DK Street Journal

Between the evening of 5 August and the close of 7 August, four of the nine business-software companies in this group put numbers on the table, and every one of them raised guidance. That matters because these are not consumer names whose fortunes turn on a season: they sell the infrastructure other firms build software on, and their contracts renew. The week's disclosures were the first hard test of whether the group's spring rally rested on anything.

It largely does — but not evenly, and not cheaply.

The four that reported

Twilio, the largest provider of programmable messaging and voice interfaces for software developers, delivered the cleanest inflection. Revenue reached a record $1.5bn with adjusted earnings of $1.47 a share against a $1.32 consensus, the fourth straight quarter of accelerating growth (14.3% to 20.0% to 22.0% year over year), dollar-based net expansion rose to 116% from 114%, and free cash flow grew 34% to $353m. Full-year revenue-growth guidance went to 18–18.5% from 14–15%. The catch is disclosed in the company's own outlook: organic growth is guided to step down to 11–12% next quarter from 17%, so the reported figure is increasingly flattered by acquisitions and carrier fees.

Atlassian, which sells Jira, Confluence and Trello — the tools software teams use to plan and document work, priced largely per user — produced the biggest single-day move. Adjusted earnings of $1.87 beat a $1.50 estimate on $1.77bn of revenue, with cloud revenue growth re-accelerating to 31% against guidance of 23% given in February; remaining performance obligations rose 44%. That is the most important fundamental datapoint in the group, because Atlassian in May reported its first-ever decline in enterprise seat counts — the event that made per-seat software the market's favourite artificial-intelligence casualty. This quarter cut against it. But the same filing guides fiscal 2027 revenue growth to roughly 13%, down from 26%, as a pull-forward from the March-2029 retirement of its self-hosted Data Center product laps, with adjusted operating margin falling from 36% to 25%.

JFrog, which runs the repository where enterprises store the compiled software packages their applications are built from, is the group's purest read on machine-written code. Revenue of $163.8m beat its own $154–156m guide and the full-year midpoint went to $650m from $630m; cloud revenue grew 53% to $87.5m, customers spending over $1m annually rose to 97 from 61, obligations grew 38%, and adjusted operating margin improved 470 basis points to 19.9%. Management tied consumption above contracted minimums directly to coding agents pushing more binaries through pipelines. Against that, 47% of revenue is still self-hosted and grew 9%.

Dynatrace, which monitors applications and infrastructure for outages, supplied the group's hardest evidence that AI adds billable volume: adjusted earnings of $0.48 beat $0.44, annual recurring revenue grew 17% to $2.14bn, and more than 1,000 customers now monitor AI workloads in production, up from 850, with that group consuming 1.5x faster than the rest.

The five that did not

Veeva, whose software is the system of record for pharmaceutical clinical trials and regulatory submissions, is growing at a steady 16.3% but has lifted operating margin from 24.8% to 30.9% in a year and raised full-year revenue guidance to $3.635–3.645bn in June. Salesforce, the customer-relationship giant, is the group's one documented divergence: revenue growth has re-accelerated from 8.6% to 12.1% to 13.3%, net income rose 37%, and it trades at 13.6x forward earnings with a 9.3% free-cash-flow yield — yet it is the only member still below its long-term trend line and is down 20.7% over twelve months.

Nutanix, which sells hybrid-cloud infrastructure software as the alternative to VMware, is the contradiction. Revenue growth has halved across four quarters, from 19.2% to 10.0%, while the shares re-rated from 4.1x to roughly 6.2x sales. Samsara, which fits trucks and heavy equipment with sensors and sells the software that watches them, is growing 30.5% but carries 11.7x forward sales and a 1.0% cash yield. Bill.com, which automates supplier payments for small businesses, is cheapest at 14.25x forward earnings and an 8.0% cash yield, and slowest, at 13.5% growth. All three report on 26 and 27 August and have disclosed nothing since spring.

Verdicts

On the business, CONFIRMS at Twilio, JFrog, Dynatrace, Atlassian's fourth quarter and Veeva; CONTRADICTS at Nutanix; INCONCLUSIVE at Bill.com and Samsara until late August.

