to-ready-to-drink plant, 570,000 square feet, whose five extract and canning lines were all commercialized by the first quarter. The quarter they produced was a record: adjusted EBITDA of $21.3m, up 39%; free cash flow of $20.2m, a quarter earlier than promised; leverage down to 3.36x, a fifth consecutive quarter of improvement. "We're on pace for estimated capital expenditures in 2026 of approximately $30 million, down from the $160 million in 2024 and $89 million in 2025, which again represents a structural shift in the capital profile of this company," chief financial officer Chris Pledger told investors on the August 6 call. Guidance of $90–100m of adjusted EBITDA for the year was reaffirmed.

The shares have followed, up 91% in six months. They carry 23.8 times trailing EBITDA, and price against gross profit rises from 4.62 times trailing to 4.89 times forward — investors are paying more for next year's gross profit than for last year's, which happens only when a ramp is already in the price. Consensus still models a $33.6m net loss for 2026, with the crossover to profit in 2027.

What the bean explains and what it doesn't

Cheaper coffee has sorted these two, and the quoted prices point the wrong way about which one it favored. Westrock has earned its six-month gain in cash: the plant is built, capital spending has collapsed to maintenance-plus, and every dollar of coffee inflation it ever booked belonged to a customer. What it has not yet earned is the forward multiple, which assumes the remaining Conway capacity fills. Black Rifle is the mirror image — revenue accelerating, margin expanding, operating income positive two quarters running, against a share price 42% below a year ago that no reported result explains. The likelier reading is that a nano-cap trading under a dollar, with an activist holding 12.2% of the class and a growing share count, was being priced as a financing question rather than a coffee one.

For two years, the answer to almost any question about either company was the price of the bean. Black Rifle has now bought its next two crops forward, and Westrock never owned the risk to begin with. Whatever decides these two from here, coffee will not.

Asana's AI products supplied about a quarter of the company's net new annual recurring revenue last quarter, chief executive Dan Rogers told investors on 3 September. That is ahead of the 15% full-year target management set in March; the next session the shares fell 12.7% to $8.81, dragging monday.com down 6.4% with them.

The reason it matters beyond one print: work-management software — Asana's platform coordinates teams' tasks and projects, monday.com sells a configurable work operating system plus dedicated customer-relationship, marketing and development products — is sold per seat per month. Its revenue base is the number of knowledge workers a customer chooses to license, precisely the count that flat white-collar hiring and AI-assisted headcount discipline are supposed to compress. Both companies are attaching a consumption-priced second meter to escape that link. Asana's is now big enough to measure, and what it shows is that the escape carries a price of its own.

The new meter arrives with cost

Asana's revenue reached $216.4m, up 9.9% year over year and above the top of guidance, with growth inching higher for a third straight quarter. Retention improved too: dollar-based net retention of 97%, with customers spending $100,000 or more at 98% against 96% a year earlier, and those spending $5,000 or more up 7% to 26,778. Five consecutive quarters of improvement, all of it still below 100% — existing customers are shrinking, just more slowly.

The cost shows up one line down. Gross margin fell to 86.0% from 89.7%, about 120 basis points of that sequentially on AI compute and the small StackAI acquisition, with roughly another 150 points of headwind flagged for the second half. Gross profit grew 5.3% while revenue grew 9.9%. Guidance compounds it: third-quarter revenue of $217–219m implies 8–9% growth, a step down, with adjusted earnings of $0.08 against the $0.09 analysts expected.

Chief financial officer Aziz Megji named the trade-off on the call: "The seeding should accelerate adoption in users, but it can push out the timing of incremental paid consumption." Management also described the free tier as clogged with "heavy tire kickers" below $5,000 of spend, a drag it assumes will not recover this fiscal year.

monday.com's version is smaller, and its recovery is not revenue

monday.com doubled its AI recurring revenue quarter on quarter to 17% of net new business — still about 1% of a book that crossed $1.5bn in July. Its expansion comes from selling more products into the same accounts: customers above $100,000 of annual recurring revenue rose 37% to 2,019 while accounts with more than ten users grew 6%. Net retention was 109%, guided down to 108% as past price increases lap.

