DK Street Journal

Kinder Morgan moves roughly 40% of the natural gas burned in the United States through some 83,000 miles of pipe, which makes it the default counterparty when a utility, a liquefied natural gas (LNG) terminal or a new gas-fired power plant needs an interconnect. In the second quarter it grew adjusted earnings before interest, taxes, depreciation and amortization (EBITDA) 12% year over year, lifted adjusted earnings per share 32%, and told investors the full year would come in at least 5% above internal budget — about $430m of incremental EBITDA. It raised the dividend 2%. Then a trend-following screen cut it to a bearish reading on 13 August, while the shares were rising.

The rates story failed a live test

The explanation on offer for a soft month in pipelines was the long end of the bond market. The 30-year Treasury yield topped 5.33% on 18 August, a 19-year high, which Bloomberg attributed to AI infrastructure spending and federal deficits — the same capex wave that fills pipeline backlogs, lifting the rate at which their cash flows are discounted.

It does not survive contact with the prices. Across the drawdown window the 30-year moved about 6 basis points, from 5.27% at the end of July. The yield ordering runs backwards: Energy Transfer, at roughly a 6.5% distribution yield, rose 4.4% over 30 days, while DT Midstream, yielding 2.7%, fell 7.9% and TC Energy fell 8.9%. Kinder Morgan (-1.5%) and Williams (-1.2%) sat closer to the regulated utilities Southern Co (-2.4%) and Duke Energy (-1.8%). Rates explain the shallow, uniform part of the move and none of the rest.

The cheapest name is the one compounding

Kinder Morgan's sanctioned project backlog was $9.6bn at the end of the second quarter, down only because completed projects entered service; gas is about 92% of it and more than 60% serves power generation and local gas utilities — a combined figure, with no separate data-center line disclosed. Federal regulators granted certificates on 31 July for Mississippi Crossing, 208 miles carrying up to 2.1 billion cubic feet a day into the Southeast: the group's largest permitting risk cleared while the shares fell. Transport volumes rose 7% and gathering 26%. Leverage fell to 3.6 times net debt to EBITDA.

Its trailing price-to-earnings ratio is 20.5x, against 21.7x in mid-May on a market capitalization essentially unchanged at $71.1bn. The multiple came down because earnings went up. Free cash flow yield is 5.4%.

DT Midstream earned its de-rating

DT Midstream is a $13.5bn pure-play gas midstream company with 588 employees, gathering Appalachian and Haynesville gas and moving it on the NEXUS and LEAP systems. Its revenue growth halved — 27% year over year in the fourth quarter of 2025, 11% in the second quarter of 2026. Adjusted EBITDA of $305m was $3m below the prior quarter, guidance was reaffirmed rather than raised, and the third quarter was guided lower. Consensus has 2027 revenue growing 4.6%.

The data-center content is real but small: more than 500 million cubic feet a day of demand pull added to NEXUS across two interconnects, roughly a third of that one pipeline and closer to a tenth of total system throughput. At 15.3x EV/EBITDA and 27.9x forward earnings it remains more expensive than Kinder Morgan on both. JPMorgan cut it to Neutral on valuation, with target cuts from Mizuho and Jefferies. Its uptrend broke on 18 August, the first non-bullish reading in some fifteen months — an analyst call and a growth curve, not a bond auction.

Williams is the contradiction

Williams owns Transco, the interstate line that feeds the eastern seaboard. Second-quarter operating margin reached 38.7%, from 32.0%; full-year adjusted EBITDA guidance went to $8.3-8.5bn and the 2025-30 growth target to 11%-plus, from 9% in May. Socrates Phase 1, a 200-megawatt plant sited behind the meter at a Meta campus in Ohio, was built in 18 months — against grid interconnection queues measured in years. Blackstone Credit & Insurance put $5.34bn into a power joint venture covering five projects, the largest 682 megawatts. The shares fell 7.8% over three months and the trailing multiple compressed from 34x to 29x on a flat market value.

The $5.5bn Momentum Midstream purchase adds more than 4,000 miles of pipe and roughly 4 billion cubic feet a day of take-or-pay capacity, making Williams the largest Haynesville gatherer tied to Transco — squarely into DT Midstream's LEAP corridor. Energy Transfer, the month's only riser, contracted about 0.9 billion cubic feet a day to three Oracle data centers. Scarcity is the shared mechanism: one large interstate line, Mountain Valley, has been completed in a decade, and new corridors take five to eight years, so only steel already in the ground serves the late-decade load.