On valuation, mostly CONTRADICTS the idea of runway from here. Against early-May readings, price-to-sales has expanded about 74% at JFrog (to 17.1x forward sales, 93.9x forward earnings, a 1.6% cash yield), 66% at Atlassian (24.5x next-year earnings against 13% guided growth), 53% at Twilio, 52% at Nutanix and 38% at Veeva. Dynatrace, up about 21%, is the least stretched of the winners; Salesforce, flat on 3.42x forward sales, is the only POSSIBLE DISLOCATION. The macro backdrop is unhelpful: the 10-year Treasury yield sits near 4.6%, and long-duration software is where higher rates bite hardest.

The tape agrees with the business, but roughly half of it arrived in four sessions. Of the group's 19.9% average 30-day gain, about 9.4 percentage points came from those four print days; strip them and eight of nine members are still up, averaging about 10.5%. Twilio's entire month is the single earnings session — it had fallen 11.6% before it. Atlassian's was not: it had already recovered 28% beforehand. Dynatrace handed back a third of its pop the next day. The wider rotation out of AI-hardware trades and back into application software that had already reset explains the quiet members better than anything they have said.

The setup

Where it stands — Four of nine reported and beat; five re-rated 38–74% on sales in three months, and three report only in late August. Would confirm — Bill.com, Nutanix and Samsara beating on 26–27 August, and Twilio's Q3 organic growth landing above its guided 11–12%. Would invalidate — Atlassian cloud growth decelerating toward the 13% full-year guide, or JFrog's net dollar retention slipping below the 120% floor. Watch next — Nutanix and Bill.com report 26 August 2026; Samsara 27 August. Valuation — JFrog 17.1x forward sales versus 10.4x in May; Salesforce 13.6x forward earnings against 22.2x trailing.

Contract manufacturers have historically been the least capital-hungry link in the electronics chain: they buy components, assemble to a customer's design at low-single-digit margins, and hand back cash. That arrangement is being renegotiated in public. Celestica, a Toronto-based builder of routers, switches, interconnects and custom server racks for hyperscale cloud operators, announced a $3 billion equity offering on 5 August to fund investments supporting what it called unprecedented multi-year demand. The 9,677,419 shares were priced at $310 against a prior close of $362.76 — about 8.3% dilution on roughly 116 million diluted shares, before a 15% underwriter option — and the stock fell 14.8% the following session, the largest single-day move in the group this year.

The raise is a symptom, not an outlier. Celestica's capital spending has gone from 1.1% of revenue a year ago to 5.6% last quarter, with $1bn budgeted this year and a $1.5bn placeholder for next. Flex cut its free-cash-flow conversion guidance to about 40% from 60% on $1.5–1.6bn of capex. Celestica, Sanmina and Benchmark have all paused or deprioritised buybacks in favour of capacity. The assembly layer is becoming an asset-heavy business while the market is asking harder questions about who finances the artificial-intelligence buildout.

The tape moved before the numbers did

All six names — not four — stepped down from their strongest uptrend classification inside 22 trading days, sequenced by AI exposure: Celestica on 30 June, Sanmina 8 July, Jabil 14 July, Flex 17 July, Plexus 21 July, Benchmark 22 July. Every one of those dates precedes the company's own results. That is a sector de-rating, and it coincides with a credit event, not a demand event: tech bonds sold off on 23 July on fears about the scale of the debt-funded AI boom, hyperscalers and Nvidia issued $225bn of bonds by midyear, up roughly 974%, with Amazon paying 18–21 basis points extra on a $25bn deal, and loan investors began demanding better terms on AI credit in the week to 1 August. Drawdowns from June highs are far deeper than the group's 2.9% monthly average suggests: Celestica −32.8%, Sanmina −28.2%, Flex −24.9%.

The businesses went the other way

Celestica's June quarter revenue rose 62.4% to $4.699bn — a fourth straight quarter of acceleration — with operating margin at 9.75% against 9.42%, and it raised full-year guidance to $20.5bn and $11.30 of earnings per share while disclosing 2027 rack programmes for OpenAI and interconnect work for AMD.

