DK Street Journal

On 26 May, Verra Mobility told investors that Avis Budget Group had served notice to terminate the agreement under which Verra processes tolls and traffic violations for Avis's rental fleet. The next morning the stock opened at $5.51 against a $13.08 close, finished the day at $3.85, and traded 75.5 million shares — roughly forty times normal volume. The chief executive was gone by 1 June.

Verra Mobility, based in Mesa, Arizona, is two businesses. Government Solutions installs and operates automated speed, red-light and school-zone cameras for municipalities, and processes the citations they generate; it handles violations for New York, Chicago, Washington DC and Phoenix, and Verra owns both ATS and Redflex, formerly its two largest US enforcement rivals. Commercial Services bills rental-car and fleet operators for managing the tolls and tickets their customers incur. Avis is a Commercial Services customer. It is not a demand problem: Avis is replacing Verra's service with its own technology, insourcing an ancillary revenue stream it had outsourced.

What actually broke

Avis came back on 28 July with agreed terms for a new seven-year contract — on terms the company itself describes as materially less favorable, and with Avis retaining the option to perform some of the work internally anyway. Hertz then renewed early, also repriced. On 5 August, interim chief executive Jon Keyser cut full-year guidance to $945-965m of revenue and $360-370m of adjusted EBITDA, the second reduction since May. Free cash flow is guided to $105-115m. Consensus now models 2027 revenue falling 3% and earnings per share falling 14%.

Set that against the quarter itself. Revenue grew 11.7% to $263.6m. Government Solutions grew 17%, added $25m of annual recurring revenue bookings for a trailing-twelve-month figure of $74m, won Los Angeles speed enforcement worth about $10m of recurring revenue, and was selected in all six California pilot cities under the state's AB 645 speed-camera law. Net leverage is 2.4x with the revolver entirely undrawn. The reported operating loss is a $104m non-cash write-down of the T2 Systems parking unit, not trading deterioration.

The regulatory tail risk that usually haunts camera enforcement is, for now, pointing the other way: New York extended its school-zone speed program and widened automated enforcement to bridges and tunnels. The real 2027 pressure is pricing, not prohibition — management guided Government Solutions margins to the low 20s, down 450-500 basis points, on New York City rate normalization and subcontractor requirements.

Where the price sits

The shares trade at 0.73x trailing gross profit and 6.3x trailing EV/EBITDA. On guided EBITDA and about $1.0bn of net debt, the enterprise is capitalized near 4.7x, against roughly 13-14x a year ago. The trailing free-cash-flow yield is 13.7%. Earnings multiples flatter and mislead here because of the impairment and the debt, but even the forward figure — under 4x guided adjusted earnings per share — sits where the market prices terminal decline, not a business whose larger segment grew 17%.

The company it keeps

Verra is filed alongside two names with nothing economically in common with it. Nayax, an Israeli supplier of payment terminals for vending machines, EV chargers and car washes, earns a take rate on every transaction its 1.55 million devices process. That engine is intact — average revenue per device up 13% to $251, take rate held at 2.62%, net revenue retention near 120%. But gross margin slipped to 46.9% and the company cut 2026 free-cash-flow conversion to 5-10% of EBITDA from about 40%, to fund fast-charger deployment. The stock fell in three consecutive sessions in August and is down 13.7% on the month. At 36.7x trailing EV/EBITDA and a 1.6% cash-flow yield, that de-rating has arithmetic behind it. Its closest listed comparable no longer exists: Cantaloupe was taken private for $848m in May.

Aurora Innovation, a Pittsburgh developer of self-driving software for heavy trucks, booked $2m of revenue last quarter against a $270m operating loss and carries a $12bn market value. It has $1.2bn of cash against guided quarterly burn of $190-220m, and fell 11.8% on 18 August on a wider loss and insider sale filings. It is a funded-runway story, not a per-transaction one.

Verra's own month tells the same story of concentration: the stock is up 5.9% over thirty days, all of it a 21.7% jump across three sessions on the Avis framework news. Strip those and the month is negative. Nothing has begun to mend; one headline arrived.

The setup

Where it stands — A repriced customer contract cut a few points of revenue and four-fifths of the equity value. Would confirm — Government Solutions recurring-revenue bookings continuing above $20m a quarter with 2026 free cash flow landing inside $105-115m. Would invalidate — Avis exercising its option to insource further work, or a second Commercial Services customer serving notice. Watch next — Third-quarter results in early November, and the signing of the definitive seven-year Avis agreement. Valuation — 0.73x trailing gross profit and 6.3x trailing EV/EBITDA, about 4.7x on guided EBITDA, versus 13-14x a year ago.

Copper for immediate delivery in London set a record this week. Metal available now costs $545 a tonne more than metal deliverable in three months — the widest such premium since the 2021 squeeze — and London Metal Exchange warehouse stocks have fallen by almost half since mid-May, according to exchange data. Cash copper printed $14,912 a tonne on Wednesday.

The miners moved further than the metal. Over the past month Teck Resources, the Vancouver producer that sold its steelmaking coal business and is now copper-led, gained 21.3%; Hudbay Minerals, a Toronto mid-cap mining copper concentrate in Manitoba and Peru, gained 34.1%; and Freeport-McMoRan, the Phoenix giant that operates the Grasberg district in Indonesia and Morenci in Arizona, gained 18.6%. Copper itself rose about 11% over the same stretch. The gap is the story.

Four sessions did most of the work

Remove each company's two best days and the advance mostly evaporates: Teck drops to +3.8%, Freeport to +5.4%. Only Hudbay survives largely intact, at +14.8%. And the big sessions were not company events. On 21 July the entire complex rose together as the premium of US futures over London metal doubled toward roughly $600 a tonne — Teck, Anglo American, Glencore, Freeport, Southern Copper, Rio Tinto and BHP all climbed. On 6 August the Democratic Republic of Congo's outright ban on copper and cobalt concentrate exports became public, and US futures hit an all-time $6.77 a pound the next day. This week's move came as one participant was shown holding a London cash position equal to 30-40% of registered inventories, with two others between 50% and 80% — a positioning squeeze as much as a demand event.