The setup

Where it stands — Kinder Morgan raised guidance and trades at the group's lowest multiple; DT Midstream's growth halved and its premium has only partly gone. Would confirm — Kinder Morgan adding the promised $1bn-plus of sanctioned projects in the second half, and DT Midstream's third-quarter EBITDA printing below $305m. Would invalidate — DT Midstream raising 2026 guidance on new data-center contracts, or Kinder Morgan's backlog shrinking without offsetting in-service additions. Watch next — Third-quarter results: Kinder Morgan in mid-October, DT Midstream and Williams in late October and early November. Valuation — Kinder Morgan 12.8x trailing EV/EBITDA and 21.0x forward earnings; DT Midstream 15.3x and 27.9x; Williams 16.1x and 30.1x.

One deal ended the comparison

On 18 May, NextEra Energy — owner of Florida Power & Light and the largest US developer of contracted wind, solar and storage — agreed to buy Dominion Energy, the regulated utility whose Virginia territory contains the densest cluster of data centers on earth. The consideration is stock: 0.8138 NextEra shares per Dominion share, plus a one-time cash payment of $360m, creating a business with roughly $420bn of enterprise value. Dominion holders would own about a quarter of the result.

That disposes of any question about which utility the market prefers. Dominion has been in an uptrend for 114 straight sessions since 28 April, the only large utility in that condition on 19 August, when nine peers including Duke, Exelon, Sempra and Xcel were all in confirmed downtrends. But the streak decomposes to a single day. Dominion closed at $62.89 when the run began and $68.04 on Thursday, an advance of 8.2%; the announcement session alone was 9.44%. Everything since has been the arbitrage discount grinding shut — from −6.8% against the exchange ratio on 18 May to −1.9% now.

The regulated side is winning the load, slowly

Dominion's operating business is doing what the thesis says it should. It reported more than 53 GW of data-center capacity in stages of contracting in the second quarter, with 12 GW under signed electric service agreements, up roughly 5 GW since year-end. Revenue rose 19.6% to $4.56bn.

The money has not arrived yet. Operating income fell 2.4%, margin narrowing to 23.5% from 28.8%, and net income fell 55% to $340m — construction of the Coastal Virginia Offshore Wind project, whose cost estimate rose nearly $300m, sits between the load growth and the shareholder. Consensus has Dominion earning $3.59 a share this year and $4.08 by 2028, mid-single-digit growth against double-digit revenue. At 18.97x forward earnings it is priced where a regulated utility with 6-8% rate-base growth normally is.

NextEra is the one being marked down, and not for operating reasons. Second-quarter revenue rose 12.4% to $7.53bn and net income rose 55% to $3.14bn. The Energy Resources backlog reached 35.1 GW after 3.6 GW of additions, and management raised Florida Power & Light's expected large-load connections to 8 GW by 2032 from 6 GW — each gigawatt roughly $2bn of capital earning a regulated return. Guidance for 8%-plus earnings growth through 2035 was reaffirmed in July. The shares fell anyway, and its forward multiple of 21.18x sits above its trailing 19.10x, because consensus adjusted earnings of $4.03 are below trailing reported earnings. The de-rating is compressing a premium, not creating a discount — and it still leaves NextEra dearer than the company it is buying. The gating item is Virginia: the merger applications went to the state corporation commission and four other regulators on 15 July, with closing targeted for the second half of 2027.

The merchant upside is being capped by rule

The purer bet on AI electricity has been the worse one, and the reason is administrative. PJM's capacity auction has now cleared at the federally approved ceiling three delivery years running, most recently at $325/MW-day for 2028/29, procuring 138,318 MW. PJM's independent market monitor calculates that the cap kept $13bn off customer bills and that prices would otherwise have exceeded $500/MW-day. The scarcity is real; the rent from it is being handed to load.

Vistra, an integrated generator and retailer with about 38,700 MW across gas, nuclear, coal and batteries, cleared roughly 10.9 GW in that auction — visibility, at the ceiling price. Its adjusted EBITDA rose 30% to $1.77bn and 2027 guidance was held at $7.4-7.8bn, yet the shares are down 27% over twelve months. It fell 8.15% on 4 August, the session after Texas ordered an audit of data-center interconnections, which management expects to resolve in about two months. Vistra trades at 10.27x trailing enterprise value to EBITDA against 14.74x for Constellation Energy, the nuclear-heavy peer that raised full-year guidance and has risen over the past month while Vistra fell. That gap is what the market pays for reactors without a gas fleet attached.

Rates are the fashionable explanation — the 30-year Treasury touched 5.33% on 18 August, a 19-year high. But a single yield cannot produce one utility rising, nine falling and one merchant generator down a quarter on the year. Ownership of the load is doing the work.

The setup

Where it stands — Dominion trades as a fraction of NextEra; the merchant generators own the demand growth but not its economic rent. Would confirm — Vistra's 2027 adjusted EBITDA guidance holds at $7.4-7.8bn when ERCOT forward curves are next marked. Would invalidate — Dominion's discount to the 0.8138 exchange ratio widening back beyond 5%, signaling doubt the deal closes. Watch next — NextEra and Dominion shareholder votes in early September 2026, then the Virginia commission's merger record. Valuation — NextEra 21.18x forward against 19.10x trailing; Dominion 18.97x forward; Vistra 15.79x forward against 23.37x trailing.