Flex, the Singapore-founded manufacturer that also makes switchgear, busway and modular power systems for data halls, grew revenue 20.6% to $7.928bn with gross margin up from 8.70% to 9.42%; its Cloud & Power Infrastructure unit grew 35% to $2.2bn at a 9.7% operating margin and is guided to 65–75% growth this year. Its supposedly soft legacy book grew 12% with margins up 130 basis points. The shares still fell about 9% on the print.

Sanmina, the San Jose maker of backplanes, enclosures and printed circuit boards now carrying the acquired ZT Systems server business, grew revenue 69.7% to $3.464bn with operating margin up 173 basis points — the best expansion in the group — as cloud and AI reached 62% of revenue. It beat on both lines but guided the current quarter about 2% below consensus on legacy programme timing.

Jabil, the Florida manufacturer serving networking, storage, healthcare and semiconductor equipment customers, is the one decelerator: revenue growth slowed to 11.8% from 23.1% and operating margin slipped six basis points, though it lifted AI-related revenue guidance to about $13.6bn, up 50%. It last reported on 17 June, so it has had no late-July reset.

The two names assumed to have no AI exposure do. Benchmark Electronics, the Arizona builder of aerospace, defence, medical and semiconductor-equipment hardware, grew its advanced computing unit 71% and guided to a record $3.0bn year. Plexus, the Wisconsin manufacturer of medical, industrial and defence electronics, disclosed roughly $0.5bn of data-centre funnel and a record $4.5bn total funnel. Both are the least damaged on the tape — and, with Jabil, are what masks the declines elsewhere.

Verdict A — does the business explain the move? CONTRADICTS. Five of six accelerated, five of six expanded operating margins, and all six raised guidance into the selloff. Demand upstream also rose: the four largest cloud buyers lifted 2026 capital spending plans to roughly $725bn, up 77% — and were sold for doing so.

Verdict B — does valuation justify it? INCONCLUSIVE, and it splits. Celestica trades at 32.7x trailing and 27.9x forward against roughly 40x forward in this Desk's May reading; Flex at 46.9x trailing but 25.8x forward, versus 59x trailing in May; Jabil 42.1x/26.8x with a 4.2% free-cash-flow yield. Multiples compressed while guidance rose, so the last leg was earnings. Sanmina is the cheapest at 16.8x forward, 17.1x enterprise-value-to-EBITDA and a 9.5% free-cash-flow yield on consensus earnings growth of 103%. Plexus is the opposite: 31.5x forward, 29.0x EV/EBITDA, a 0.9% cash yield, 19.5% expected growth and the only contracting reported margins in the group. Benchmark sits at 27.8x forward with growth slowing to 13% next year.

The unresolved question is not orders. It is that Celestica's top three customers are about 65% of revenue, up from 51%, and it just funded their capacity with discounted equity.

The setup

Where it stands — Six assemblers beat and raised into a July de-rating; Celestica then diluted holders 8.3% to fund capacity.

Would confirm — Flex's Cloud & Power segment printing 45–55% growth next quarter and Celestica holding $20.5bn full-year revenue.

Would invalidate — Any of the six cutting AI-segment guidance, or a second equity raise in the group at a discount.

Watch next — Jabil's fiscal fourth-quarter results in late September, the first fresh guidance since 17 June.

Valuation — Group forward earnings multiples run 16.8x (Sanmina) to 31.5x (Plexus), against roughly 28–59x trailing readings in May.

Getty Images' board voted on 30 June to walk away from its $3.7bn takeover of Shutterstock, its largest rival, after Britain's Competition and Markets Authority demanded Shutterstock divest its editorial business as a condition of clearance — a remedy the parties would not accept, despite US antitrust approval in February. The agreement lapsed in early July.

For Getty, which licenses photographs, video and a proprietary archive covering roughly 160,000 news, sport and entertainment events through the Getty Images, iStock and Unsplash brands, the consequences were financial rather than strategic. The deal's failure triggered a special mandatory redemption on Getty's $628.4m of 10.500% senior secured notes due 2030, cost the company a $162m cash benefit the merger would have delivered, and prompted S&P Global Ratings to cut its credit rating from B to CCC+. Those notes were only settled and closed as part of an exchange offer that the merger was meant to take out. The board says it will retain an adviser on financing alternatives.