Revenue growth is decelerating — 26% a year ago, 22% last quarter, 16–17% guided for the current one. What improved is the profit and loss account: GAAP operating margin turned positive at 5.5%, free-cash-flow margin is guided to 19–20%, the $870m repurchase authorization has been completed, and about 630 jobs went in July, roughly a fifth of staff, for some $100m of annual savings.

What each side earns

monday.com's advance since its March low is paid for by margin and cash, not by faster sales; at 16.7x forward earnings and 2.95x forward gross profit for a 19–20% grower, that is a defensible price for a decelerating business. Asana trades at 2.77x forward gross profit — a discount of only about 6% for less than half the growth — and its trailing price-to-earnings ratio is negative because the company still loses money on a GAAP basis. On 4 September the seat-metered complex sold off together, Workday and Paycom included, after 162,000 August payrolls against a 55,000 consensus lifted rate-rise odds; Samsara, which meters connected devices, rose almost 10%.

Consumption pricing does not remove the growth problem so much as relocate it. Asana has proved a second meter can reach a quarter of new business within two quarters — the clearest evidence in the category that customers will pay for something other than headcount — and proved at the same time that those dollars land at a lower margin than the seats they replace. Sell-side reaction split accordingly, with Citi raising its target while DA Davidson stayed neutral, both citing the same 25%.

From mid-September, Asana folds AI Teammates, AI Studio and Dash into every paid tier at no extra list price and charges for usage on top. New logos and renewals move onto it immediately, which makes the next print the first honest read on whether customers who stopped adding seats will pay for requests instead.

Vistra had sold forward roughly 94% of the electricity it expects to generate in 2027, and about 72% of 2028, as of August 3, according to its second-quarter earnings release. Expected 2026 generation was essentially fully hedged. Those three percentages decide more about the next two years of reported profit than any data-center announcement will.

Vistra runs about 38,700 MW of nuclear, gas, coal, solar and battery capacity and sells power and gas retail to roughly 4.3 million customers in 20 states. It does not sell electricity at the forward curve the artificial-intelligence narrative trades on. It sells at prices struck one to three years earlier, and its retail arm is a deliberate counterweight whose margins widen when wholesale prices fall. The scarcity everyone is pricing therefore arrives on a delay — and when it arrives, it lands only on the open tail, which does not begin to widen until 2028.

The ladder cuts both ways

That is why the quarter held. Adjusted EBITDA rose about 30% to $1.77bn, generation EBITDA up 68% to $994m against $773m from retail, on realized prices roughly 5% higher per megawatt-hour — while spot power in Texas went the other way. Houston round-the-clock power averaged $33/MWh, 8% below a year earlier. Vistra reaffirmed 2026 guidance of $6.8bn-$7.6bn in adjusted EBITDA.

The softness is physical, not seasonal noise. Chief executive Jim Burke told analysts on the August 7 call that the market was showing a "recency bias" from weather plus a significant influx of batteries performing below expectations. Management pointed to July 22, when the market cleared at $57/MWh despite tight availability, and suggested it could have cleared $400-$500 absent battery competition. Vistra's 2027 range of $7.4bn-$7.8bn is now trending toward the lower end on a declining ERCOT forward curve, partly offset by stronger PJM prices and the nuclear production tax credit's price floor.

What is actually contracted, and when it pays

The merchant frame misses real signed business. Vistra holds 20-year agreements supplying Meta with more than 2,600 MW of PJM nuclear output and a separate 20-year, 1,200 MW arrangement at Comanche Peak in Texas, and in August won approval for Cogentrix, ten gas plants totaling 5,500 MW for a net $4.0bn. None of it sits in the 2027 range; management calls the Meta contract and Cogentrix together worth roughly $700m.

