DK Street Journal

The security industry just staged a rally in which the tightest link between business results and share price was an inverse one. Fortinet, which sells firewall appliances with attached security subscriptions, delivered on July 29 what was arguably the group's best quarter — billings up 33% to $2.37bn, revenue growth accelerating to 25.6% from 14.8% two quarters earlier, free cash flow tripling to $966m — and the stock finished the month down 2.7%. Meanwhile Rubrik and Zscaler, which have reported nothing new since early June and May 26 respectively, gained 27% and 24%.

The explanation for Fortinet is arithmetic, not disappointment: the shares had already doubled over twelve months and trade at 46.8x forward earnings, so a record quarter bought digestion rather than applause. The explanation for the rest of the group is a narrative. The Black Hat conference in Las Vegas pushed CrowdStrike and Palo Alto Networks to record highs on August 10 on the argument that AI is both a new attack surface and a new budget line, with industry forecasts putting 2026 security spending above $520bn, double 2021's level.

Who has the numbers

Only two of the seven can point to reported acceleration. Cloudflare, which runs a global edge network billed by usage rather than seats, grew Q2 revenue 35.9% to $696m and raised full-year guidance to about $2.87bn — its fourth straight quarter of accelerating growth, with net retention up six points in a year to 120% and management saying more than half its network traffic now comes from AI agents. Fortinet is the other, and its quarter is described above.

The rest are riding along. Rubrik, which secures backup data and bills by volume, is decelerating — revenue growth has slowed from 51% to 39% over four quarters — and its August surge came from an AI-agent identity product launch that lifted the stock 18% plus the Black Hat pop. Zscaler, which routes employee traffic through its cloud to enforce zero-trust access, grows a steady 25% with no acceleration in three quarters; its move is entirely multiple, though at 7.6x forward sales it started from the least demanding valuation in the group. Palo Alto Networks, the sector's largest vendor at a $313bn market cap, printed 31% headline growth, but that includes the $21.1bn CyberArk acquisition; organic growth ran near 15%, and the quarter swung to a $183m GAAP operating loss on deal costs. CrowdStrike, which sells endpoint-sensor subscriptions, is quietly re-accelerating toward 26% growth, yet remains down 49% over twelve months and still costs about 43x trailing sales. Okta, which bills per employee identity managed, is the group's clearest laggard for a reason: growth of 11% is guided to 9-10% next year, net retention sits at 107%, and consensus price targets sit below the shares.

The month, decomposed

Strip each stock's two best sessions and the group's average 30-day gain collapses from roughly +11% to +1%. Those best days cluster on the same dates — around Cloudflare's August 6 report and the August 10 Black Hat session — so this was a sector-wide sentiment sequence, not seven independent stories. Trend-wise the group is not one trade either: Cloudflare and Fortinet have held unbroken uptrends since May 5, while CrowdStrike's 50-day average still sits below its 200-day and Zscaler turned positive only on August 13.

The price being paid for the story is steepest at Cloudflare: 39.9x forward sales and roughly 264x forward earnings, against a spring level near 25-28x enterprise value to forward revenue — a re-rating of about 50% on one earnings report. Whether the AI-security budget is real will be settled by the companies that haven't spoken yet.

The setup

Where it stands — Two of seven security vendors reported accelerating spend; the other five rallied, or didn't, on narrative alone.

Would confirm — CrowdStrike, Zscaler and Rubrik each print accelerating billings or revenue growth at their next reports.

Would invalidate — Palo Alto's organic growth stays near 15% and Zscaler's growth slips below 25%, leaving the rally two names wide.

Watch next — Palo Alto's fiscal-Q4 report in late August, then CrowdStrike, Zscaler and Rubrik in early September.

Valuation — Cloudflare at 39.9x forward sales versus roughly 25-28x in May; Zscaler at 7.6x forward sales for 25% growth.

The memory chips inside a broadband access box are the same commodity that artificial-intelligence data centers are now buying by the pallet. That is the single most important fact about Calix this year, and it explains a share price that has moved in the opposite direction to the business for seven months.

The record quarter nobody rewarded

Calix supplies access systems and premises equipment to broadband service providers — mostly small, regional and rural operators — alongside a subscription cloud platform that those operators use for marketing, support and network operations. Its June quarter produced revenue of $293.3m, up 21.3% from a year earlier, the fourth straight quarter above 20%. Operating income came in at $21.8m against $0.4m a year ago. Remaining performance obligations, the contracted revenue not yet recognized, reached a record $386m. Software and services revenue hit a record $50m. Management raised full-year guidance toward the top of its 15-20% growth range and bought back $69m of stock in the quarter.

The shares fell about 7% on the day, to roughly $35.78. The reason sat one line below the beat: third-quarter gross margin guided to 50.5-53.5%, with appliance gross margin already down 460 basis points sequentially on memory costs that customer surcharges only partly offset. Reported gross margin has fallen from 57.7% in the December quarter to 54.6% in June.

That pressure is not a Calix decision. Gartner expects prices for dynamic random-access memory (DRAM) to rise 47% in 2026, with roughly 70% of memory production allocated to AI data centers. Anyone building a box with memory in it is paying. ePlus's chief executive, Mark Marron, named the same shortage as the cause of shipment delays in his own June quarter.

Calix shares have traded below both their 50-day and 200-day averages continuously since 13 January, 146 sessions — the longest such run among the four names it is grouped with. The stock is down 30.6% over twelve months while earnings turned positive. It changes hands at 22.97x forward earnings against 50.05x trailing, and against roughly 31x implied at February's $53.32 price. Consensus has earnings per share going from $1.72 this year to $2.30 next. Management's own caution is the counterweight: it told investors 2027 could be "harder than 2026," and flagged satellite broadband as a threat to 5-10% of its addressable market.

One stock is the whole group

The four companies filed together under cloud infrastructure share no business model — and Calix is classified by data vendors as communication equipment, not software at all. Over the 30 sessions to 14 August, Fastly rose 49.7%, OpenText 7.9%, ePlus 2.4% and Calix 0.2%. Remove each name's two best sessions and the average turns slightly negative. Over twelve months Fastly is up 330.6%; the median of the four is roughly -0.8%.