The companies that supply the eyes, hands and nerves of automated machinery reported second-quarter results within a week of each other in early August. Almost all of them named the same new customer. It was not a robot maker.

The shipped-units test

Vishay Precision Group, which makes strain gauges, load cells and force transducers for steel mills, farm equipment and test rigs, has been the loudest of the group about humanoid robots — it announced a production nomination from its first humanoid customer and a possible ramp to thousands of robots a week by late 2026. In the quarter just reported, humanoid bookings were $500,000 and related sales $320,000, against $83.9m of total revenue. That is four-tenths of one percent.

What did grow at Vishay Precision was its Sensors segment, up 26% to $33.4m on record bookings of $48.1m — a book-to-bill of 1.44 — driven by AI infrastructure, semiconductor equipment makers, data-center fiber optics and defense. Weighing Solutions, the industrial and transportation half, grew 3% with bookings below shipments. The shares fell 24.5% on 5 August and another 10.7% the next session around the report, and are down 58% from a 30 June peak.

The same pivot, three times

Cognex, which sells machine-vision systems and barcode readers that inspect and locate parts on production and logistics lines, and which with Japan's Keyence holds close to half the global machine-vision market, posted record revenue of $291.3m, up 16.9%. Gross margin was 70.6% and operating income nearly doubled. Management raised semiconductors, electronics and packaging to double-digit growth on AI infrastructure spending, held automotive flat on European weakness, and described its data-center supply-chain business — server and rack assembly inspection — as growing more than 30% from a low-single-digit share of revenue. It called that business nascent. It is nonetheless the accelerant.

Novanta, a photonics and precision-motion supplier to medical and industrial equipment makers, grew 9% organically with gross margin up 100 basis points to 47%. Its generative-AI-linked applications — lithography, GPU drilling, advanced packaging — were about 17% of company revenue and grew roughly 25%. Novanta did book its first material servo-drive orders for humanoid training centers, a step past prototyping, but the segment carrying it is semiconductor capital equipment. Reported operating income fell 34.3% on deal costs from the $1.2bn Riverpoint Medical purchase, struck at about 19x estimated 2026 earnings before interest, tax, depreciation and amortization and funded with roughly $800m of gross debt, lifting pro forma leverage to 2.7x.

Sensata, an automotive and heavy-vehicle sensor maker with 16,700 employees, is the industrial-scale member. Its automotive revenue grew 1.8% organically to $545m against flat global production — roughly two points of content-per-vehicle outgrowth, achieved while S&P Global Mobility cut its 2026 forecast to 89.4m light vehicles from 91.9m. Aerospace and defense grew 10.9%. And Sensata has won five hyperscaler platform concepts this year, including a coolant-distribution-unit sensor ramping in early 2027; its industrial components revenue roughly doubled in the first half.

Where the prices disagree

This is not one trade. Allient, a motion-and-controls maker, gapped 18% higher on 6 August after orders rose 49% to a 1.31 book-to-bill. Badger Meter — a water-metering business, misplaced in any sensor grouping — lost 7% of revenue and fell 13.2% in one session on 22 July. Strip those two and Vishay Precision out and the rest are roughly flat over a month.

Most of what remains is one macro session: on 18 August the 30-year Treasury yield hit a 19-year high near 5.3% and the Philadelphia Semiconductor index fell 5.4%. Novanta lost 12.5% over three sessions, Cognex 10.6%, Sensata 8.5%, with no company news.

The de-rating story does not survive the arithmetic. On consensus 2026 earnings of $1.68, Cognex's February price implied about 33x; today it is about 35x. Novanta sits at 39.0x forward, flat to February, and 33.6x enterprise value to EBITDA for roughly 7% organic growth. Sensata is the exception: 11.2x forward earnings, 11.1x EV/EBITDA, a 9.3% free-cash-flow yield, and a fourth straight organic-growth quarter with free cash flow up 61% to $186m and leverage cut to 2.4x from 3.0x — down 6.1% on 18 August alone and 11.7% over three months.

The setup

Where it stands — Data-center and semiconductor capex, not robotics, is the growth engine in industrial sensing; only Sensata's price has fallen while its numbers improved. Would confirm — Cognex third-quarter revenue lands in its $300-320m guide with semis and electronics again double-digit. Would invalidate — Humanoid-related bookings at Vishay Precision or Novanta exceed 3% of revenue, making robotics a real line item. Watch next — Third-quarter reports in late October; Sensata guided to $957-987m, sequentially flat to down. Valuation — Cognex 56.7x trailing and 35.4x forward against ~33x implied in February; Sensata 11.2x forward, 11.1x EV/EBITDA.