Capacity revenue offers no rescue either. PJM's auction announced July 14 cleared at $325 per megawatt-day, 2.5% below the prior $333.44, and did so at the approved administrative cap while falling 6,831 MW short of PJM's own reliability requirement. ERCOT, an energy-only market, holds no capacity auction at all. Meanwhile Texas ordered a verification audit of every data center in ERCOT's interconnection queue — about 474 GW of requests, some 90% of them data centers — which the grid operator aims to finish by December 10. Vistra's own realistic estimate is 12-15 GW of large-load additions in ERCOT by 2030.

The other engine is winding down. Vistra has repurchased about $6.5bn of stock since November 2021, cutting the diluted count from 482.2m in 2021 to 339.8m in 2025, with roughly $1.2bn of authorization left that it expects to exhaust no later than the end of 2027. Operating income fell 78.6% in 2025 to $1.34bn.

Where the group trades

The shares are down 21.3% over twelve months to $149.30, 31.5% below their 52-week high, at 17.3x forward earnings against 24.9x trailing and 10.8x trailing EV/EBITDA. Constellation Energy, which raised 2026 guidance to $11.50-$12.50 a share and signed roughly 920 MW of nuclear contracts at an 18.5-year average tenor, commands 24.8x forward and is down just 3.3% over the year. NRG, the more retail-weighted operator serving about six million customers, is the cheapest at 13.4x forward against 31.2x trailing — and has earned it: Texas EBITDA fell $131m, adjusted earnings per share slipped to $1.49 from $1.73, and its diluted share count rose from 190.4m to 210m as LS Power stock consideration swamped $932m of buybacks. Its headline 1.2 GW hyperscaler plant is still only aligned on principal commercial terms, pending a final investment decision, with operation targeted for late 2029.

The verdict

The division is not contracted versus merchant. Talen Energy is contracted — 1,920 MW to Amazon through 2042 — and fell 16.7% over the year, because its volume ramp is 840-1,200 MW in 2029 building to 1,920 MW by 2032. What is being discounted is anything dated 2029 and later, in favor of realized earnings now. On that test Vistra's de-rating is unexplained by its results and NRG's is not: one reaffirmed guidance with earnings up 30%, the other planned on $52 power and got $33.

The same hedge book that carried Vistra through a soft Texas summer is what will keep it out of a strong one. The first genuine choice management faces is what to do with the roughly quarter of 2028 still unsold — and that choice gets made long before any of the announced load actually plugs in.

Palo Alto Networks reported its fiscal fourth quarter on September 1 with the best growth number in its recent history — next-generation security annual recurring revenue up 63% to $9.10bn, nearly $1bn of it added in the quarter, on revenue of $3.41bn and remaining performance obligations of $21.2bn. The guidance published alongside it says most of that lift does not repeat. The company set fiscal 2027 next-generation security recurring revenue at $11.075bn to $11.175bn, growth of 22% to 23%.

The gap between the two figures is the CyberArk and Chronosphere acquisitions rolling into the year-ago base. That matters beyond one guidance line, because both of the largest platform security vendors are selling the same idea to the same budgets — consolidate a dozen point products into one subscription — and the question a buyer of either stock is answering is whether the vendor consolidates by acquiring the products or by selling more of its own.

What the acquired route costs

Palo Alto, the $311bn Santa Clara maker of hardware and software firewalls and the subscription services layered on them, run by Nikesh Arora with 21,491 employees, shows the price of the first route in two places. In the April quarter, revenue grew 31.1%, but roughly 14% of that was organic once CyberArk and Chronosphere are stripped out. And gross margin fell to 67.6% from 74.2% two quarters earlier, as purchase-accounting amortization from the deals landed in cost of revenue.