Fastly runs an edge network billed by traffic delivered and requests served. Its June-quarter revenue grew 23.3% to $183.3m, the fastest in four years, gross margin climbed from 58.4% to 63.3% over four quarters, and gross profit grew 51.9%. Net revenue retention — spending by existing customers against a year earlier — reached 117%, from 104%. It raised full-year revenue guidance to $732-746m. It also remains loss-making under standard accounting, at -$15.6m. Growth is narrow: the top 10 customers, 37% of revenue, supplied 87% of sequential growth, and management called roughly $10m of the quarter episodic, noting three-quarters of World Cup matches fell in the June period. Its biggest single day, a 20.9% gain on 10 August, followed target hikes rather than any new disclosure. Forward price-to-sales is 6.34x, against roughly 3.6x implied at May's $17.00 close.

OpenText, which licenses enterprise information-management software to large corporations and governments, is the cheapest of the four at 5.56x forward earnings with a trailing free-cash-flow yield of 25.0% — and the reason is on the page. Fiscal 2026 revenue grew 1.5%, and fiscal 2027 is guided to $5,135-5,185m, a decline of 1-2%, with adjusted margin falling from 36.3% to 32-33% as it hires 300-plus salespeople.

ePlus, which resells other vendors' hardware and software for a gross-profit spread, is the mirror image of Calix. Revenue growth fell from 23.4% a year ago to 1.9% in the June quarter, gross margin narrowed to 23.3%, and net income dropped 19.7% to $30.3m. Its forward multiple went the other way, from roughly 13.4x a year ago to 16.17x.

The setup

Where it stands — Calix is growing above 20% with record backlog while memory costs have cut gross margin 310 basis points in two quarters. Would confirm — Third-quarter gross margin printing at or above the 53.5% top of guidance. Would invalidate — Third-quarter revenue below the $301m low end, or full-year growth guided back under 15%. Watch next — Calix's third-quarter report in late October, against a 52% gross-margin midpoint and $301-307m revenue guide. Valuation — 22.97x forward earnings versus 50.05x trailing and roughly 31x implied at February's $53.32 price.

The contested question in chip packaging this year has been who gets to do the work. Packaging — the step that bonds finished silicon into a working part, stacks memory beside a processor and tests the result — has become a harder constraint on artificial-intelligence hardware than wafer fabrication itself. The assumption underneath a lot of trading was that Taiwan Semiconductor Manufacturing Co. (TSMC), which fabricates most of the world's accelerators, would keep the high-value packaging in its own plants and leave the independent contractors with low-margin wirebonding and test hours.

The past three weeks made that assumption harder to hold. TrendForce, a Taiwanese research house, reported on 5 August that TSMC is expanding its outsourcing of the chip-on-wafer front-end step — the part of its CoWoS (chip-on-wafer-on-substrate) flow that had stayed almost entirely in-house — to ASE, Amkor and SPIL. TSMC's own monthly CoWoS capacity is set to roughly double by the end of 2026, and the supply gap is still projected near 20%. Separately, DigiTimes has put TSMC's 2026 outsourced CoWoS volume at 240,000 to 270,000 wafers, the bulk of it to Amkor.

The businesses agree

ASE Technology Holding, the world's largest outsourced packaging and test provider and a supplier of flip-chip, fan-out and 2.5D interposer services to fabless chip designers, reported second-quarter revenue of TWD 191.06bn, up 34.1% year over year. That was an acceleration from 17.4% in the prior quarter. Gross margin reached 21.0%, against 16.1% a year earlier; gross profit grew 75.5% and operating income more than doubled. On the 30 July call management put utilization at 80-85%, described pricing as "very friendly," and said fourth-quarter margin in the assembly and test business should break through the 30% ceiling it has historically treated as structural.

Amkor Technology, the Arizona-headquartered number two and the main independent alternative to TSMC's in-house packaging, grew revenue 25.6% to $1.898bn. Gross margin rose to 16.8% from 12.0%, and factory utilization moved from the low 50s to the high 70s. Its Computing segment set a record and was guided to grow about 30% sequentially in the third quarter. In July it announced a multi-year, $1.5bn agreement with NVIDIA, with NVIDIA prepaying to fund US capacity, alongside a ten-year packaging partnership with TSMC.

ChipMOS Technologies, a Hsinchu-based specialist in memory test and display-driver packaging, is the sharpest turn of the three. Revenue rose 28.7% to TWD 7.38bn, its highest quarterly level since 2014, with memory revenue up more than 46%. Gross margin went to 18.0% from 6.6%, and a loss a year ago became TWD 892m of net income.

The capital spending says the same thing. ASE raised 2026 capex by $2bn to a record $10.5bn, with 70% of equipment spending aimed at leading-edge capacity, and its SPIL subsidiary broke ground on 11 August on a plant near NT$100bn dedicated to integrating CoWoS. Among the Asian names, JCET of Shanghai reported first-quarter net profit up 42.7% and has committed RMB 7.8bn to a high-end packaging plant in Lingang; Tongfu Microelectronics is part of the same wave of Chinese expansion, which totalled over RMB 27bn in the first half. Powertech Technology of Taiwan and Hana Micron of Korea, both memory-weighted packagers, do not publish comparable figures in our data.

What the shares did, and what they cost

All seven fell hard between 28 July and 3 August, inside the sector-wide rout that shed more than $1trn from chip stocks on reports SK Hynix was slowing high-bandwidth memory expansion. Intel, AMD and Micron fell with them; this was not a judgment on packaging. Every one of the seven has since rallied off its trough, by an average near a third — Hana Micron by 48%, Amkor by 38%, ASE by 27% — which leaves the group up about 180% over twelve months.