Teck: the earnings arrived first

Teck is the one name where the business outran the shares. Second-quarter revenue rose 78% to C$3.61bn, about $2.6bn, and gross margin widened to 44.3% from 23.3%. Adjusted EBITDA tripled to $2.2bn at a record 61% margin. Copper output rose 25%, with net unit cash cost improving 19% to $1.64 a pound after by-product credits, and full-year guidance of 455,000-530,000 tonnes was left unchanged.

That delivery compressed the multiple. Teck trades at 9.0x trailing gross profit against roughly 10.3x on 1 May, because trailing gross profit grew 30% while the shares rose 14%. Forward earnings put it at 11.0x, below its trailing 17.8x. The caveat is corporate: the Anglo American merger that underpins the equity story is not closed, with final approval expected between September and March and Chinese clearance outstanding.

Hudbay: gold is doing the arithmetic

Hudbay's headline consolidated cash cost was minus $0.40 a pound last quarter — the company is, on paper, paid to produce copper. That number is manufactured by by-product credits: gold was 38% of gross revenue, and Hudbay attributed its improved full-year cost guidance to those credits offsetting external cost pressure. Sustaining cash cost was $1.39 a pound. Gold rose 3.66% to $4,493 an ounce on Wednesday alone, tracking long-dated Treasuries, and bullion is up around 12% over the month while flat across three.

Underneath, the copper business is decelerating. Revenue growth slowed to 21.4% from 27.3%, and gross margin fell to 41.0% from 48.6% one quarter earlier. The multiple went the other way, expanding about 21% to 12.6x trailing gross profit. Its forward price/earnings of 18.3x sits above the trailing 16.3x, because consensus 2026 earnings of $1.50 a share are below the $1.69 already delivered. Hudbay's own copper growth — roughly 150,000 tonnes in 2027, Copper World from the second half of 2029 — is years out, and the feasibility study is expected to carry capex above the 2023 estimate.

Freeport: the case is entirely 2027

Freeport's reported quarter went backwards. Revenue fell 7.3% to $7.03bn, gross profit fell 15.5% and operating income fell 17.6%. The shares rose anyway, and the multiple expanded about 29% to 14.3x trailing gross profit; on enterprise value to EBITDA it stands at 11.7x, against 7.4x at both Teck and Hudbay.

What investors are paying for is volume that has not shipped. The Grasberg block cave doubled to 69,000 tonnes a day in June from 34,000 in April, second-half copper sales are guided more than 20% above the first half, and 2027 more than 20% above 2026. Management quantifies the price leverage at about $390m of EBITDA per 10-cent move in copper. The tariff angle is thinner than the premium implies: refined cathode remains exempt, with a 15% duty phased in only from January 2027, and Freeport itself dates any benefit to "2027 or later."

The structural demand is real — data centers consume far more copper per rack than conventional servers, and S&P projects global demand rising 50% to 42m tonnes by 2040 against a deficit this year. But none of it explains four sessions in a single month.

The setup

Where it stands — Copper is at a record on a London squeeze; Teck's quarter supports its move, Hudbay's and Freeport's do not. Would confirm — Freeport's second-half copper sales land more than 20% above the first half, as guided. Would invalidate — The London cash premium collapses and copper retreats toward $13,400 a tonne while these multiples hold. Watch next — Hudbay's Copper World definitive feasibility study, due early in the fourth quarter of 2026. Valuation — Teck 9.0x trailing gross profit, 6.0x forward; Hudbay 12.6x; Freeport 14.3x, against 11.1x on 1 May.

Figma reported second-quarter results on 5 August that beat its own guidance and lifted the full-year outlook. The shares fell 14.85% the following session. The reason was not demand — it was what the company now pays to serve it.

A beat that cost five points of gross margin

Figma sells a browser-based canvas where design and product teams draw interfaces, prototype them and hand them to engineers, billed per editor and per developer seat. Revenue in the quarter ended 30 June was $370.1m, up 48.2% — the fourth straight quarterly acceleration, from 38% a year earlier. Paid customers spending more than $10,000 a year rose 34% to 15,964, and those above $100,000 rose 46% to 1,635, the company disclosed. Net dollar retention — what existing customers spend this year against last — was 136%.

Against that, cost of revenue more than doubled to $60.5m, and reported gross margin fell to 83.66% from 88.83%. Gross profit grew 40% while revenue grew 48%. That gap is the whole story of the quarter. Figma's newer products — Make, which turns a prompt into a working prototype, and an in-product agent — are metered in credits, and every credit call runs a model on rented accelerators. A seat carried no variable cost; a credit does. More than 80% of Figma's $10,000-plus customers now use AI credits weekly. Management says it routes across model providers and is shifting work to its own models, and warns margin will stay volatile while new products scale ahead of monetization.

Two other numbers frame the quarter. Figma guided the current quarter to roughly $373m, or 36% growth — a visible step down from 48%. And its reported operating loss of $117.3m sits against a non-GAAP operating margin of 10%, a gap of about $154m, or 42% of revenue, driven principally by stock compensation. A lock-up releasing 77.7m shares, some $1.86bn or roughly 15% of the float, opened on 8 August; the stock is up about 15% since, so the supply cleared into strength rather than crushing the price.

Adobe has the same disease and a tenth the multiple

Adobe sells the same creative professionals a bundle — Photoshop, Illustrator, Acrobat — on annual subscriptions, plus marketing software to enterprises. Fiscal second-quarter revenue was $6.618bn, up 12.7%, itself an acceleration from 10.5% two quarters earlier. Ending annual recurring revenue reached $27.1bn, and AI-first recurring revenue crossed $500m, roughly tripling. Diluted shares fell 7.2% year over year against a fresh $25bn buyback authorization.

And the margin told the same story as Figma's. Operating margin fell to 33.8% from 35.9%, and net income grew 1.2% on 12.7% revenue growth. Generative features are bundled into subscriptions whose price did not rise, so cost per subscriber climbs while revenue per subscriber does not. Adobe is also mid-transition at the top: Shantanu Narayen said in March he will step down once a successor is installed, with fiscal third-quarter results due around 10 September.