What the organic route produced

CrowdStrike, the Austin cloud security vendor whose Falcon platform sells endpoint, identity, cloud-workload and log-management modules on subscription, reported its July quarter on August 26: revenue up 25.8% to $1.471bn, a fourth consecutive quarter of acceleration, with none of it bought. Gross margin widened to 74.6% from 73.5%. Recurring revenue reached $5.84bn, up 25%, and net new recurring revenue was a record $332.8m, up 51%.

The number that answers the standing bear case — that Falcon Flex, the pooled multi-year commitment customers reallocate across modules at will, books commitments faster than it collects cash — is the cash line. Free cash flow was $377m, up 33% and 26% of revenue, growing faster than the recurring revenue it supposedly outruns. Both gross and net dollar retention rose sequentially, and 51% of subscription customers now run six or more modules, 35% seven or more.

"We are in an arms race. AI is driving more cyber attacks. AI is driving more cyber spending," co-founder and chief executive George Kurtz told investors on the August 26 call. The company raised its full-year net new recurring revenue growth outlook by 630 basis points to 34% and lifted revenue guidance to $5.99bn-$6.01bn.

The premium

CrowdStrike shares, split four-for-one effective July 2, are up roughly 110% over twelve months on a comparable basis, rose 20.5% the day after results to a record close, and have since eased about 5% from that high. The stock trades at 172.7x forward earnings and 54.3x trailing gross profit, against roughly 32x gross profit in early May. Palo Alto trades at 93.3x forward earnings and 40.8x trailing gross profit.

So the market pays CrowdStrike an 85% premium on forward earnings for growth that is slower on the headline and cleaner underneath. That premium is earned on the composition — organic, accelerating, cash-converting — and nothing in the results explains its near-doubling in four months. Palo Alto's problem is the mirror image: its guidance concedes the underlying rate, and a 22% grower carrying purchase-accounting drag is a different security company from the one the 63% print described.

Palo Alto's next quarterly report is the first in which CyberArk appears on both sides of the year-over-year line. That is when the acquired growth stops flattering the comparison and starts having to be sold.

Omnicom said on 28 July that organic revenue from its core operations grew 6.1% in the second quarter, and raised its full-year growth target to 5% from a range of 4% to 4.5%. Nine days later WPP reported that its revenue less pass-through costs had fallen 4.7% on a like-for-like basis in the first half. The two figures are not opposites, because the two companies do not measure the same thing.

What separates them is money that passes through an agency on its way to a media owner. WPP, the London group behind media buying arm WPP Media and a stable of creative and public relations brands, reported £6,373m of first-half revenue and £4,745m after pass-through costs — £1,628m, or 25.5% of the headline line, never touches the fee base. Stagwell, the challenger group run by Mark Penn, posted $632m of net revenue against $786.3m reported in the quarter, implying roughly $154m of the same. Omnicom, the New York holding company that closed its acquisition of Interpublic and now employs about 120,000 people, does not make the deduction: its organic growth is calculated on gross revenue including pass-through costs, and third-party service costs in the quarter rose 65.8% to $1.5bn, about 23% of the $6.56bn revenue line.

The buying practice that inflates the line

The gap is not bookkeeping trivia. Omnicom acts as principal — owner and reseller of media inventory rather than agent for the client — in experiential marketing and in parts of media buying. The Association of National Advertisers' March 2026 study of the practice found 90% of marketers now cite as their top concern whether recommended principal media is in their interest, up from 79% in 2024, with only 57% of companies operating any guidelines governing it. Integrated Media was 53% of Omnicom's quarter and grew above 10% organically; Advertising, at roughly 16% of revenue, fell high single digits.