Genpact, the business-process outsourcer spun out of General Electric's back office that runs accounts payable, invoice-to-cash, procurement and risk-and-compliance operations for banks, insurers and drugmakers, told investors on 6 August that quarterly revenue rose 7.1% to $1.34bn. That was a fourth consecutive quarter of accelerating growth. Gross margin widened 62 basis points to 36.5%, and gross profit grew faster than revenue. Management raised its full-year outlook for the second time this year, lifting expected growth in its Advanced Technology Solutions unit to at least 25% and adjusted earnings growth to at least 12%.

The shares fell 5.2% the next session, to $34.29, and drifted to $33.68 by 13 August before recovering.

The line that did the damage

Advanced Technology Solutions — data, engineering and AI-led programs — produced $363m in the quarter, up 24.1%, and is now 27% of the company. Core Business Services, the seat-based legacy processing that is the rest of it, grew 1.9%, and management guided it flat to slightly down for the current quarter. That is the whole argument in two numbers: when a client's invoice queue is automated, the seats that used to be billed for it disappear immediately, and the replacement AI mandate is smaller and lumpier. Genpact is the purest labor-arbitrage business of the three, and it is the first place the substitution shows up in reported revenue.

The other two moved on nothing they said

The group's year has been defined by a single session. On 18 June, Accenture — the Dublin-based professional-services firm that sells strategy, systems integration and managed operations to large enterprises and governments — reported new bookings down 2% to $19.3bn and cut full-year local-currency revenue guidance to 3-4%. The stock fell 18.0% that day, its worst on record. Cognizant, the Teaneck-based offshore-heavy delivery incumbent, fell 10.5% in sympathy; Genpact fell 6.8%. None of the three had a lower close afterwards than on 30 June.

The recovery since has been uneven in a way that matters. Accenture's 30.7% gain over the past month is spread across many sessions; strip its two best and roughly 15% remains. Cognizant's 39.6% falls to about 8% on the same treatment, and Genpact's 22.2% falls to approximately nothing — its entire month is two days in late July, when a $3.3trn selloff in semiconductor shares pushed money out of expensive AI hardware and into cheap labor-heavy value. Accenture has disclosed nothing new since June; its fiscal fourth quarter is not due until late September.

Cognizant is the one whose own earnings day was an up day. On 29 July it raised full-year adjusted earnings guidance to $5.70-$5.82, reported a sixth straight quarter of adjusted operating-margin expansion and trailing-twelve-month bookings of $29bn. But revenue growth decelerated to 4.5%, gross margin slipped to 33.4%, and the operating leverage came from the Project LEAP cost program — $84m of charges in the quarter, mostly severance. The pyramid is being reshaped by layoffs, not by demand.

The offshore names refuse to confirm an industry turn. Infosys rose 3.8% over the same month and Tata Consultancy Services 1.9%, after Infosys cut its full-year growth guidance to 1.5-3% in constant currency on pricing pressure.

What the prices still imply

Even after the bounce, none of the three has recovered its own multiple. Genpact trades at 10.8x trailing and 8.9x forward earnings, against roughly 14.7x a year ago — a de-rating delivered entirely through the multiple while earnings rose. Accenture is at 14.6x trailing and 13.3x forward, versus about 21x a year ago, and remains the premium name on price-to-gross-profit at 4.79x forward against Genpact's 3.14x. Cognizant sits at 13.1x and 10.6x.

The competitive question underneath is who sells the automation. Microsoft committed $2.5bn and 6,000 employees to Frontier Co. on 2 July, embedding its own engineers inside customers — the same implementation work Accenture and Cognizant bill for, sold by the vendor whose software they deploy. Accenture, meanwhile, stopped breaking out advanced-AI bookings after its fiscal first quarter, when the figure was $2.2bn. The clearest disclosure of AI's effect on this industry now comes from its smallest member, and it is subtraction.

The setup

Where it stands — Genpact has the best growth and margins of the three and the lowest multiple, because its legacy book is shrinking as its AI book ramps. Would confirm — Core Business Services returning to growth in the third quarter after guidance of flat to slightly down. Would invalidate — Advanced Technology Solutions growth slipping below the raised 25% full-year floor while core stays negative. Watch next — Accenture's fiscal fourth-quarter results in late September, the first bookings figure since the 18 June cut. Valuation — Genpact at 10.8x trailing and 8.9x forward earnings, against roughly 14.7x trailing a year ago.

A beat that read as a warning

Dave Inc., a consumer app that fronts members cash against their next paycheck for a flat fee instead of an interest charge, told investors on 5 August that its 28-day past-due rate had improved to 2.12%, fourteen basis points better than a year earlier. Volume on the ExtraCash advance grew 27% to $2.3bn over the quarter, with the average advance at a record $215. Monthly transacting members reached 3.08m, up 17%, and the company lifted full-year revenue guidance to $725-735m, adjusted cash profit to $315-325m and adjusted earnings to $17.00-17.50 a share. Over the following two sessions the shares lost 26% — more than the entire month's decline.