The business is not what broke

Getty's revenue rose 1.1% year on year in the first quarter to $226.6m, and 4.5% for full-year 2025 to $981.3m. Gross margin slipped from 73.1% to 66.8% and the company lost $206.1m last year, but this is not a collapsing top line. Getty also largely lost its landmark copyright case against image-generator Stability AI, when the English High Court ruled in November 2025 that AI model weights are not a 'copy' of training images; an appeal was granted in December.

Shutterstock is the genuine substitution casualty. The marketplace, which licenses images, footage, music and 3D models under the Shutterstock, Envato, Pond5 and TurboSquid brands, reported second-quarter revenue of $221.8m, down 17%, with content revenue down 17% on weak new-customer acquisition and a $163.4m non-cash goodwill impairment producing a $155.9m net loss. Gross margin fell from 60.3% to 57.7%. Management cancelled the 6 August earnings call and will issue no guidance for the rest of 2026, having already cut over $70m of annualised costs with $60m more targeted by year-end.

Verdict on the business: CONFIRMS at Shutterstock, CONTRADICTS at Getty.

The software names went the other way

Adobe, the $105bn maker of Photoshop, Illustrator and Acrobat, is the name most often cast as generative AI's victim. Its revenue growth has accelerated three quarters running — 10.5%, then 12.0%, then 12.7% to $6.618bn — and AI-first annual recurring revenue tripled past $500m, with its Firefly image generator nearing $300m and free creative users above 90 million. The soft spot is pricing, not seats: Adobe deferred planned second-half Creative Cloud price rises, which accounts for roughly half its reduced recurring-revenue growth expectation. Operating margin fell from 35.9% to 33.8%. It also closed a $1.9bn all-cash purchase of search-analytics firm Semrush in April.

Figma, the browser-based collaborative design platform that listed last year, grew revenue 48.2% to $370.1m, its third straight quarter of acceleration, with existing customers spending 36% more than a year earlier and full-year guidance raised $40m. It still fell 16.5% after hours on 5 August because third-quarter guidance implies barely 1% sequential growth and reported operating expenses have nearly doubled to $426.9m, producing a $117.3m operating loss.

Chegg, the shrunken online-homework service, is the arithmetic curiosity: revenue down 50.7% to $51.8m, yet adjusted EBITDA of $9.1m, $6.4m of free cash flow and $38.5m of net cash against a $102.5m market value.

Where the multiples sit

Adobe: 15.2x trailing and 10.9x forward earnings, 10.8x enterprise value to EBITDA, a 10.1% free-cash-flow yield, against consensus for 12.0% revenue growth this year and 9.1% next. Getty: 0.19x sales looks cheap, but 17.6x trailing enterprise value to EBITDA on $183m of equity says the stock is a leveraged option on refinancing. Shutterstock: forward price-to-sales of 0.28x sits above trailing 0.24x — the market stating outright that revenue shrinks from here — despite a 40.6% free-cash-flow yield and 0.60x book value. Figma: 7.72x forward sales, down from 8.86x trailing. Valuation verdict: possible dislocation at Adobe; justified de-rating at Shutterstock and Figma; INCONCLUSIVE at Getty, where equity value depends on a bond negotiation.

Against cybersecurity software, which averaged roughly +7% over the same 30 days, Adobe led and Figma matched. Whatever is happening to the stock-media libraries, it is not happening to the creative-software vendors.

The setup

Where it stands — Getty and Shutterstock have lost roughly half their value on a failed merger and a forced bond redemption, not on licensing collapse.

Would confirm — Shutterstock full-year revenue landing near the consensus $783.6m, and Getty announcing refinancing terms on the $628.4m notes.

Would invalidate — Getty's next quarter showing licensing revenue down more than 5% year on year, making AI substitution the cause after all.