That mix is why the merger, not the market, is the Omnicom story. Adjusted operating margin before amortization expanded two percentage points to 17.8%, and the company expects to deliver 75% to 80% of a $900m synergy target this year en route to $1.5bn annualized by mid-2028. Omnicom and Interpublic cut 8,200 roles across the two groups in 2025, against a stated aim of about 105,000 heads from 128,000 at end-2024. "We're now more of an operating company than a holding company," chairman and chief executive John Wren told investors on the July call. Consensus has 2027 revenue falling 3.3% as divestitures annualize while earnings per share still rise 13.5%.

One of the three is genuinely losing accounts

WPP's decline halved sequentially, from -6.7% in the first quarter to -2.8% in the second, and WPP Media ran worse than the group at -5.4% for the half. Net new business subtracted roughly six percentage points. The losses have names: Coca-Cola's North America media and the $1.7bn Mars account went to Publicis, while adidas and IBM global media went to Omnicom, which has won $4.1bn of new client billings in 2026. Global ad sales are forecast to rise 6.3% in 2026 and pass $1trn, so this is share changing hands, not a cycle. Chief financial officer Joanne Wilson warned on the 6 August call that "we expect second half margins to be down by as much as 200 basis points year-on-year." The shares still jumped about 25% in London that day, against a consensus that had penciled in a 6.5% decline; the New York listing gained 26% on five times its normal volume.

Stagwell grew organic net revenue 5%, with digital transformation up 18% to $107m at a 30% adjusted margin and record quarterly net new business of $171m — taken, Penn said, from incumbents including a 30-year holder of IBM's creative account. Adjusted earnings guidance went up to $1.17 from $1.03. The reported accounts disagree: operating income fell 50.2% to $11.5m and the company posted a net loss of $8.8m, with net leverage at 3.04x.

What the fee test settles

Fee income is not contracting across advertising agencies; it is contracting at one of these three. Omnicom earns its margin expansion, but part of its top-line growth is a gross meter carrying media costs that its rivals exclude, at the exact moment clients are asking harder questions about that practice. Stagwell is taking work, and its cash earnings are real even where its statutory ones vanish. WPP's improvement is a second derivative — the business is still smaller each quarter, only less so.

The market has barely begun to tell them apart. Omnicom trades at 8.1x forward earnings, against a trailing 45.3x made meaningless by a $941m merger charge, with a 10.3% free-cash-flow yield. Stagwell sits at 8.0x forward on consensus revenue growth of 12.1%. WPP is cheapest at 7.3x forward and 5.1x trailing enterprise value to earnings before interest, tax, depreciation and amortization — but its forward price-to-sales ratio of 0.43x sits above its trailing 0.31x, which is what it looks like when buyers pay more per pound of revenue because they expect fewer pounds of it.

WPP won Mars in 2018 and lost it this year, and management has said the account still has to come out of the fourth quarter.

AMN Healthcare filled more travel-nurse shifts in the second quarter than in any comparable stretch since 2022, and it did so without charging hospitals more per hour. Travel-nurse volume rose 6% year on year and allied-health volume 7%, both four-year highs, with average bill rates flat.

That combination is the first hard evidence in four years that the labor-arbitrage engine underneath hospital staffing has stopped deflating on price. It matters because almost nobody has marked it: consensus has AMN's 2027 revenue at $2.66bn, below the $2.73bn the company actually reported for 2025, against a book of orders management describes as growing at double digits. Astrana Health, which reported the same day, is running the opposite experiment — its revenue line barely moves and its entire improvement sits in the cost of care.

The price per hour stopped falling before the volume turned

AMN places travel and local nurses, allied clinicians, locum tenens physicians and interim hospital executives, and sells the vendor-management software hospitals use to buy that labor. Its annual revenue fell 27.7% in 2023, 21.3% in 2024 and 8.5% in 2025 — a decelerating collapse as hospitals rebuilt permanent staff and pushed contingent hours into managed-service programs that squeeze the spread on every shift. The clinician's own pay tracks it: the average US travel nurse earned just under $2,300 a week in 2025, down 42% from the pandemic peak.