One number explains it. June-quarter revenue of $170.8m was up 29.6% year over year, after 46.7%, 62.4% and 63.0% in the three quarters before it. Growth halved in a year, and the reaction was read at the time as a valuation reset rather than a credit one. Dave's pricing power is thin — rivals including EarnIn, Chime's MyPay and MoneyLion's Instacash charge no subscription and no mandatory advance fee, while Dave takes up to $5 a month plus roughly 5% per advance — so what it sells is underwriting and limit size. Its CashAI V6 model, live with about a third of users, is meant to push maximum advances well above $500. The regulatory overhang has eased: the Consumer Financial Protection Bureau (CFPB) withdrew its proposal to treat paycheck advances as credit under the Truth in Lending Act, replacing it with an advisory opinion in December. A 2024 Federal Trade Commission suit over Dave's "tip" disclosures and separate Justice Department litigation remain open.

Deceleration, not defaults

The pattern repeats across consumer fintech. Sezzle, a buy-now-pay-later provider, fell 33.9% on 7 August after beating on revenue and raising full-year guidance to 35% growth — punished solely for guiding second-half growth down toward 30%. Klarna's 22% single-session drop on 18 August came from a revenue guidance cut driven by about $600m of currency headwinds and softer German demand, with credit provisions running at 0.52% of goods volume. The control is the two large US consumer lenders with no results in the window: Affirm rose 2.1% and Upstart 4.0% between 3 and 19 August. Nor was it rates — the 30-year Treasury yield touched 5.33%, a 19-year high, on 18 August, and Dave rose 3.1% across that session and the next.

OppFi is the exception that proves the rule. The Chicago lender enables partner banks to write subprime installment loans and is a credit business, not a software one. Revenue of $145m grew 1.9%, originations fell 9% on deliberate tightening, adjusted net income dropped 27%, and net charge-offs reached 52% of receivables from 43% — though management attributes about 500 basis points of that to shrinking the loan book itself, and calls the consumer environment stable. Its guidance cut was pinned on a line-of-credit launch slipping to September; 2027 and 2028 targets, including roughly $3 of earnings by end-2028, were left alone. Shares fell about 25% in one session anyway. OppFi has applied to the Office of the Comptroller of the Currency to buy BNC National Bank, which would remove its dependence on wholesale funding, and started a $40m buyback.

The software name went the other way

Bill Holdings, which runs the largest bill-payment network for small businesses in the US and reaches most customers through their accountants, processed $98bn of payments in the June quarter, up 14%, while transaction fees grew 17%. That gap widened the take rate to roughly 33.1 basis points from 32.3 — the direct rebuttal to the long-standing worry that competition from Ramp, Brex and Melio would compress B2B payment fees. Float income on customer money in transit, at $35.7m, was about 8% of revenue and slightly lower than a year ago: the software business is carrying the growth, not the interest-rate position. The catch is fiscal 2027 core revenue guidance of 11-14%, below the 16% just delivered. Shares rose 8.4% over the month.

What the prices leave is uneven. Dave at 8.56 times trailing gross profit, down from about 11 times at its July high, is still above the 6.7 times it fetched in May and roughly 7.3 times a year ago — the drawdown has not returned it to its own anchors. Bill trades at 3.57 times trailing and 3.29 times forward gross profit, against 3.04 times in May and 3.58 times a year ago, with 13.4 times forward earnings and an 8.9% free-cash-flow yield; its trailing price-to-earnings ratio of -384 is noise on a small accounting loss. OppFi sits at 1.17 times book and 5.1 times forward earnings.

The setup

Where it stands — Three fintechs fell on slowing growth and one guidance delay, not on deteriorating loan books.

Would confirm — Dave's third-quarter loss rates hold near the second quarter's 2.12% past-due level with revenue growth still near 30%.

Would invalidate — Dave's or OppFi's delinquencies rising while origination volumes also grow, indicating looser underwriting.

Watch next — OppFi's line-of-credit launch in September and its OCC decision on BNC National Bank, targeted to close in the fourth quarter.

Valuation — Dave 8.56x trailing gross profit versus 6.7x in May; Bill 3.57x trailing and 3.29x forward against 3.58x a year ago.

MKS Inc. supplies the layer beneath the fab tools everyone quotes: vacuum and pressure subsystems, gas and flow controls, radio-frequency power delivery, plasma sources and lasers, sold to chipmakers and to the companies that build their equipment. It also owns, since a debt-funded 2022 purchase of Germany's Atotech, an electroplating-chemistry business that plates copper onto printed circuit boards and advanced packaging substrates. That second half has been treated for four years as the thing to look past.