The competitive threat is real and dated. Anthropic launched Claude Design in April, a conversational prompt-to-prototype tool aimed squarely at Figma's estimated 80–90% share of interface design. Figma's answer is that professional work needs a fast canvas and direct manipulation, and that its file becomes the context agents read from — calls writing into Figma through its model-context protocol grew 75% in a quarter. Adobe's is file-format dominance in the professional tier.

What the two prices assume

Figma is at 8.88 times forward sales and 11.2 times forward gross profit, with a forward price/earnings of 93.6 on losses and a trailing free-cash-flow yield of 1.77%. Consensus models it growing 24% next year with no path to reported operating profit through 2029. Adobe is at 11.2 times forward earnings and 4.08 times forward sales, the lowest multiple in over a decade against a five-year median near 41.6, with a 9.8% free-cash-flow yield — about five and a half times Figma's.

The advance that produced both is narrower than it looks. The two bottomed in the same session, 25 June, and Figma's trend turned positive on 7 August, Adobe's on 18 August — the day money rotated out of semiconductors into beaten-down software. Remove each stock's two best sessions and Figma's 11.5% month becomes minus 7.8% and Adobe's 16.1% month becomes minus 3.8%. Over six months Figma is up 2.7%. The businesses mended steadily; the prices moved in a handful of days.

The setup

Where it stands — Both companies are growing faster than a year ago and earning less per dollar of revenue, and only Adobe's multiple reflects the doubt.

Would confirm — Figma's gross margin recovering above 85% on a reported basis while third-quarter revenue growth holds at or above the 36% guided.

Would invalidate — Net dollar retention falling below 130%, or Adobe's annual recurring revenue growth slipping under 10%.

Watch next — Adobe's fiscal third quarter, due around 10 September, with a chief executive successor expected to follow.

Valuation — Figma: 10.2x trailing, 8.9x forward sales. Adobe: 15.6x trailing, 11.2x forward earnings, against a five-year median near 41.6x.

The most valuable input Corsair Gaming sells is not made by Corsair. Its Vengeance and Dominator memory kits are DRAM chips bought from the three companies that fabricate them, screened, clipped into a heatsink and resold to people building their own PCs. That makes Corsair — a Fremont, California maker of gaming keyboards, mice and headsets, Elgato streaming gear and DIY PC components — one of the few consumer names with direct revenue leverage to the memory cycle. In the June quarter, the leverage worked in reverse.

Revenue in Gaming Components and Systems, the larger of Corsair's two segments, fell 9% year on year to $198.5m. Memory revenue within it rose 17%. The gap is the whole story: units are going down while prices go up, because the enthusiast staring at a $550-to-$700 sticker for a 32GB DDR5 kit — roughly $9 to $14 per gigabyte, up 40% to 60% from the 2024 floor — waits. Corsair's management calls that deferral rather than destruction, arguing that buyers waiting for cheaper memory will give up waiting because memory is not getting cheaper. It probably is not: server DRAM contract prices were set to rise 13% to 18% in the third quarter and to keep climbing through 2027 as fabs redirect capacity to AI systems.

The cost side is guided down too. Corsair told investors memory gross margin in that segment goes from about 23% in the September quarter to the high teens in December. Its edge in components is procurement and shelf space, not technology: Kingston, G.Skill, TEAMGROUP and Crucial buy the same die for the same socket.

What is actually branded

The other half is better. Gamer and Creator Peripherals grew 13% to $115.9m at a 44.9% gross margin, up from 40%, led by Fanatec sim-racing hardware. Elgato Marketplace revenue and transactions each more than doubled, and a minority stake in show-control software maker Bitfocus makes the Stream Deck the default control surface in professional live-events rigs. That is where switching costs live.

Group revenue still fell 1.8% to $314.3m, a second straight decline. Reported gross margin hit a record 33.2%, but roughly 500 basis points of that was a one-off $15.6m tariff refund; underlying is nearer 28%. Cash flow was real — operating cash flow up 148%, and $75.1m of net cash.

The shares rose 35.25% on 7 August, the day after the beat and raise. Every bit of the month's advance is that one session; strip it and the same 30 days are down about 11%. They have fallen 20.6% from the 7 August close, and on 19 August Goldman Sachs initiated coverage at Sell with an $11 target, citing AI build-outs competing for the same memory supply and a forecast 15% drop in industry PC shipments this year. Price-to-trailing-gross-profit has expanded about 69% since mid-May, to 2.65x from roughly 1.57x, on 36.2x trailing earnings — while consensus has fiscal 2026 revenue falling 2.0%.

Logitech pre-bought the problem

Logitech, the $14.4bn Swiss maker of mice, keyboards and Microsoft Teams and Zoom room systems, has the same input and none of the exposure. It secured memory through fiscal 2027 and raised video-collaboration prices 13% in May to cover the cost. June-quarter revenue grew 6.9% to $1.227bn, gross margin reached 49.5% (44.8% excluding tariff refunds) and operating income rose 59.5%. Pointing devices grew 14%; video collaboration grew 9%, against an installed base of meeting rooms now six years old and a refresh cycle of five to seven. Fewer than half of PCs have a mouse — the growth mechanism is attachment and premiumisation, not PC unit shipments, which is why Corsair's headwind is not Logitech's.

The stock went nowhere anyway, down 1.2% over 30 days and 17.5% below its 1 June close. The cause is specific and disclosed: a semiconductor supplier's plant shut in late June, costing about $20m of September-quarter sales and up to $200m in December, with no reopening date. Price-to-gross-profit has de-rated roughly 21% from the June high to 6.52x, and forward earnings of 17.9x sit below trailing 18.3x — consensus now models fiscal 2027 revenue up 1.2%. Rivals Yealink and HP's Poly are pressing on room-system share while the outage runs.

One of these companies is priced for a memory cycle that is shrinking its unit base. The other has contracted its way around the cycle and is discounted for a factory it does not own.