The interesting number is what did not recover. Amkor is priced at 12.7 times trailing gross profit and 12.3 times forward, against 18.8 times in May, when its market value was $19.1bn versus $14.7bn now. Its forward price/earnings ratio of 23.8 sits below its trailing 26.3, EV/EBITDA is 11.3, and it is the only one of the three generating positive free cash flow, at a 3.5% yield. ASE has de-rated too, despite a higher share price: trailing P/E of 44.3 against 52.6 in May, because earnings outran the stock. ChipMOS is the cheapest on cash profits at 8.7 times EV/EBITDA.

The honest bear threads are not about AI. Amkor's Communications division, still its largest, was guided down high single digits sequentially and faces headwinds into the first half of 2027, and depreciation on the Arizona campus is expected to dilute margins in 2027-28. ASE expects negative free cash flow to persist into 2027 while it runs roughly 20 construction projects at once, and says execution — not demand — is the risk.

The setup

Where it stands — Advanced packaging capacity is a binding constraint on AI hardware, and the foundries are outsourcing more of it, not less. Would confirm — Amkor third-quarter revenue landing at or above the $2.05bn top end of guidance, with Computing up near 30% sequentially. Would invalidate — ASE's fourth-quarter assembly-and-test gross margin failing to exceed 30%, or its 2027 leading-edge doubling target being pulled back. Watch next — Amkor and ASE third-quarter results in late October 2026, and the first CoWoS output from newly outsourced lines. Valuation — Amkor at 26.3x trailing and 23.8x forward earnings; 12.7x gross profit against 18.8x in May. ASE 44.3x trailing, versus 52.6x in May.

The companies that build quantum computers for a living have started booking real revenue. IonQ, the biggest of them, reported $80m for the June quarter — more than the other four listed pure-plays combined, and its fifth consecutive record. But almost all of the growth across the group came from shipping hardware to national labs and universities, or from government research subcontracts. Very little came from customers renting time on qubits. That distinction is now showing up in the gross margin line.

The mix is the story

IonQ, based in College Park, Maryland, builds trapped-ion machines and sells access to them through Amazon's Braket, Microsoft's Azure Quantum and Google Cloud. Revenue rose 287% year on year, remaining performance obligations reached $485m, and full-year guidance went up to $280-290m. Gross margin went the other way: 56.5% in the December quarter, 23.8% in March, 24.9% in June. Operating margin was -421%. Management closed a $1.8bn all-stock purchase of chip foundry SkyWater Technology on 31 July, which the guidance excludes, and ended the quarter with $3.0bn of cash and investments. It also reported hard technical results — 99.99% two-qubit gate fidelity without ground-state cooling, and error correction validated on its Tempo system.

Infleqtion, the renamed ColdQuanta, makes neutral-atom computers, quantum sensors and atomic clocks for defense primes and government labs from Louisville, Colorado. Revenue of $12.63m was up 116%, all organic, and it raised full-year guidance to about $43m while reaffirming 30 logical qubits this year. Gross margin fell to 11% from 21% in the prior quarter, and the operating loss was $30.6m. It listed publicly this year; diluted shares went from 118.2m in March to 219.7m in June.

Quantum Computing Inc., a 75-person firm in Leesburg, Virginia, sells optimization software, photonic chips and is building a thin-film lithium-niobate foundry. Revenue was $5.55m against $61k a year earlier — but gross margin was -21%, meaning it sold below cost. Backlog stood at about $42.5m, and management disclosed on the call that it had not grown over the preceding 40 days. Between 70% and 80% of revenue is government subcontracting.

Rigetti, a 163-employee superconducting-chip maker in Berkeley, is the exception. Revenue rose 185% to $5.14m and gross margin improved to 42.6% from 31.4%, on sales of its Novera on-premises processors. It holds $541.3m in cash with no debt, and has an $8.4m order from India's C-DAC for a 108-qubit system due in the fourth quarter. Its Cepheus system runs at 99.1% median two-qubit fidelity against a stated 99.5% target for year-end; management named coherence time, currently 25-30 microseconds, as the binding constraint.

D-Wave sells the past, books the future

D-Wave Quantum, of Burnaby, British Columbia, sells annealing machines and cloud access to them. Revenue of $3.08m was flat year on year, and first-half revenue of $5.9m was down 67% against a period containing a $13.7m system sale. Gross margin lost 8.4 points to 55.4%, and the adjusted loss before interest, taxes, depreciation and amortization widened 85% to -$37.1m. The offsetting number is the order book: first-half bookings rose more than 1,120% to $35.5m, including a $20m system for Florida Atlantic University, and contracted future revenue rose 668% to $40.7m. Cash was $546.2m. Its gate-model roadmap runs to 100 logical qubits in 2032, and management does not expect meaningful gate-model service revenue before then.

The shared catalyst is federal. In May the Commerce Department signed letters of intent with nine companies for $2.013bn of CHIPS Act quantum incentives, taking minority equity stakes in each. D-Wave, Rigetti, Infleqtion and Quantinuum were each slated for roughly $100m; Rigetti's tranches of $19.9m, $22.2m and $18.5m are contingent on milestone completion.

What the buyer is paying

Forward price-to-sales runs 60.5x at IonQ, 67.1x at Infleqtion, 92.5x at Quantum Computing Inc., 181.6x at D-Wave and 264.7x at Rigetti. Those are lower than a year ago in the sense that matters — IonQ's market value rose about 65% while its forward revenue base grew 119%, Rigetti's rose 17% against 233% — so a genuine de-rating has happened. It has happened from levels where the de-rating barely registers. Infleqtion's thin margin puts it at 438x forward gross profit; Rigetti's and Quantum Computing Inc.'s price-to-gross-profit cannot be computed because gross profit is negative.

Per-share gains flatter the picture further. Diluted share counts rose 46.5% at IonQ, 58.9% at Quantum Computing Inc. and 26.0% at D-Wave over the year. Rigetti, at 11.7%, was the least dilutive.

The month, honestly

From 14 July to 14 August the five gained 15.5% on average, and 33.8% from the 24 July low. Remove each name's two best sessions and the month is negative — IonQ from +17.7% to -3.8%, Rigetti from +16.8% to -6.0%, Quantum Computing Inc. from +8.3% to -9.6%, Infleqtion from +23.1% to +0.8%. Average daily volume in the week to 14 August was 78% below the May-June run rate at D-Wave and 57% below at Infleqtion. Over twelve months, the four with a full year of trading are down 1.9% on average. This was a handful of headline days on falling participation, not steady buying.