Watch next — Shutterstock's promised strategic update after its cancelled 6 August call; Adobe's third-quarter results in September.

Valuation — Adobe 15.2x trailing / 10.9x forward earnings; Shutterstock 0.24x trailing sales versus 0.28x forward; Getty 17.6x EV/EBITDA.

Amkor Technology and ASE Technology Holding sit at a physical chokepoint in the AI chip supply chain: they bond stacks of high-bandwidth memory (HBM) to logic processors and test the finished package before it ships to Nvidia, AMD or a hyperscaler's custom-chip designer. Both stocks fell sharply over the past month, but neither drop originated in a weakening AI business. Amkor's AI-exposed segment posted record revenue and ASE's core packaging-and-test unit had its best quarter on record. The declines instead trace to a smartphone-driven guidance miss at Amkor and a sector-wide chip selloff that swept up ASE despite its own strong results.

Amkor: a legacy miss overshadows an AI beat. Amkor, which packages and tests chips for customers including Apple and Nvidia and is building a new advanced-packaging plant in Peoria, Arizona, reported second-quarter revenue of $1.9 billion, beating estimates, with earnings per share of $0.70 against a $0.47 consensus. Its Computing segment — the AI/data-center piece — grew 20% sequentially to a record and was guided up roughly 30% more in the third quarter, driven by a data-center CPU packaging ramp. But Amkor's largest segment, Communications (smartphone and memory-linked assembly), grew just 6% sequentially and was guided down high-single-digits for the third quarter on declining Android unit volumes and memory supply constraints. Because that segment still dominates the revenue base, third-quarter total-revenue guidance of $1.95-2.05 billion missed the $2.11 billion analysts expected, and the stock fell 24% in a single session even though gross-margin guidance (18.5-19.5%) and EPS guidance ($0.72-$0.82) both beat consensus. Amkor also disclosed a 10-year advanced-packaging agreement with TSMC and a multi-year Nvidia partnership, with its Arizona Phase 1 capacity fully committed and a second Korea facility on track for year-end.

ASE: a record quarter, sold off by sympathy. ASE, the largest outsourced assembly-and-test house outside TSMC and the primary second source hyperscalers use for custom AI chip packaging, reported second-quarter revenue up 34% year-over-year. Its assembly-test-materials segment — 66% of revenue and 94% of operating profit — hit a record TWD 126.1 billion, up 36% year-over-year, with gross margin up 5.4 points and operating margin up 6.2 points. Its leading-edge advanced-packaging line is tracking a few hundred million dollars above its own $3.5 billion full-year target, and management raised 2026 capital spending by $2 billion, directing 70% of new equipment spending toward advanced capacity now running at 80-85% utilization — management called the pricing environment "very friendly." None of that stopped the stock from falling 6.6% on July 24 amid a broader "chip risk reset" with no ASE-specific news attached, part of a week in which chip stocks lost more than $1 trillion in market value on reports that memory maker SK Hynix was slowing HBM capacity expansion and on growing skepticism about AI-infrastructure spending returns.

Fundamentals: CONTRADICTS the selloff, valuation INCONCLUSIVE. For both companies, the businesses that explain this story — Amkor's Computing segment, ASE's leading-edge packaging line — accelerated in the same quarter the stocks fell, which contradicts a reading of the drop as fundamentally driven. Amkor now trades at 22.3x forward earnings, below its own trailing 24.7x and below the 25-30x forward range the desk's own notes flagged for it in May, with EV/EBITDA of 10.6x — cheaper against its recent history after the drawdown, not more expensive. ASE's trailing P/E of 42x looks rich against Amkor, but its own forward multiple data was internally inconsistent in this dataset and is not used here; the business case rests instead on 36% segment growth and expanding margins funded by negative free cash flow during a capex buildout that won't ease until 2027. Independent of either earnings report, TrendForce reported August 5 that TSMC is actively expanding outsourcing of its CoWoS advanced-packaging front-end step to outside assembly houses amid rising Nvidia and custom-chip demand — confirming, after the selloff, that the capacity-overflow dynamic underlying both stocks' AI thesis is still operating.