The setup

Where it stands — Corsair has given back most of a one-day earnings pop; Logitech is growing and de-rated on a supplier outage. Would confirm — Corsair's components segment returns to growth in the September quarter as deferred builds convert. Would invalidate — Corsair's memory gross margin lands below the guided high teens, or Logitech's outage extends past fiscal Q4. Watch next — Logitech's fiscal Q2 report in late October, guided to 0-3% constant-currency growth and roughly 44% gross margin. Valuation — Corsair: 36.2x trailing, 15.8x forward earnings; 2.65x gross profit against 1.57x in mid-May. Logitech: 18.3x trailing, 17.9x forward.

A cut that spared the loan book

Klarna Group, the Swedish-founded, London-headquartered payments company that lends shoppers small sums at checkout and funds itself as a licensed retail bank, delivered a good quarter and a bad year on 18 August. Second-quarter revenue was $1.042bn, up 27% and above its own guidance range, with net income of $9m against a $53m loss a year earlier — the first profit in its short life as a listed company. The stock fell 22% that session on 34.4m shares, against a normal day near 2m.

The damage came from the outlook. Klarna took full-year revenue guidance to $4.08–4.16bn from $4.34bn, against $4.42bn of consensus, and gross merchandise volume to $149–151bn from more than $155bn. Management attributed the reduction to roughly $600m of negative currency translation and a more cautious view of Germany, its largest market, where retail sales slowed. Its chief financial officer and chief marketing officer will both step down in early 2027.

What did not happen is the thing the share price implies. Provisions for credit losses were 0.52% of volume, better than 0.56% a year earlier, with provision dollars up 11% against volume up 18%. Klarna simultaneously raised its full-year transaction-margin guide to $1.62–1.65bn — the gross-profit line went up while the revenue line went down. Roughly 90% of its book is funded by consumer deposits, which is why the long end of the Treasury curve, where the 30-year touched 5.323% on the day of the crash, a 19-year high, does not reach its cost of money.

Affirm's advance was bought, not earned

Affirm, the San Francisco lender that spreads purchases over one to 48 months across some 29,000 integrated merchants and keeps the credit risk on its own balance sheet, is the direct competitor for the same American checkout. It is also, on the last comparable quarters, winning it: Affirm's volume grew 35% against Klarna's 18%, and Klarna's US volume grew 27%. Five firms account for more than 95% of US buy-now-pay-later volume, with banks including JPMorgan now bolting on their own instalment features.

Affirm's credit is flat to better. At 31 March, 30-day-plus delinquencies on US monthly installment loans stood at 2.8% versus 2.7% in December, and the 90-day-plus rate improved to 0.7%. Revenue less transaction costs — the company's own gross-profit measure — grew 41% to $498.2m, faster than volume, lifting the take rate to 4.3%. Funding got cheaper rather than dearer: costs fell about 125 basis points year over year, and the 2026-2 securitization was upsized to $750m from $500m and more than twice oversubscribed.

The soft spot is price. Affirm has risen 18.3% in three months on a trailing gross-profit base that has not changed since the 7 May report, taking price to trailing gross profit from roughly 8.1x to 9.6x, or 7.3x forward. Forward earnings sit at 43x. Nothing in the business deteriorated; the shares simply got more expensive per dollar of the same profit. The company guided fiscal fourth-quarter volume of $13.15–13.45bn and reports on 27 August.

The one with no borrowers

Remitly, the Seattle app immigrants use to send money home across nearly 150 countries, is grouped with these two and shares almost nothing with them: it earns a foreign-exchange spread and a fee per transfer, and holds no consumer loans. Second-quarter revenue was $495.2m, up 20%, on send volume of $23.5bn and 10.2m active customers. It raised full-year guidance. The new US remittance tax, widely read as a threat, pushed senders toward digital channels and produced a record quarter of customer acquisition. Its trailing price/earnings ratio of 17.3x flatters: second-quarter net income included a $140.6m one-off tax benefit, which is why forward earnings, at 24.6x, are more expensive than trailing.

The macro backdrop does not carry a credit story either. The New York Fed's 11 August report showed 4.7% of household debt delinquent, a slight improvement, with credit-card delinquency transitions easing to 8.6%. On the crash session, Affirm traded roughly flat and PayPal higher; Affirm closed 5.0% higher the next day. Klarna now trades at 1.26x forward sales against Affirm's 4.91x and Remitly's 2.66x — a gap that is either a verdict on European volumes or a mispricing of the same instalment product.

The setup

Where it stands — Klarna's break was a currency-and-Germany volume event; its credit metrics and gross-profit guide both improved, and neither peer followed it down.

Would confirm — Affirm's 27 August print landing revenue less transaction costs inside its $535–550m guide with 30-day-plus delinquencies at or below 2.8%.

Would invalidate — Klarna's third-quarter provisions rising above 0.56% of volume, or Affirm's take rate falling below 4.0%.

Watch next — Affirm fiscal fourth-quarter results, 27 August 2026; Klarna guided third-quarter revenue of $940–980m.

Valuation — Affirm 9.6x trailing and 7.3x forward gross profit, against about 8.1x in May; Klarna 1.26x forward sales.

The June quarter separated Appalachia's gas producers more than any move in the gas price did. CNX Resources, which pairs shale drilling with coalbed-methane wells and its own gathering pipelines, grew revenue 14.3% to $618.5m and widened its operating margin to 38.3% from 35.2%. EQT, the basin's scale producer at a $33.6bn market capitalization, saw revenue fall 29.2% and operating income fall 60%, with operating margin narrowing to 25.1% from 44.3%. Gulfport Energy, a dry-gas driller in the Utica shale and Oklahoma's SCOOP play, missed and posted a gross margin of 29.1% against 70.7% a year earlier.

The split is the barrel, not the balance sheet

The cleanest divide across the eight is not leverage or hedging. It is whether a company sells liquids. Forward price-to-earnings sits below trailing at only three names — Antero Resources at 8.86x against 10.56x, Range Resources at 9.82x against 11.08x, and Infinity Natural Resources at 4.45x against 4.72x. All three carry oil and natural gas liquids in the mix. At every dry-gas producer the relationship inverts: Expand Energy 10.48x forward against 8.22x trailing, EQT 12.67x against 11.79x, Gulfport 7.43x against 6.87x. Analysts are modeling 2026 earnings down at the gas-only names and up at the liquids names.