The setup

Where it stands — Revenue is real and growing fast, but arrives as low-margin hardware and government subcontracts, priced at 60-265x forward sales. Would confirm — Gross margin stabilizing or rising sequentially at IonQ and Infleqtion in the September quarter. Would invalidate — Another quarter of margin compression, or D-Wave's $40.7m of contracted revenue failing to convert as guided. Watch next — Third-quarter results in early November; Rigetti's 99.5% two-qubit fidelity target is dated end-2026. Valuation — IonQ 70.1x trailing and 60.5x forward sales against consensus 2026 revenue of $285m; Rigetti 468.5x and 264.7x.

In the space of five trading days this month, the American advertising-software business delivered a verdict on itself. AppLovin, which runs the auction that routes mobile advertising dollars from marketers to app publishers, missed the midpoint of its own guidance for the first time since going public and lost a fifth of its value in one session. Two days earlier, Zeta Global and Digital Turbine had both raised full-year forecasts. Two days later, DoubleVerify agreed to be taken private.

That week is doing nearly all the work. Over the 30 sessions to 14 August, the six companies returned an average of 7.2%. Strip each name's two best days and that average becomes -9.3%: Digital Turbine goes from +26.7% to -15.6%, Viant from +2.4% to -10.5%. Only Zeta survives, at +7.0%. Every one of those gap sessions falls between 4 and 10 August.

The budget itself is not growing

The read-across is unusually clear this quarter. The Trade Desk, the largest independent buyer of programmatic advertising, reported quarterly revenue of $715m — up 3% — and guided the current quarter to at least $650m, roughly a 12% year-over-year decline. The agency group dentsu cut its 2026 global ad-spend growth forecast to 5.0% from 5.8% growth in 2025, and flagged search moderating to 3.4% as AI answers erode query volume. Whatever the three growing companies here are doing, they are not riding a rising market. They are taking share inside a flat one.

The one that grew and fell

AppLovin, a $106bn company that is paid a cut of every ad dollar routed through its bidding engine, reported June-quarter revenue of $1.92bn, up 52.8%. Operating margin was 77.7%. The shares closed at $311.14 on 12 August, a 52-week low, and sit 57% below their high of $733.60.

The growth is decelerating on a visible line — 68.2% a year ago, 59.0% in the March quarter, 52.8% now, with current-quarter guidance implying 46% to 48%. Management told investors the shortfall was timing: a model release slipped into the following quarter. Analysts at Bank of America, Piper Sandler and Wells Fargo cut ratings or targets anyway, questioning the 30% long-term growth target. The structural worry is take rate: third-party tracking showed e-commerce pixel installations slowing to about 750 in June from roughly 950 in May, with Google's AI ad stack encroaching and the European Union's Digital Markets Act threatening the attribution modelling underneath AppLovin's AXON engine.

The price now sits at 19.7x forward earnings against 24.1x trailing, on consensus 2026 earnings per share of $15.97 — about 70% growth. At the 30 June close of $515.23, the same estimate implied roughly 32x. Six weeks compressed the multiple by a third.

The share-takers

Zeta Global, a $7.3bn marketing platform that ingests opted-in consumer data to predict purchase intent for large enterprises, grew revenue 43.5% to $442.8m and turned a GAAP profit of $8.2m against a loss a year earlier. It raised full-year revenue guidance to $1.818bn, a 20th consecutive beat-and-raise. The catch is mix: gross margin compressed 300 basis points to 59.1%, and at 6.52x forward price-to-gross-profit Zeta is the most expensive name here, sitting at a 52-week high.

Digital Turbine, which pre-loads and recommends apps on carrier handsets, beat revenue estimates by about 8.5% at $166.0m, up 26.8%, and raised full-year guidance to $650-670m. Net leverage fell to 2.5x from above 5.0x. It is still loss-making on a GAAP basis, at -$3.2m, and has risen 195% in three months — but short interest near 11% of float argues this was a repricing rather than a squeeze. It trades at 17.2x forward earnings.

Viant, an $854m demand-side platform paid on media dollars transacted, is the one the market has not rewarded. Revenue grew 33.9% to $104.3m and the current quarter is guided to 27% growth. Connected-TV spend on its platform rose about 50% and now exceeds half the total. Chief executive Tim Vanderhook drew the comparison directly, contrasting Viant's growth with The Trade Desk's guided decline. Viant holds $193m of cash and no debt, at a 6.4% trailing free-cash-flow yield — and its month is -10.5% once two sessions come out.

Two that stopped being businesses

DoubleVerify, which measures whether digital ads were seen by real people in safe places, grew revenue 2.5% to $193.8m. Its Activation segment, the buy-side product, shrank 1%. Nielsen, backed by Elliott, agreed to pay $13.60 a share in cash; the stock closed at $13.29, a 2.3% spread, and the company has withdrawn all guidance. Semrush, the search-visibility subscription business, was bought by Adobe for $12.00 a share in a deal that closed on 28 April. Its price has not moved since. Its final reported quarter grew 14.6%, down from 22.4% a year before.

The setup

Where it stands — Three of these companies are growing 27%-44% inside a flat ad market; the largest is growing 53% and trading at a 52-week low. Would confirm — AppLovin's September-quarter revenue lands within the $2.055-2.085bn guided range, ending the miss. Would invalidate — AppLovin guides December-quarter growth below 40%, or Zeta's gross margin falls further from 59.1%. Watch next — AppLovin, Zeta, Viant and Digital Turbine all report again in early November; the Nielsen-DoubleVerify deal is expected to close by the first quarter of 2027. Valuation — AppLovin at 19.7x forward earnings against 24.1x trailing, versus roughly 32x implied on the same estimate at its 30 June close.