Technicals. Amkor's crash was a single violent session (July 28, -24%, bottoming near $42.73 on July 29 before recovering to about $55), not a gradual grind; ASE's decline was smaller and smoother, consistent with sector sympathy rather than a company-specific shock, and both had round-tripped roughly a third of the peak-to-trough loss by early August.

The setup

Where it stands — Both stocks are down roughly 20-35% from June/July highs after each posted record AI-segment growth in the same quarter they sold off. Would confirm — Amkor's Communications segment stabilizes or its Computing segment share of revenue keeps rising toward 2027, offsetting the phone-cycle drag. Would invalidate — Amkor's Q4 2026/H1 2027 guidance shows Communications weakness persisting without Computing-segment revenue large enough to offset it. Watch next — Amkor's Q3 2026 print (guided $1.95-2.05B revenue) and ASE's follow-through on its raised $10.5B 2026 capex plan. Valuation — Amkor 22.3x forward vs 24.7x trailing (below its own recent 25-30x anchor); ASE 42.1x trailing, forward multiple data unreliable in this dataset.

Four stocks, four different earnings, one misleading average

Over the past month, a group of companies that make the steel pipe, pumps, valves and cryogenic equipment used to move natural gas from the wellhead to export terminals and gas-fired power plants gained an average of 11.2%. That number implies a steady re-rating of the whole gas-equipment supply chain. It isn't one. The average is built from four wildly different outcomes — one stock fell on its own earnings report — and three of the four gains are entirely attributable to a single day's earnings release, not a gradual trend.

Tenaris, the world's largest maker of OCTG (oil country tubular goods) — the steel casing and tubing that line oil and gas wells, plus line pipe for gas transport — fell 6.7% on August 5-6 after reporting second-quarter net income down 10% year over year, a decline the company attributed mainly to U.S. Section 232 steel tariff costs. Flowserve, which makes pumps, seals and valves for oil and gas, chemical, power and water plants — including cryogenic pumps for liquefied natural gas (LNG) facilities — jumped 7.7% on July 30 after posting bookings up 26% year over year to $1.35 billion against $1.17 billion of revenue, an implied book-to-bill above 1.15x. DNOW, a Houston-based distributor of pipes, valves and fittings (PVF) to refineries, LNG terminals and power plants formed by its late-2025 merger with MRC Global, jumped 8.0% and 7.3% on August 6 and 7 after an earnings beat. Graham Corp, which builds vacuum and heat-transfer equipment mainly for Navy submarines, space programs and energy plants, rose 6.7% around its August 6 report of record revenue and backlog.

The business case is not uniform

Only Flowserve shows a gas-specific acceleration that matches its stock move. Its energy-segment bookings rose 48% year over year, aided by large LNG awards in the Middle East and Canada, and backlog is up 16.9% year over year to $3.34 billion. Even so, Flowserve trimmed its full-year organic sales outlook to roughly a 1% decline while raising earnings guidance — the near-term earnings growth is coming from margin expansion, not the top line, even as bookings point to gas-driven demand arriving later.

Tenaris's OCTG business is the opposite: contracting, not accelerating. U.S. tariffs pushed foreign steel costs up roughly 50% and cut OCTG imports about 35% over the past year, which helped domestic pricing but did not offset the tariff-cost drag on Tenaris's own margins. DNOW's headline growth is almost entirely merger consolidation — Q2 revenue more than doubled to $1.31 billion from $628 million a year earlier because of the MRC Global combination, not organic demand, while operating margin collapsed to roughly breakeven. The company has captured $23 million of first-year merger synergies, 35% above target, but that cost-cutting has not yet shown up as per-share cash generation. Graham Corp's headline growth is a defense and space story, not a gas story: total revenue rose 29%, but the Energy & Process segment — the part of the business actually exposed to gas and refining capital spending — grew only 5%, held back by refinery and petrochemical project pushouts, while defense revenue jumped 40% on submarine and torpedo contracts.