The reason is supply. Henry Hub futures fell nearly 15% in July, to $2.75 per million British thermal units, and the Energy Information Administration cut its 2027 forecast to $3.18 from $4.60, with inventories expected to end October about 7% above the five-year average. Expand Energy's management traces it to associated gas riding out of the Permian Basin on 3.5 billion cubic feet a day of new pipeline capacity by year-end, which it says keeps the market oversupplied through at least the first half of 2027. The offset is dated and local: EQT has signed 1.5 Bcf/d of in-basin supply agreements this year for two data-center-linked power projects, and its finance chief says Appalachian basis differentials should tighten through the end of the decade.

Range is the clearest beneficiary today: it captured a $3.49 per barrel premium over the Mont Belvieu benchmark on its liquids in the quarter and raised full-year guidance to a $2.50 premium. Antero has hedged 34% of 2027 gas at $3.84 and has made completion of its next two dry-gas pads contingent on locking in $3-plus per thousand cubic feet equivalent for 2027. Infinity, a 2025 listing producing 348 million cubic feet equivalent a day, has 78% of remaining 2026 volumes hedged — the heaviest near-term coverage disclosed in the group.

Gulfport is shrinking itself

Gulfport's diluted share count fell from 19.4m in the December quarter to 17.9m in June — roughly 7.5% of the company retired in two quarters, about $242.8m of repurchases against a $3.11bn market capitalization. It is doing this while earnings fall, funded by roughly 1.0x net debt to EBITDA, the lowest in the group; S&P Global Ratings upgraded the company to BB- from B+ on that basis. What knocked the shares down 7.4% in a single session on 6 May was not a miss — first-quarter earnings beat at $8.87 a share — but the disclosure of an $83m state land auction and a $140m discretionary leasehold budget competing with the buyback. Its trailing free-cash-flow yield of 8.0% is the lowest among the drillers for exactly that reason, and management calls 2026 the peak year for that spend.

Gulfport now trades at 4.30x trailing enterprise value to EBITDA against EQT at 6.31x, Antero at 6.70x and Range at 7.30x. The moving-average trend that turned down in mid-May has still not repaired, though the shares have led the group higher since 20 July. The business and the trend disagree in both directions at once.

What the non-gas revenue actually pays

CNX is the one name with a real stream that does not track Henry Hub. Its revised model for the 45Z clean-fuel production credit values it at about $40m a year, and it targets roughly a $90m annual run rate from environmental attributes by 2027, with $30m already monetized in early July. That is about 9% of consensus 2026 EBITDA of $963m — material, but management notes it lands on the tax expense line, not EBITDA. CNX generated $138m of free cash flow in the quarter, its 26th consecutive positive quarter, and bought 5.6m shares for $199m.

Diversified Energy is the capital-structure outlier the thesis expected Gulfport to be. It does not drill; it buys mature producing wells and runs them for cash, financing them with securitized asset-backed notes — 76% of its debt, one tranche carrying a 7.076% coupon amortizing to 2033. Leverage rose to 2.45x from 2.2x in the March quarter. Its 1.68x trailing price-to-earnings is a derivative artifact; the usable anchor is 3.03x EV/EBITDA, the cheapest of the eight. The quarterly dividend of $0.29 annualizes near 8%, and sits behind both securitization principal and the plugging costs of retiring old wells in the cash waterfall. Consensus models 2027 revenue down 5.9% and earnings per share down 16.8%.

The setup

Where it stands — Estimates for 2026 fall at every dry-gas Appalachian producer and rise only at the three liquids-weighted names. Would confirm — Third-quarter liquids realizations at Range and Antero hold their premium to Mont Belvieu while dry-gas revenue keeps declining. Would invalidate — The 2027 Henry Hub strip recovers toward $4, lifting dry-gas earnings estimates back above trailing. Watch next — Gulfport's third-quarter report, with second-half liquids volumes guided more than 50% above the first half. Valuation — Gulfport at 4.30x trailing EV/EBITDA against EQT 6.31x and Range 7.30x; forward price-to-earnings 7.43x versus 6.87x trailing.

A charger business waiting on a rack that slipped

Navitas Semiconductor, a fabless designer of gallium-nitride (GaN) and silicon-carbide (SiC) power chips sold into chargers, appliances, motor drives and, lately, data centers, told investors on 27 July that artificial-intelligence infrastructure would account for more than a third of its sales by the end of this year. The quarter it was reporting carried $10.5m of revenue, down 27% from a year earlier. Full-year 2025 revenue was $45.9m, itself down 45% from 2024, as the consumer-charger base eroded. Consensus has this year at $47.0m before a step up to $74.2m in 2027 and $124m in 2028.

The shape of that curve is the whole argument. Nvidia's proposed shift to distributing 800 volts of direct current through the rack, rather than converting to 12 or 48 volts upstream, exists because accelerator clusters now draw currents approaching 100,000 amps; at those levels the copper losses in low-voltage distribution become unaffordable, and single-stage conversion from 800V down to roughly 6V at the chip becomes mandatory. Navitas is one of the few listed companies with parts spanning the whole of that chain, from 100V GaN to 6,500V SiC.

But it is one of fourteen silicon suppliers Nvidia names in its 800VDC ecosystem, alongside Infineon, Texas Instruments, onsemi and Power Integrations. That is a place on a list, not an exclusive socket. Navitas has disclosed no dollar-quantified 800VDC design win and no named customer, describing instead "record backlog and book-to-bill extending beyond 2026." Its own dated milestones put 800V sidecar racks in mid-2027, GaN chip-level conversion in late 2027 and solid-state transformers from 2028.