Roughly a year ago, Atlassian did something the rest of the software industry had been dreading. The Sydney-based maker of Jira, Confluence and Trello — project-tracking and collaboration tools billed largely per named user — disclosed a decline in enterprise seat counts, the first in its history. If artificial-intelligence agents were going to write code and file tickets in place of people, this was what it would look like on an income statement. The shares lost about a third of their value.

On 6 August, Atlassian reversed the narrative. Revenue for the June quarter came in at $1.77bn, up 27.6%, against consensus of $1.66bn. Cloud revenue accelerated to 31% growth. Gross margin reached 86.9%, and operating income swung to a positive $211.7m from a small loss. Management said seat expansion in core Jira and Confluence was strong and that AI adoption was increasing demand for planning and coordination — its Model Context Protocol server, the plumbing that lets outside agents act inside Jira, passed 1m monthly active users. The stock rose 35.8% in one session.

That single day is the largest fact in this story, but it is not the whole of it.

Three defenses, one capitulation

JFrog, which runs Artifactory — the repository every binary and container passes through between source control and production — reported the same week. Revenue grew 28.7% to $163.8m, cloud revenue rose 53%, and net dollar retention reached 121%, three points better than a year earlier. Customers paying more than $1m a year went from 61 to 97. Management's explanation is the mirror image of the bear case: AI tools generate more builds, so more artifacts flow through the toll booth. It also announced integrations with Anthropic's Claude Code and the Cursor editor.

GitLab, whose single-application DevSecOps platform bundles source control, continuous integration and security scanning at a per-developer price, is the ambiguous one. Its first-quarter results on 2 June showed revenue up 23% and net retention of 117% — decent numbers, but the fourth straight quarter of deceleration from 29.2%. Gross margin has compressed about 250 basis points. Alongside those results the company disclosed a 14% workforce reduction and guided the year to roughly 17.8% growth. Consensus earnings for the year have fallen, not risen. GitLab has reported nothing since; the past month's 29.8% gain rests on an expanded Google Cloud partnership and a broad rotation back into enterprise software that also lifted Salesforce, ServiceNow and Workday sharply on 27 July.

PagerDuty is the capitulation. It routes software alerts to on-call engineers and charges per responder. Annual recurring revenue was flat at $496m in the April quarter, revenue grew 1.0%, and the company guided the full year to 0.7% growth. Consensus has revenue declining by fiscal 2029. The shares are up 10.7% on the month and 75.8% over three, on speculation about activist involvement and a possible sale rather than anything the business did.

The one that fell

Cadence Design Systems is not a DevOps vendor at all — it licenses the electronic-design-automation software chipmakers use to lay out and verify silicon — but it is the only name here going down, and it is the most interesting workup. June-quarter revenue grew 24.2% to $1.584bn at a 28.4% GAAP operating margin. Backlog hit a record $8.1bn, first-half bookings rose about 55%, and management delivered what it called the largest single-quarter guidance raise in company history. The shares are down 12.6% over the month, dating to 17 July, when Moonshot AI said its Kimi K3 model completed a full chip-design flow using only open-source tools. Its forward multiple has compressed to 39.9x earnings from 64.2x trailing — the only compression among these five.

What the prices now assume

Everyone else has re-rated hard. Measured against gross profit, Atlassian has gone from 4.46x in May to 7.64x; GitLab from 4.57x to 8.26x; JFrog from 13.50x to 24.93x; PagerDuty from 1.36x to 2.59x. So the advance is not purely delivered earnings. Atlassian is the cheapest of the four on 26.7x forward earnings and 6.85x forward gross profit, and it has guided next year to roughly 13% growth as legacy Data Center revenue falls about 17%. JFrog at 100.8x forward earnings and GitLab at 52.5x on declining forward earnings leave the least room. PagerDuty, at 9.05x forward earnings and an 11.5% trailing free-cash-flow yield, is priced as an asset rather than a growth business.

All four DevOps names have held rising trends since spring, with GitLab and PagerDuty crossing into clear uptrends in mid-August. Two of them earned it with results. Two have not yet been asked.

The setup

Where it stands — Three of four seat-billed vendors reported evidence that AI is expanding their billable units; PagerDuty reported the opposite and rallied regardless.

Would confirm — Atlassian holding cloud growth near 25% and continued core seat expansion in the September quarter.

Would invalidate — GitLab net retention slipping below 115%, or Atlassian conceding renewed seat contraction at any point in fiscal 2027.

Watch next — PagerDuty reports second-quarter results on 27 August 2026; guidance midpoint currently implies flat revenue.

Valuation — Atlassian 26.7x forward earnings and 6.85x forward gross profit, against 4.46x gross profit in May; JFrog 100.8x forward earnings.

In the last week of July, five businesses that sit between investors and the markets they trade reported quarterly results within six days of one another. In the middle of that week, Intercontinental Exchange — owner of the New York Stock Exchange, 13 regulated exchanges and a growing bond-data arm — agreed to buy MarketAxess, the electronic venue where institutions trade corporate bonds, for $167 a share in cash, a 33% premium and about $6.0bn of equity value. ICE filed the deal with the Securities and Exchange Commission on 30 July; it expects to close in the first half of 2027.

That week did the work. Across the six names — ICE, CME Group, Tradeweb, Morningstar, CoStar and Starwood Property Trust — the past month is worth about 10% on an equal-weight basis. Remove each company's two best sessions and the group is down about 2%. The 22–28 July stretch alone delivered an average 7.3 points, led by Morningstar's 16.4%. (Local price records are missing three sessions in that stretch, so the 28 July close bundles roughly four days of move.)

The fear has not lifted for everyone

The premise being tested here is that markets have been de-rating anyone who sells financial information by subscription, on the argument that large language models make that data cheap to reproduce. That trade is still running. Over the same 30 days S&P Global, the ratings and index company, fell 5.8% and index provider MSCI fell 8.5%; over twelve months S&P Global is down 25.8% and FactSet down 24.4%. S&P Global's shares hit a two-year low in February after guiding 2026 profit below expectations. Morgan Stanley's equity strategists have argued the threat is uneven — owners of proprietary datasets fare better than aggregators of other people's.

By that test the six divide cleanly.