Grace put the floor in plain terms on the August 6 call: "pre-COVID, you would have seen that premium of contingent to permanent labor be in the mid- to high teens. During COVID, you got up to 100% premium just because of the significant spike in demand. We're now back down into the mid- to high single digits." On orders, she said: "Year-over-year, travel nurse orders turned positive in May and accelerated in June. As of early August, the improvement continued with orders up about 40% year-over-year and 20% higher than August 2024."

The recovery is one segment deep. Nurse and Allied Solutions revenue was $422m, up 11%; Physician and Leadership Solutions fell 6% to $165m and Technology and Workforce Solutions fell 15% to $87m. Reported 2026 figures also flatter: the first quarter carried $722m of one-off strike-coverage revenue, and revenue round-tripped to $673.2m in the second. Brian Scott, chief financial and operating officer, told investors on August 6 that residual items "added about $27 million to revenue, 290 basis points to our consolidated gross margin and 370 basis points to our adjusted EBITDA margin." Net leverage is 1.5x, and the company refinanced in October 2025, issuing $400m of 2031 notes to retire the 2027s.

AMN trades at just under 5x trailing enterprise value to EBITDA against Pediatrix's 9.5x. Its headline forward price-to-earnings of 10.2x is an artifact of the strike quarter; on 2027 consensus earnings of $0.94 a share the same price is nearly 36x. The market and the sell side are not describing the same company.

Astrana's improvement is in the denominator

Astrana takes capitated payments from Medicare Advantage, Medicaid and commercial plans across roughly 1.5 million value-based lives and keeps what it does not spend on care. Gross margin — the inverse of its blended cost ratio — has widened three straight quarters, from 8.27% to 10.70%. Full-year adjusted EBITDA guidance rose to $255–280m and net leverage fell to 2.26x, ahead of the company's own sub-2.5x commitment.

But quarterly revenue has gone $956m, $951m, $965m, $973m — up 1.7% across a year. The 48.5% year-on-year headline is Prospect Health, acquired on July 1, 2025, where synergies are tracking to the high end of a $12–15m target. Only 42% of membership sits in full risk against 81% of capitation revenue, which management frames as conversion upside. Chief executive Brandon Sim said roughly three-quarters of first-half outperformance went back into new provider partnerships: "We believe that allocating some of our outperformance towards these growth opportunities is among the highest return capital allocation decisions available to us." The policy backdrop helped — CMS finalized a 2.48% growth rate for 2027 and declined to recalibrate the v28 risk model now fully in the base — while California's Medicaid work requirement lands January 1, 2027, with 4.8 million Medi-Cal enrollees in the affected category. At about 9.2x guided 2026 EBITDA against 14.0x trailing, most of the cost-ratio repair is already priced.

The other two names in the same corner of care facilities frame the range. agilon health is up 476.5% over six months but still loss-making, with negative trailing EV/EBITDA and consensus showing a $34m 2026 loss — a distress recovery, not an earnings re-rating. Pediatrix, the neonatology group, grew revenue 4.0% on 4% same-unit pricing while volumes fell, and anchors the group at 9.5x.

What each rally has earned

Astrana's advance is paid for: three quarters of margin, a guidance raise, a turn of leverage retired. What it has not bought is growth — strip Prospect out and the book is flat, and the multiple already assumes the cost ratio keeps improving. AMN's is the unpaid one, and that is why it is the more interesting. Its move rests on a single segment, one clean quarter and an order figure only management has published. If those orders convert, the 2027 revenue line the market has penciled in is wrong by a wide margin; if they do not, a company at 5x EBITDA is simply cheap because it is still shrinking. Note that AMN's chief competitors did not consolidate: Aya Healthcare walked away from its $615m purchase of Cross Country in December after the Federal Trade Commission objected, so nothing structural has tightened pricing.

A 40% order book and a forecast of shrinking revenue cannot both survive two more quarters. The third-quarter print is where they meet.