Then the schedule moved. On 6 July, SemiAnalysis reported that Nvidia's Kyber rack — the platform carrying 800VDC to Rubin Ultra — had slipped more than a year, to 2028, on yields for a 78-layer orthogonal backplane circuit board. Nvidia said its roadmap is intact. The market did not wait: Asian board and substrate suppliers shed over $10bn of value that day. Five days earlier, Meta's announcement that it would resell spare AI capacity had already knocked the scarcity premium out of power-equipment shares.

What the accounts show

Add up Navitas's four quarters through June and gross profit comes to about minus $0.6m on $36.5m of revenue. There is no price-to-gross-profit multiple for the company because there is no gross profit to divide into. Reported gross margin last quarter was minus 9.5%, against management's guide of 39.7% on a non-GAAP basis for the September quarter, on revenue of $13.5m — a return to growth, and still less than Vicor books in a fortnight.

The funding tells the same story. Navitas raised $373m during the June quarter at $21.89 a share, after completing a $122m at-the-market offering on 12 May. Diluted shares are up 18.5% year over year. Cash stands at $557m with no debt — enough runway to reach 2028 — but the shares closed at $12.94, some 41% below the price at which that stock was sold.

Competitors are attacking the intellectual property directly. Wolfspeed sued Navitas in Delaware on 7 July over five patents covering essentially its entire GaN and SiC range. Renesas followed on 22 July with trade-secret claims naming chief executive Chris Allexandre; Navitas countersued in Texas on 10 August and has called the suits a harassment campaign.

Where the peers landed

The same architecture is producing hard numbers elsewhere. Bel Fuse, which makes magnetics, connectors, fuses and power supplies, grew revenue 25% to $210.7m last quarter, lifted gross margin 120 basis points to 39.9% and raised operating income 47%, with six consecutive quarters of bookings above sales — though it still earns more from defense ($66.5m) than from data centers ($58m), and trades at 27.7 times forward earnings. Vicor, which sells complete power modules, reported a one-year backlog of $379.7m, up 145%, at a 58.0% gross margin, and now sits at 28.8 times forward gross profit against 61.8x in mid-May. Navitas trades at 91.7 times trailing sales and 71.2 times forward; Bel Fuse trades at 4.3x and 3.9x.

The shares tell you the market has noticed. Navitas is 59% below its 26 May close of $31.79, and its uptrend broke at the end of July when the 50-day average crossed below the 200-day. Its 12% gain over the past month is two sessions — 30 July and 7 August — and nothing else; strip those and the month is minus 11.5%.

The setup

Where it stands — Navitas's 800VDC content is real engineering with no quantified revenue attached, and the ramp is dated to 2027-28. Would confirm — September-quarter revenue of $13.5m delivered with AI infrastructure disclosed above a third of sales. Would invalidate — Another quarter of negative reported gross profit, or a further equity raise below $12.94. Watch next — GlobalFoundries 8-inch GaN sampling, promised before year-end, with qualified production in early 2027. Valuation — 91.7x trailing and 71.2x forward sales; no gross-profit multiple exists, against Vicor's 28.8x forward.

Talen Energy, an independent power producer that sells electricity, capacity and grid services into US wholesale markets from roughly 10.7 gigawatts (GW) of nuclear, gas, coal and solar plants, told investors on 5 August that 2026 would be better than it had previously said. Management raised full-year adjusted EBITDA guidance to $2.03-2.23bn and adjusted free cash flow to $1.20-1.35bn. Second-quarter revenue grew 64.5% to $747m, and gross margin widened to 49.3% from 20.7% a year earlier. The shares have fallen about a tenth since.

The standard explanation for August weakness across power stocks is the long end of the bond market. Utilities are bought for their yield and fund their construction in the debt markets, so a rising discount rate should hurt them. It is a clean story. It is also the wrong one.

The session that separated the two explanations

On 18 August the 30-year Treasury yield touched 5.323%, its highest since 2007, on deficits, heavy corporate issuance funding AI construction and uncertainty about policy under new Federal Reserve chair Kevin Warsh. If duration were the mechanism, the regulated utilities would have led the decline. They did not move: NextEra Energy rose 0.36%, American Electric Power 0.19%, Dominion Energy slipped 0.14%. What broke were the merchants whose equity value rests on long-dated fixed-price contracts with cloud computing companies — Talen down 11.44%, Vistra down 4.19%, Constellation Energy down 4.09%.

The day before, Microsoft fell around 3% after Morgan Stanley warned on hyperscaler creditworthiness and the widening gap between what AI infrastructure costs and what it earns. A megawatt contracted for twenty years is worth what the buyer is good for. Talen carried its own catalyst too: Raymond James cut its price target to $449 and Oppenheimer to $400, both citing execution risk, with a $984m shelf registration adding dilution overhang. Both targets still sit far above the $322.58 close.

Nor is this the slow bleed it looks like in aggregate. Remove each name's two worst sessions from the past 30 days and Vistra's 9.7% decline turns into a 10.3% gain, Talen's 10.5% into 17.0%. Only NextEra's small 2.3% slide is genuinely spread out — worst single day, 1.57% — and NextEra runs an interest-rate hedging program of more than $46bn against precisely that risk.

The businesses went the other way

Vistra, which runs about 38,700 megawatts of gas, nuclear, coal, solar and battery capacity and sells retail electricity to some 4.3m customers, posted second-quarter adjusted EBITDA of $1.767bn, up 30%, with the generation arm up roughly 68%. It reaffirmed 2026 guidance of $6.8-7.6bn and held 2027 at $7.4-7.8bn even against softer Texas forward power curves. Neither its pending Cogentrix gas acquisition nor its 20-year Meta nuclear contracts covering more than 2,600 MW sits inside those numbers.

NextEra pairs Florida Power & Light, serving about 5.7m accounts, with the largest US clean-energy development business. Adjusted earnings per share rose 9.5%, the development backlog reached 35.1 GW, and the utility raised its expectation for large industrial load in Florida to 8 GW by 2032 from 6 GW — roughly $2bn of rate base per gigawatt, earning an allowed return. Guidance for 2026 through 2035 was left untouched.