In the June quarter it was the faster half. Group revenue reached $1.248bn, up 28% year on year and accelerating from 15% in the March quarter. The Semiconductor segment contributed $554m and grew 28%. Electronics & Packaging — the Atotech-derived arm — contributed $381m and grew 44%. Specialty Industrial added $313m. So semiconductors are 44% of MKS, not most of it, and the chemistry is not the drag. Management said artificial-intelligence-related chemistry has gone from roughly 5% of chemistry revenue in 2024 to 15-20% now, with lead times of six to nine months and visibility into 2027.

The leverage question, answered in cash

The standing objection to MKS has been the balance sheet, not the business. That objection is smaller than it was. Net leverage is 3.0x, down from 4.0x a year earlier. Free cash flow was $188m in the quarter, about 15% of revenue, and the company made a $100m voluntary prepayment on its term loans with more planned for the second half.

One wrinkle is worth knowing. Because the share price cleared the conversion trigger, MKS reclassified $1.4bn of convertible senior notes into short-term debt, which are convertible at holders' option this quarter. Cash stood at $611m. It is an accounting reclassification of an equity-linked instrument, not a funding cliff — but it makes the short-term debt line look alarming to anyone reading only the balance sheet.

Operating margin reached 20.1% against 13.9% a year earlier, and operating income grew 86% on 28% revenue growth. The September quarter is guided to $1.35bn, with the semiconductor segment around $630m — more than 50% growth.

What the shares did

None of that stopped two hard falls. Between 22 and 29 July, MKS lost 25% as the Philadelphia Semiconductor Index entered bear-market territory and more than $1trn of chip value was erased on doubts about hyperscaler returns; Applied Materials fell 21% in the same window. Then 17-19 August took another 13%, as the 30-year Treasury yield hit a 19-year high and the chip index dropped 4.96% in a session. Both were index-wide. Neither was about MKS.

The result is that MKS trades at 21.4x forward earnings against 42.9x trailing, and at 9.89x trailing gross profit — below the 10.77x it carried in early May, when growth was half its current rate. The demand pool behind it was raised, not cut: Lam Research lifted its 2026 wafer-fab-equipment forecast to the low $150bn range from $140bn, and Applied Materials says leading-edge foundry, memory and advanced packaging drive more than 80% of this year's spending growth.

Two neighbors, two different answers

FormFactor makes probe cards — the consumable interfaces that test chips while still on the wafer, bought per wafer rather than per tool. Its gross margin hit 50.7% against 37.2% a year earlier, gross profit grew 80%, and its memory probe business set a record with high-bandwidth memory roughly two-thirds of it, on two customers adopting its full-wafer contactor for HBM4 testing. At 21.9x trailing gross profit it is priced at less than two-thirds of its early-May multiple. The caveat is the guide: September memory revenue is flat, with mix shifting toward lower-content standard DDR chips.

Ichor is the different case. It builds gas and fluid-delivery panels to its customers' designs, at low-teens gross margins, with Lam and Applied together 76% of sales. The business improved — revenue up 23%, gross margin recovered to 13.9% from a 4.6% trough, and full-year growth guided to at least 30%. But even after a 40% fall from July, it trades at 20.5x trailing gross profit, slightly above its early-May level, because gross profit grew faster than the price fell. On 38.6x forward earnings for build-to-print work, its decline has removed less than it appears.

The setup

Where it stands — MKS's chemistry arm is outgrowing its semiconductor arm while the forward multiple sits below May's, and leverage is falling.

Would confirm — September-quarter revenue near the $1.35bn guide with Electronics & Packaging growth above 30%.

Would invalidate — Semiconductor revenue below the roughly $630m guided, or net leverage rising back above 3.5x.

Watch next — MKS reports the September quarter in early November; FormFactor's memory mix shift shows up first, in late October.

Valuation — MKS at 21.4x forward and 42.9x trailing earnings; 9.89x trailing gross profit against 10.77x in early May.

When a company runs payroll through Automatic Data Processing, the cash leaves the employer's account days before it lands in employees' accounts. ADP holds it in between, and invests it. For most of the past two decades that float was a rounding line in a story about employment growth. In fiscal 2026 it became the story.

ADP reported its year on 29 July. Employer Services revenue, the payroll and benefits business, grew 7%, on new bookings of $2.2bn and record client retention of 92.1% — it now loses fewer than eight clients in a hundred each year. But the volume metric that ties revenue to hiring barely moved. US pays per control, the count of employees on client payrolls ADP bills for, grew 1% in the fourth quarter and for the full year, and the company guided fiscal 2027 to 0-1%. The professional employer arm, which co-employs staff at smaller firms, ended the year with 775,000 average worksite employees, up 2%, and its margin contracted 110 basis points.