Two businesses the numbers support

Morningstar, which sells fund research, ratings, indexes and the PitchBook private-markets database to advisers and asset managers, is the cleanest case. Second-quarter revenue was $663.2m, up 9.6%. Operating income rose 28.4%, taking operating margin to 24.2% from 20.7% a year earlier — the third straight quarter above the 21.5% it managed for all of 2025. Free cash flow nearly doubled to $122.5m and adjusted earnings of $3.10 a share beat the $2.83 consensus. The stock trades at 17.0x forward earnings against 19.5x trailing, the cheapest forward multiple of the six, on consensus 2026 earnings of $12.20 a share. The blemish is PitchBook: $174.7m of revenue, but growth of only 4.9% and a slightly lower margin.

ICE is a milder version. Revenue of $3.611bn grew 10.7%, with operating margin at 55.0% against 39.8% a year earlier. Recurring revenue reached a record $1.4bn, and management raised full-year growth guidance for its fixed-income and data unit to 7–8% from mid-single digits. Index exchange-traded fund assets under management hit $922bn, up 29%. The quarterly buyback went to $400m from $350m. Shares change hands at 19.2x forward earnings against 21.8x trailing.

Two the numbers do not

CME Group, the Chicago futures exchange where traders hedge interest rates, equity indexes and crops, grew revenue just 0.8% to $1.706bn — down from 14.5% growth the prior quarter. Average daily volume of 29.8 million contracts was the second-highest second quarter ever, but the average rate per contract slipped to $0.678 from $0.690 a year ago. Market data revenue was a bright spot at a record $238m, up 20% and the 33rd consecutive quarter of growth. At 22.8x trailing and 22.0x forward earnings, less than a turn separates the two — the market is pricing almost no growth. CME's own event contracts run above 4 million a day, while combined monthly volume at prediction venues Kalshi and Polymarket reached about $44.8bn in June, from under $5bn last September.

Tradeweb, which runs electronic platforms for trading rates, credit and money markets for roughly 2,500 institutions, grew revenue 9.0% to $558.9m, down from 21.2% a quarter earlier. Average daily volume rose 18.2% to $3.0tn and market share hit records, but average fees per million fell 10.3% in long-tenor swaps and 11.4% in cash credit — volume outrunning revenue. It is the only one of the six whose forward multiple, 26.4x, sits above its trailing 25.3x.

The one the chart got right

CoStar, which sells commercial-property data and runs Apartments.com and Homes.com, has been in a downtrend for 191 straight trading sessions since 4 November, its 50-day average below its 200-day throughout. Revenue grew 18.4% to $925m, the 61st consecutive quarter of double-digit growth, and the residential segment turned its first quarterly profit at $12m of adjusted earnings before interest, taxes, depreciation and amortization. But net new bookings — the forward order book — fell 26% to $69m, and management cut 2026 revenue guidance to $3.715–3.755bn after restructuring its auction arm and cutting the Homes.com inside sales force by 21%. Trailing earnings are near zero; the usable anchors are 3.51x forward sales and 1.65x book.

Starwood Property Trust is not an exchange or a data business at all. It is a $6.0bn commercial mortgage lender, trading at 0.92x book, and it fell 2.2% over the month — the only decliner. It earned $0.40 a share of distributable earnings against a declared $0.48 dividend, the second straight uncovered quarter.

The setup

Where it stands — A month's gain concentrated in one earnings week and one takeover bid, with only Morningstar and ICE delivering results that match it. Would confirm — Morningstar holding operating margin above 24% and CoStar's net new bookings returning to year-on-year growth in the third quarter. Would invalidate — CME's rate per contract falling below $0.678 again, or Tradeweb's revenue growth slipping below 9%. Watch next — Third-quarter results in late October; ICE's MarketAxess purchase is due to close in the first half of 2027. Valuation — Morningstar 17.0x forward against 19.5x trailing; ICE 19.2x versus 21.8x; CME 22.0x versus 22.8x; Tradeweb 26.4x above 25.3x.

Paycom, which sells payroll, benefits and time-and-attendance software to small and mid-sized American employers, told investors on 5 August that its own use of artificial intelligence had removed more than $100m a year from its research and development budget, plus another $30m of fees paid to third-party AI model providers. On the same call, management said the thing that actually generates its revenue — the number of people its clients employ — was stable, consistent with historical levels, and not accelerating.

Both statements are in the same quarter, and together they explain why the shares have run without the underlying employment base doing anything.

Billed per head, paid on margin

Paycom and Paylocity, a rival selling the same cloud payroll and human-capital management stack through its own direct sales force, are paid per employee per month. Their top line moves with client headcount, which is why the debate about AI agents replacing white-collar workers lands on them harder than on most software. The macro backdrop is not helping: the US economy shed 23,000 jobs in July, and ADP counted just 44,000 private hires, down from 95,000.

So the growth line is decelerating on schedule. Paycom's annual revenue growth has gone 23.2% in 2023, 11.2% in 2024, 8.9% in 2025; the raised 2026 guide of $2.197bn-$2.212bn implies 7-8%. Second-quarter revenue was $531.2m, up 9.8%.

The profit line went the other way. Operating income rose 50% to $168.5m, lifting operating margin to 31.7% from 23.2% — 850 basis points in a year, on single-digit revenue growth. Full-year adjusted EBITDA is guided to a record 46% margin at the midpoint, with free cash flow above $650m. Diluted shares fell from 56.3m to 45.9m after $1.4bn of buybacks year-to-date, roughly a fifth of the company retired in six months. Consensus 2026 earnings per share of $11.96 is up 29.5% on revenue growth of 7.6% — earnings compounding at four times the rate of the top line.

One leg the bear case expected to break has not. Interest earned on client payroll balances held between collection and disbursement was $26.0m in the quarter, on an average balance of about $2.9bn, up 9%, guided to roughly $105m for the year. The Federal Reserve is holding at 3.5%-3.75%, with the market now leaning toward hikes rather than cuts.