The capacity market is not signaling weakness either. PJM Interconnection's auction for 2028/29 cleared at the $325 per megawatt-day price cap for the third straight year, and came in 6.8 GW short of its own reliability requirement, with the uncapped price estimated at $554.72. Revenue is limited by regulation, not by demand.

Where the prices now sit

Vistra trades at 15.7 times forward earnings against 18.8 times at its February price on the same $9.06 consensus estimate, and at 10.4 times trailing EV/EBITDA — the fitting anchor for a leveraged generator. Talen's trailing multiple is meaningless on a GAAP net loss of $92m; on forward earnings it is 15.2 times, against 20.5 times at its 18 June high on the identical estimate. NextEra has de-rated to 8.60 times gross profit from 11.73 times in mid-May, while gross profit per share rose; on management's own adjusted guidance the shares are near 21.5 times.

One control matters. Dominion, which votes on its all-stock combination with NextEra on 3 September, has held an uptrend for 75 unbroken sessions since late April, its 50-day average above its 200-day. This is not a utilities selloff. It is a repricing of who is on the other side of the contract.

The setup

Where it stands — Merchant generators with AI offtake have de-rated on counterparty credit while guidance rose; regulated peers are untouched.

Would confirm — Regulated utilities keep holding as long yields rise, while further hyperscaler capital-spending warnings knock Talen, Vistra and Constellation.

Would invalidate — A data-center counterparty renegotiating or walking from a signed power purchase agreement, or 2027 EBITDA guidance being cut.

Watch next — Dominion and NextEra shareholder votes on the merger, 3 September 2026; Texas's Batch Zero data-center audit resolves in roughly two months.

Valuation — Vistra 15.7x forward versus 18.8x in February; Talen 15.2x versus 20.5x in June; NextEra 8.6x gross profit versus 11.7x in May.

The rung nobody prices

Every chip that goes into an artificial-intelligence server has to be electrically exercised before it ships — probed at wafer level, heated, cycled and graded. That step used to be an afterthought in a fab's budget. Teradyne's management told investors on 29 July that test equipment has risen from 4% of global semiconductor capital spending in 2023 to 7% in 2025 and 8% in the first half of this year, and that the automated-test market could reach $20bn by the end of the decade against $12–14bn today. The reason is physical: a memory stack sixteen dies tall takes longer to validate than one eight tall, and warpage and thermal stress during cycling only get worse as HBM4 stacks grow.

The quarter just reported shows that arriving in revenue.

Four winners, one laggard

Advantest, the Japanese maker of memory and system-on-chip testers sold to fabless designers, foundries and test houses, grew revenue 39.3% in the June quarter at a 69.5% gross margin — an operating margin of 51.5%. On 29 July it lifted its full-year forecast to ¥1.714trn of revenue and ¥846bn of operating profit, implying about 70% operating-profit growth against a prior guide of 26%. That revenue figure sits above the ¥1.616trn analysts had modelled: the company is ahead of the sell side, not catching up to it. Management attributed the upgrade to AI test demand expanding beyond plan, with the memory business driven by high-bandwidth memory and the AI market broadening from training into inference.

Teradyne, which sells semiconductor and system test alongside the Universal Robots collaborative-robot arm, reported June-quarter revenue of $1.329bn, up 103.9%. Its Compute line grew more than 600% year on year and is now 70% of system-on-chip revenue; memory test set a record $212m with bookings running above twice billings. Operating margin went from 13.9% to 32.9%. The robotics arm that was supposed to be dead weight grew 33% to $100m. Advantest took system-on-chip tester share from 56% to 66% in calendar 2025, by one industry count; Teradyne is attacking socket by socket through dual-vendor qualification, a nine-to-twelve-month cycle whose share effects it expects to show only in 2027.

FormFactor sells probe cards — the consumable contactors that touch a wafer during test, a per-wafer stream rather than a capex approval. Revenue rose 31.9% to $258.2m and gross margin expanded 1,350 basis points to 50.7%, lifting operating income 370%. It set DRAM probe-card records on HBM4 and has qualified on Nvidia's Rubin processor. The razor-blade model cuts both ways: it guided September-quarter DRAM revenue flat as customers shift wafer mix toward conventional DDR memory.

Onto Innovation, in defect inspection and packaging lithography, grew 35.3% to $343.1m. Camtek, the Israeli inspection and metrology supplier aimed at advanced packaging, is the group's laggard on results: revenue up only 8.0%, operating income down 14.8%, margin compressed from 25.9% to 20.4%. Its order book argues the other way — more than $600m booked year to date, 80% of it advanced packaging, with September-quarter revenue guided to about 20% sequential growth.

The stocks got cheaper

The striking fact is not the run. It is that gross profit compounded faster than the share prices did. Teradyne trades at 22.42x trailing and 19.45x forward gross profit, against 25.15x trailing in mid-May and 32.04x in early May. FormFactor is at 21.76x trailing versus 34.35x in May, Camtek at 27.59x versus 34.42x. Advantest is the most expensive of the profitable four at 30.99x trailing, 23.63x forward. Gross profit is the comparable lens here because margins span 69.5% at Advantest to 40.6% at Aehr, which makes sales multiples meaningless across the group.

Aehr Test Systems, a 138-person Fremont company whose FOX platforms burn in and test whole wafers at once, has no analogue in that range: 199.5x trailing and 80.66x forward gross profit. Its fiscal year to May saw revenue fall 15.2% to $50.0m with a $14.1m operating loss, as the silicon-carbide burn-in business that once carried it collapsed. The pivot is real but sits in the order book: fourth-quarter revenue rose 40.2%, bookings hit a record $60.7m and effective backlog $100.6m against $116.5m of cash, and guidance for the current year is $130–150m. On 12 August a lead AI-processor customer placed a $22m follow-on order for systems testing nine 300mm wafers apiece. Consensus already carries $125.9m for the year — the pivot is fully marked in.

Three sessions took a fifth off

Between 17 and 19 August, Aehr fell 25.9%, FormFactor 17.0%, Camtek 15.0%, Onto 14.7%, Teradyne 14.3% and Advantest 6.2%. No company-specific catalyst was reported for any of them, and the broad semiconductor exchange-traded fund fell only 2% on the 19th. What moved was the discount rate: the 30-year Treasury yield touched 5.33% on 18 August, a 19-year high, having added more than 60 basis points in a month. Aehr's break is not the first name where the pivot failed — it is the highest-multiple name in a group being repriced on duration.