The float is the margin

Interest on funds held for clients rose 14% to about $1.35bn, on average balances of $40.4bn and a yield up to 3.4% from 3.2%. Management guided that line to $1.54-1.56bn for fiscal 2027, on a yield of 3.7%. Set the roughly $195m increment against the $23.1bn revenue base implied by the 5-6% growth guide and it is worth about 84 basis points of margin — against the 70-90 basis points of adjusted operating-margin expansion the company guided. Float income carries almost no cost to serve, which is why $1.35bn equalled 6.2% of last year's $21.95bn of revenue but roughly 31% of its $4.41bn of net income.

The rate backdrop is doing the work. The 30-year Treasury yield touched 5.323% on 18 August, its highest since 2007. ADP shares rose 1.1% that session and 2.9% the next. The employment backdrop is doing the opposite: ADP's own National Employment Report showed private payrolls adding just 44,000 jobs in July, the weakest since January.

Paycom cut its way to the same place

Paycom, which sells a single-database payroll and human-capital platform to small and mid-sized US employers, reported second-quarter revenue up 9.8% to $531.2m. Operating margin went from 23.2% to 31.7% — sourced not from volume but from more than $100m of annual research savings after moving workloads into its own data centers, plus a $30m-plus cut in third-party AI token fees. Diluted shares fell 18.5% year over year, which is how 20% net-income growth became 48% earnings-per-share growth. On the call, management described client employment as stable with no acceleration. Full-year revenue guidance of 7-8% compares with 23.2% growth in 2023, and consensus models 7.1% for 2027. Paycom competes against ADP Workforce Now, Dayforce and UKG in a market where scale sits with ADP's million-plus employers; its answer is product velocity — a 45th product shipped in July — rather than price.

Workday's month has no numbers in it

Workday, the large-enterprise human-resources and financials suite, has disclosed nothing operational since 21 May, when it reported subscription revenue up 14.3% and a 12-month subscription backlog of $8.806bn, up 15.5% — committed spend running slightly ahead of the 12-13% full-year guide. Its shares are up about 35% in a month anyway. Two sessions explain it: the late-July software rally after Nvidia's Jensen Huang said of AI disruption fears that "the markets got it wrong", and 14 August, when Reuters reported that Silver Lake is in talks to take Workday private at up to $43bn. The stock rose 12.5% on 8.2m shares, roughly six times its recent norm. Strip those sessions and the month is slightly negative. At a $51.97bn market value, the shares already sit about 17% above the reported deal price.

What the multiples say

Gross margins differ too much for revenue multiples to compare — 83.2% at Paycom, 76.2% at Workday, 53.0% at ADP in the latest quarter — so price against forward gross profit is the usable lens. Workday trades at 6.43x, against roughly 4.19x three months ago and 7.37x a year ago. Paycom is at 5.66x versus about 3.08x six months ago and 5.78x a year ago: the de-rating has essentially been undone. ADP, at 9.91x forward gross profit and 22.6x forward earnings, is the only one of the three still below its year-ago level of about 11.04x — and the only one whose recent results contain new operating numbers.

The setup

Where it stands — Payroll volumes across all three are flat; the profit growth on offer comes from interest rates, cost cuts and buybacks. Would confirm — ADP fiscal 2027 float income tracking to the guided $1.54-1.56bn while pays per control stays at 0-1%. Would invalidate — Pays-per-control growth re-accelerating above 1%, or ADP raising margin guidance without help from the yield. Watch next — Workday reports fiscal second-quarter results on 27 August, its first disclosure since 21 May. Valuation — ADP 9.91x forward gross profit versus 10.48x trailing and 11.04x a year ago; Workday 6.43x versus 4.19x in May.

Commvault, which sells software that backs up corporate data and rebuilds it in a clean environment after a ransomware attack, beat both lines of its fiscal first quarter on 28 July and raised guidance for subscription revenue and margins. Its shares fell 19.3% over the following two sessions, from $149.46 to $120.66, and by 19 August had recovered only to $137.32. Adjusted earnings came in at $1.42 against $1.16 expected, with Piper Sandler downgrading the stock for want of a catalyst.

That is the sharpest illustration of a problem running through this whole rung of software. Seven vendors — data recovery, fraud approval, identity checks, regulated filings — rose an average of roughly 10.5% over the past month. They are not one demand curve, and the month was not a grind: Workiva was up 32.4% and Freshworks 18.3%, while Commvault fell 6.4%.

The bid is doing the work at Commvault

The reported numbers are decelerating. Revenue growth has gone 19.5%, then 13.3%, then 11.4%, and first-quarter operating income rose just 2.5% on that 11.4% — negative operating leverage. Underneath, the conversion from perpetual licences to subscription is genuinely hiding demand: subscription annual recurring revenue (ARR) rose 22% to $1.05bn, software-as-a-service (SaaS) ARR rose 38% to $424m, and SaaS gross margin crossed 70% for the first time, up 635 basis points to 70.6%. Nearly a third of net new subscription ARR came from identity-resilience and data-security products rather than backup.