Paylocity pays more for less

Paylocity's fiscal fourth quarter, reported 4 August, brought revenue of $444.7m, up 11.0%, and operating margin of 19.0% against 16.5%. Its guidance is the problem: fiscal 2027 revenue of $1.880bn-$1.895bn, roughly 7% growth, on a deceleration that runs 37.8%, 19.4%, 13.7%, 11.0%. Consensus fiscal 2027 earnings of $8.83 sit only 9.0% above this year's.

That is the divergence inside the pair. Paycom trades at 23.1x trailing and 18.2x forward earnings, 11.8x trailing EV/EBITDA and a 7.7% free-cash-flow yield. Paylocity is at 29.7x trailing and 16.8x forward, 15.6x EV/EBITDA, 5.6% free-cash-flow yield — a similar forward multiple on roughly a third of the earnings growth. Both multiples have expanded sharply from mid-May: Paycom's forward P/E from about 11.2x, Paylocity's from about 11.8x. Neither has been accompanied by estimates running ahead of management; consensus for both sits at the midpoint of guidance issued this month.

The third name isn't in this business

ZoomInfo, which trades as GTM and sells business contact data and sales-engagement workflow, is grouped with the payroll vendors and belongs nowhere near them. Its second-quarter revenue was $310.4m, up 1.2%, with a $650.5m goodwill impairment driving a $643.7m net loss. Net revenue retention slipped to 89%. It cut headcount 15% year on year and is moving customers off per-seat pricing toward consumption billing as go-to-market work migrates into large language models. Analysts expect revenue to decline 2.6% this year and 2.1% next. Its price-to-gross-profit has gone from 0.89x in late July to 1.13x with nothing in the numbers changing. This is the one name where AI is visibly eating seats, and it is the one where the rally has no fundamental support.

What actually moved the shares

The advance is narrow and partly borrowed. Strip each name's two best sessions from the past 30 days and the group's average gain falls from about 34% to roughly 3%. Paycom's 6 August jump of 23.8% on four times normal volume was a genuine earnings reaction. Paylocity's two best days, 27 and 29 July, came a week before it reported anything — they were the software-wide AI rotation that lifted Salesforce, ServiceNow and Workday; its own print moved it 2.5%. The mid-August leg came from reports that Silver Lake was weighing a roughly $51bn purchase of Workday, the HR and financial software vendor, which lifted peers in sympathy. Workday rose about 40% over the month; ADP and Paychex, the incumbent payroll processors, rose about 10% each.

The setup

Where it stands — Paycom's re-rating rests on AI-driven cost savings and buybacks, not on a recovery in the client headcount it bills for. Would confirm — Third-quarter operating margin holding above 30% with 2026 revenue guidance maintained at 7-8%. Would invalidate — Paycom naming client employment declines, rather than stability, as a drag on recurring revenue growth. Watch next — Paycom's third-quarter results in early November; Paylocity's fiscal first quarter, guided to roughly 7% growth. Valuation — Paycom 23.1x trailing, 18.2x forward, up from about 11.2x forward in mid-May; Paylocity 29.7x and 16.8x.

Getty Images filed its quarterly report on 10 August. It contained the sentence no operating company wants to publish: substantial doubt about the company's ability to continue as a going concern. The next session, shares of the visual-content marketplace, which licenses news, sports and stock photography under the Getty Images, iStock and Unsplash brands, fell 42.6% on 11.6m shares.

The same filing did not show a collapsing business. Second-quarter revenue was $229.1m, down 2.5% from a year earlier. Annual subscription revenue, now roughly 59% of the total, rose 7.1%. Editorial revenue, the news and sports archive, rose 9.2%. Adjusted earnings before interest, taxes, depreciation and amortization were $62.3m, down 8.4%.

The break was financial

What broke was the other side of the ledger. Getty carries $2.1bn of debt against $51.6m of cash and a market capitalization of about $129m. Free cash flow was minus $122.6m in the quarter, after $110.9m paid out on warrant litigation. The $150m revolving credit line is fully drawn at 7.76%. Management declined to give 2026 guidance and hired Guggenheim Securities on 21 July to review financing alternatives.

The proximate cause was a deal that did not happen. Getty's board resolved on 30 June not to divest Shutterstock's editorial business as Britain's Competition and Markets Authority (CMA) required, and the $3.7bn merger was terminated on 7 July. The termination triggered redemption of $628.4m of Getty's 10.5% senior secured notes. Eighteen months and more than $100m of professional and financing costs produced nothing.

Shutterstock, which licenses images, footage, music and 3D models under the Shutterstock, Envato, Pond5 and TurboSquid brands, took the other half of the damage. Its shares fell 19.0% on 28 July, when it reported revenue down 16.9% to $221.8m, wrote off $163.4m of goodwill, withdrew guidance for the rest of 2026 and cancelled its investor call. Analysts now model revenue falling 23.3% this year and another 8.0% next. At 0.27x forward sales and 0.57x book value, the shares price an orderly wind-down, not a mispriced asset.

The large names accelerated

Adobe, the $105bn maker of Photoshop, Illustrator and Acrobat and the presumed first casualty of generative design tools, grew revenue 12.7% year on year last quarter to $6.62bn. That is the fifth consecutive quarter of acceleration, from 10.5%. Gross margin was 89.2% and operating margin 33.8%. The company raised full-year revenue guidance to $20.5-20.6bn and disclosed artificial-intelligence-first annual recurring revenue above $500m, more than triple a year ago, with its Firefly image-generation product approaching $300m.

Adobe trades at 15.1x trailing and 10.8x forward earnings, against a forward multiple that spent most of the past decade between 25x and 40x. Free cash flow yield is 10.1%. The stock bottomed at $193.41 on 25 June and has risen 36.5% since, including a 14.1% session on 28 July after a CLSA upgrade on a day the technology sector fell.

Figma, the browser-based collaborative design tool that went public last year, grew 48.2% to $370.1m, its third straight quarter of acceleration, with net dollar retention of 136% and full-year guidance raised by $40m. The cost of that growth is visible: GAAP operating income was minus $117.3m, against plus $2.1m a year earlier, and the stock fell after hours on the expense line despite beating estimates. At 10.65x forward gross profit it is valued at roughly 2.4x Adobe on the same measure, and holders of about 54.1% of Class A stock see their lock-up end on 31 August.