The gains were fragile in a specific way. Strip each name's two best sessions from the past 30 days and all six turn negative, Teradyne from +13.75% to −11.30%, FormFactor from +9.32% to −21.71%. Four of the six had their best day on 30 or 31 July, the sessions immediately after Teradyne and Advantest reported. A month's return delivered on two earnings reactions is a month's return that a bond move can take back.

The setup

Where it stands — The testers' results confirm rising test intensity, but every name sits well below its 52-week high after a rate-driven, cohort-wide unwind. Would confirm — Advantest's next quarter tracking toward the raised ¥1.714trn full-year revenue guide, ahead of the ¥1.616trn consensus. Would invalidate — Teradyne's flagged reversion of test to 6–7% of semiconductor capex in 2027, or memory book-to-bill falling below 1. Watch next — Aehr's fiscal first-quarter report, the first test of whether $100.6m of backlog converts toward the $130–150m guide. Valuation — Group forward price-to-gross-profit of 19.0x to 25.9x, down from 34x in May; Aehr at 80.66x forward, 199.5x trailing.

FactSet Research Systems sells the workstation an equity analyst opens each morning: data, models, screens and order workflows, billed per seat at roughly $12,000 to $24,000 a year. That is comfortably below a Bloomberg Terminal's roughly $32,000 per user, and it is precisely the product a competent AI research assistant is supposed to render unnecessary. That fear drove the Norwalk, Connecticut company's shares to multi-year lows earlier this year.

The disclosures point the other way. Organic annual subscription value — the contracted run-rate, the only clean demand measure for a business billed in seats — reached $2.49bn at 31 May, up 7.1%, the fifth consecutive quarter in which that growth rate accelerated. Retention held above 95%, and enterprise renewals lengthened by 30% on average in the quarter. Users and clients rose rather than shrank: 241,352 users at the end of February, a net 1,489 more in three months, plus 98 net new clients. Management raised the fiscal 2026 subscription-value target on 31 March, to $130-160m of net additions from $100-150m, and reaffirmed it on 1 July.

The bill arrives on the margin line

What FactSet is not doing is converting that demand into profit. Quarterly revenue grew 6.4% to $622.9m; operating income fell 14.3%. Operating margin went from 33.2% to 26.7% and gross margin from 52.1% to 49.9%, the arithmetic of embedding AI features into a product whose price per seat cannot rise as fast as the cost of running it. More than 90% of FactSet's fifty largest clients now use four or more of its AI products — adoption that shows up in the subscription line and in the cost line, not yet in earnings.

The shares have repaired about half the de-rating. Against consensus fiscal 2026 earnings of $17.79 a share, FactSet traded at 21.0x a year ago, 11.9x at its 15 May low and 16.9x now — still a fifth below where it started. Consensus models 6.5% revenue growth this year, below the organic subscription growth just reported, an unusual position for a company the market has been treating as structurally impaired. The average sell-side target has been cut toward $305 even as the stock rose to $301.59; Jefferies lifted its target to $253 and kept a Hold.

The one with the best numbers fell

MSCI licenses the benchmarks that index funds are built on, so it is paid a fee on assets, not on analysts. Its asset-based fee run rate hit a record $948m, up 25%, on roughly $40bn of second-quarter ETF inflows and $2.8trn of ETF assets tracking its indices. Revenue grew 12.2% to $867m at a 56.2% operating margin — the strongest of the three by both measures. The shares fell 10.1% in the 21 July session anyway, because the company raised its full-year expense guidance to $1,340-1,370m from $1,305-1,335m for the First Street acquisition and incentive pay. That single day is the entire month's decline. The result is a stock unchanged over twelve months on a forward multiple of 28.8x, identical to a year ago, while consensus 2026 earnings for it rose about 14% — a full year of growth accruing to the holder as nothing. Trailing, 31x compares with a ten-year average near 41.6x. The soft spot is sustainability data, where management guided net new sales to roughly zero for two quarters; private-assets subscriptions are growing past 16%.

Morningstar, the smallest at $8.0bn, is the one whose improvement is real and whose source is misread. Revenue rose 9.6% to $663.2m and operating margin widened from 20.7% to 24.2%. But Morningstar Credit — the DBRS ratings business — grew 23.4% to $104.9m on structured-finance issuance and supplied about a third of all revenue growth from a sixth of revenue, while PitchBook, its seat-billed private-markets platform, grew 4.9% amid corporate churn. The fastest-improving investment-data company is being carried by the debt-issuance cycle.

What the month actually was

Morningstar's 24% month is one gap: the sessions spanning its 28 July results moved the stock 16.4%, and stripping that plus its next-best day leaves about 2%. FactSet's is not — it round-tripped a July spike, then built 14.6% through August, and even without its two best August sessions retains roughly 4% of drift. On 19 August, when FactSet gained 5.7% with no company news beyond a routine $1.16 dividend, thirty-year Treasury yields fell 10 basis points to 5.18% and the S&P 500 rose 0.4%. That is multiple repair on a long-duration subscription stream, not a fresh disclosure.

Underneath all three, the industry pool is still growing: global spending on financial market data rose 6.5% to a record $49.2bn in 2025, with AI integration cited as a driver of demand rather than a substitute for it.

The setup

Where it stands — FactSet's subscription growth is accelerating while its margin compresses; MSCI's numbers are the group's best and its shares the worst. Would confirm — FactSet's fiscal fourth-quarter organic subscription value growth prints at or above 7.1% in September. Would invalidate — Adjusted operating margin lands below the reaffirmed 34-35.5% range, or client and user counts turn negative. Watch next — FactSet reports fiscal fourth-quarter and full-year results in late September; MSCI reports third-quarter results in October. Valuation — FactSet 19.5x trailing and 16.9x forward, against 21.0x on the same estimate a year ago and 11.9x in May.