But net dollar retention was 114%, flat year on year — a rate at which existing customers are not doing the heavy lifting. And management reiterated, rather than raised, its full-year ARR target of $1.20-1.21bn. That was the line that broke.

The competitive backdrop explains the reluctance. Rubrik lifted its share of data protection from 3.9% to 5.5% and Veeam from 9.2% to 12.9%, leaving Commvault grouped with Cohesity behind them. Rubrik grows 39% and trades at 18.3x trailing gross profit; Commvault grows 11% and trades at 5.79x, against 9.2x a year ago. What lifted the stock was not the P&L. It gapped 10.2% on 10 April on nine times normal volume, when it emerged that Commvault was weighing a sale after approaches from private equity, with Goldman Sachs advising and Thoma Bravo reported to have renewed interest.

Riskified is buying its acceleration

Riskified is not a subscription business. It decides whether to approve or decline an online checkout and assumes the fraud liability itself, taking a fee on the merchant volume it clears. Second-quarter revenue accelerated to 21.8% growth, its fastest in over four years, and guidance was raised for the second time this year to $400-410m. Gross merchandise volume grew only 13%, so the take rate rose.

The cost of underwriting rose faster. Cost of revenue climbed 31.4% to $54.3m, gross margin fell 490 basis points to 45.0%, and gross profit grew 11.7%. Management blames a heavier mix of lower-margin ticketing and new merchants ramping, and calls it temporary — but its own third-quarter guidance pairs about 27% revenue growth with 11-14% gross-profit growth. Retention is about 105%; the acceleration is new logos, not existing merchants. Against that, $223.6m of cash and no debt is 24% of the market value. Competition is thickening: Stripe Radar added a standalone multi-processor product in 2026, alongside Signifyd, Forter and Accertify.

Workiva is the one the numbers carry

Workiva sells the software that assembles regulated filings and controls documentation, with audit trails, for roughly 6,200 corporate clients. The fear priced into it was that agentic AI automates exactly this work first. The quarter says otherwise: revenue up 18.6% to $255.3m, gross margin up 343 basis points to 80.4%, and non-GAAP operating margin of 16.8%, up 1,300 basis points — its 2027 target reached a year early. Net revenue retention was 111%, contracts above $300,000 grew 34%, and premium AI tiers carry a price premium north of 20%. Its largest session of the month was 29 July, when it launched three compliance agents and an intelligence layer. At 5.32x trailing gross profit, up from 3.68x three months ago, it is still cheaper than the 6.99x it commanded a year ago.

The remaining four are a reminder that the grouping is loose. Alarm.com, a cloud platform for home and commercial security systems, grew 9.2% and is the cheapest name at 4.22x gross profit. Freshworks sells helpdesk and IT service management software, not security, and its 16% growth rests on an employee-experience line at $567m ARR. Mitek, in identity and check-fraud verification, moved 1.0% with no earnings in the window. Red Violet, an identity-data business, posted record revenue of $26.7m and trades at 11.38x gross profit, the group's richest. The 19 August session that lifted Workiva 8.0% was a rotation out of semiconductors into beaten-down enterprise software — the same day Commvault fell 3.8%.

The setup

Where it stands — One month of shared gains covers three unrelated engines: a takeover bid, a fraud-loss squeeze, and a genuine margin inflection.

Would confirm — Commvault raising its $1.20-1.21bn ARR target, or Riskified's gross profit growth closing on its revenue growth.

Would invalidate — Commvault's sale process ending without a deal, or Riskified's gross margin falling below 45% again in the third quarter.

Watch next — Commvault's fiscal second quarter in late October; Riskified's third quarter, guided to 27% revenue and 11-14% gross-profit growth.

Valuation — Commvault 5.79x trailing gross profit and 5.39x forward, versus 9.2x a year ago and Rubrik's 18.3x.

On 6 August, Datadog — which sells cloud monitoring software billed by the host watched, the gigabyte of logs ingested and the trace stored — told investors its largest customer would reduce usage starting in the third quarter. The account is a nine-figure artificial-intelligence company running 17 Datadog products. Chief executive Olivier Pomel declined to name it; Wall Street analysts widely believe it is OpenAI. Datadog had just beaten and raised. The shares fell 20.4% that day.

That is the arithmetic of consumption billing. Revenue is a direct read on telemetry actually emitted, so a single customer's cost-engineering decision arrives in the quarter it is taken. Datadog's quarter was otherwise its best in years: revenue of $1.121bn, up 35.6%, a fourth consecutive acceleration. Customers spending over $100,000 a year reached 4,720, up 22.6%, and now supply 91% of annual recurring revenue. Net revenue retention sits in the low 120s. Management still guided the September quarter to 28–29% growth. Bernstein's Peter Weed cut the stock to Market Perform with a $226 objective, arguing the valuation required near-flawless execution.

The opposite contract