Chegg, which sold homework-answer subscriptions to students, is the one name where the substitution story is undisputed: revenue fell 50.7% to $51.8m and third-quarter guidance is $43-44m. Even so it produced $9.1m of adjusted EBITDA and $6.4m of free cash flow, and holds $38.5m of net cash against an $88m market value while pivoting to a job-search and interview-coaching product launching this quarter.

Where the AI evidence actually sits

The clearest management statement linking generative AI to lost volume came from Getty, which said search engines prioritizing AI-generated answers hurt iStock's new-customer acquisition. Subscription retention fell to 88.4% from 93.4%. That is real, and it is happening at the cheap, self-serve end of image licensing, not in professional creative suites. One analyst arguing the market overestimates the threat noted in August that there is still no material impact from AI-powered design competitors.

One technical note, because the averages mislead. Over the past month Adobe rose 14.5% and Figma 8.6% while Getty fell 56.3% and Shutterstock 31.0%. Remove each company's two worst sessions and three of the five are higher. This was not a slow bleed; it was three dated accidents.

The setup

Where it stands — Getty and Shutterstock broke on a failed merger and a balance sheet, while Adobe and Figma both posted accelerating growth. Would confirm — Adobe's next quarter holding revenue growth above 12% with artificial-intelligence recurring revenue above $500m. Would invalidate — Adobe reporting Creative Cloud subscription growth below high single digits, or Figma guiding below 35%. Watch next — Figma's lock-up on about 54.1% of Class A shares expires 31 August; Getty's financing review is open-ended. Valuation — Adobe at 15.1x trailing and 10.8x forward earnings, against a forward multiple historically between 25x and 40x.

Only one company in this corner of data software has reported a quarter since the spring, and it was punished for it.

Datadog, which sells software that watches other companies' software and bills by the server monitored and the log stored, published second-quarter results on 6 August. Revenue reached $1.12bn, with growth accelerating to 35.6% year over year — a fourth consecutive quarterly acceleration, the clearest business improvement any of these names can show. Management raised full-year guidance to $4.45-4.47bn, above the $4.374bn analysts had modelled. The shares closed down 20.4% that day, at $229.29 against $288.13 the session before, because its largest customer — a nine-figure account running 17 Datadog products, which analysts widely believe to be OpenAI — will reduce usage from the third quarter. By 14 August the stock had recovered to $255.46, still 11% below where it went in.

That is what usage-based billing looks like when the meter turns down, and it is worth holding next to what the other two did on no numbers at all.

The rally without a print

Snowflake runs a cloud data platform that stores and queries corporate data and charges by the computing credit consumed. Its last set of financials, published on 27 May, showed product revenue of $1.33bn, up 34% and accelerating from 30% the prior quarter, with net revenue retention of 126% — existing customers spending a quarter more than a year earlier. Everything since has been product news rather than delivered revenue: the Cortex AI Gateway launched to govern enterprise AI agents, with Alteryx, Aembit and 1Password integrating in late July and early August.

The re-rating on those announcements is the largest thing in this story. Snowflake's market capitalization has gone from $49.3bn in early May to $114.0bn. Its price-to-gross-profit — the cleanest lens on a company still losing money at the operating line — has moved from 15.7x to 33.7x over the same stretch. The operating loss is genuinely narrowing, to -23.4% of revenue from -42.9% a year earlier, and consensus has fiscal-2027 revenue at $6.098bn, up 30.9%, slowing to about 26% thereafter. Growth is real; the price has moved considerably faster than it.

Elastic sells Elasticsearch, the open-source search and log-analytics engine that companies use for security monitoring and application troubleshooting. It rose 39.3% in the month, the most of any member, including six consecutive up sessions into 14 August worth 23%, on an upgrade to its security operations product. Strip out its two best days and the month is still +18.4%, so this is a trend, not a spike. But Elastic has not reported since February, and its revenue growth is flat to slowing in the mid-teens — 16.0% last quarter, with consensus at 15.0% and then 14.5%. Its trailing price-to-earnings of 24.5x is an artifact of a one-off gain that produced $435.9m of net income on $450.7m of revenue. On gross profit it is the cheapest name here by a wide margin: 5.94x forward, against Snowflake at 27.8x and Datadog at 26.1x. It is also the only one where a discount is still visible.

The bitcoin half went the other way

Strategy, formerly MicroStrategy, holds bitcoin funded by equity and preferred issuance, atop a shrinking legacy analytics software business that grew 6.9% last quarter to $122.4m. It fell 4.5% over the month and 50.2% over three months, against a 22.1% fall in bitcoin — roughly 28 points worse than the asset it exists to hold. Its market capitalization dropped below the market value of its coins for the first time on 27 June, and it now trades at 0.70x book. Diluted shares still rose 15.2% year over year, to 343.3m.

The financing has inverted. Strategy added $525m from at-the-market sales of common stock to a reserve now at $3.75bn, about 25 months of expected preferred dividends, while buying back 288,930 preferred shares for roughly $25.0m. Common holders are funding the preferred. That is why STRK, Strategy's perpetual preferred paying $8 a year per $100 of stated amount, rose 11.7% on the month while the common fell — a transfer down the capital structure, not coin exposure.

One name has left the group entirely: Clearwater Analytics, which ran investment-accounting software for insurers and asset managers, was taken private for $8.4bn at $24.55 a share on 25 June by Permira and Warburg Pincus.

The setup

Where it stands — The month's gain came from Snowflake and Elastic, neither of which has reported since spring; Datadog reported and fell. Would confirm — Snowflake's next quarter holding product revenue growth above 30% with net revenue retention at or above 126%. Would invalidate — Datadog's third quarter landing below its $1.135bn guidance floor as the large AI customer's usage rolls off. Watch next — Snowflake's fiscal second-quarter results, due in late August 2026. Valuation — Snowflake at 33.7x trailing and 27.8x forward gross profit, against 15.7x in early May; Elastic at 5.9x forward.