Valuation: cheap, supported, and stretched — all in the same group

Tenaris trades at roughly 14x trailing and forward earnings and 8.6x trailing EV/EBITDA, multiples that have barely moved and sit against minimal forecast revenue growth — the recent stock decline reads as a justified reset for tariff-driven margin pressure, not a re-rating. Flowserve's forward price/earnings ratio of 19.4x sits well below its trailing 27.8x, implying the market expects real earnings growth behind its bookings — a supported advance. DNOW is priced for a recovery that hasn't printed yet: a forward P/E of 48x against a roughly 1% forecast net margin and a trailing free-cash-flow yield of just 1.1%. Graham Corp is the richest of the four by far, at 68x forward earnings and 56x trailing EV/EBITDA, multiples justified by defense and space contracts rather than the gas-equipment story this group is nominally about.

Sector context: upstream capex isn't uniformly falling

The hypothesis that gas-equipment makers are advancing just as Appalachian gas producers retrench doesn't hold up cleanly either. EQT, one of the largest Appalachian gas producers, lowered maintenance capital spending while simultaneously raising its production forecast on efficiency gains, not distress. Expand Energy is holding capital spending near $2.75-2.95 billion for 2026, not cutting it. Separately, 2025-2026 has been the highest period on record for LNG project sanctioning, which is the demand pool Flowserve's LNG bookings are drawing from — though gas earmarked for data centers is still described as "playing second fiddle" to LNG exports for Gulf Coast supply.

The tape versus the business

Three of the four stocks' 30-day gains are single earnings-day pops, not gradual re-ratings; DNOW's own trend signal flipped from a bearish to a bullish reading in the space of days around its report, consistent with a sharp reversal rather than a sustained advance. That technical pattern agrees with the fundamentals only for Flowserve, where bookings and backlog data support the move. For Tenaris, DNOW and Graham Corp, the stock reaction and the underlying business trend point in different directions or rest on different drivers than natural gas.

The setup

Where it stands — Four gas-equipment stocks rose together on four separate earnings events; only Flowserve's gains are backed by gas-specific bookings growth. Would confirm — Flowserve's book-to-bill holding above 1.0x and Energy segment bookings still growing double digits in its next quarterly report. Would invalidate — DNOW's operating margin failing to climb meaningfully above breakeven, or Graham Corp's Energy & Process segment staying near flat, in their next reports. Watch next — Flowserve's Q3 2026 earnings release, expected late October 2026, for backlog conversion into revenue. Valuation — FLS forward P/E 19.4x vs. trailing 27.8x; TS 14x trailing/forward; DNOW 48x forward on <1% margins; GHM 68x forward, richest in the group.

What happened

The group of companies that sell corporate technology work by the hour — consulting, software development, IT outsourcing — spent the past year being priced as the industry most exposed to AI writing its own code and closing its own tickets. Over the last 30 days, all eight of the sector's most-followed names reversed sharply, gaining between 9% and 31%, clawing back part of a 12-month slide that ran as deep as -53% for one name. The rebound is broad, not a one- or two-stock pop — but the fundamentals behind it are not uniform, and reading the earnings calls alongside the price charts shows the group splitting into companies whose demand is actually recovering and companies whose stocks are simply re-rating ahead of the numbers.

Accenture (ACN), the largest consulting and technology-services firm globally, staffing everything from cloud migrations to government modernization projects, rose 23.2% in 30 days but remains locked in an unbroken downtrend — its 50-day average has sat below its 200-day since March 6, 154 sessions with no upgrade. Infosys (INFY), an Indian IT-outsourcing giant that staffs offshore technology work for global corporations, is in the same boat: up 11.1% but still in that same deep downtrend as of August 6. Cognizant (CTSH), an outsourcer known for staffing bank and insurer technology projects, only broke its 94-day downtrend on August 7, the day this rebound is being written up — even after a 31.2% bounce, the largest in the cohort. By contrast, EPAM Systems (EPAM), a custom software-engineering consultancy; Grid Dynamics (GDYN), a small AI-engineering and analytics shop; CGI Inc. (GIB), a Montreal-based IT and government-outsourcing firm; Globant (GLOB), a digital-design and software consultancy; and Wipro (WIT), another Indian IT-outsourcing firm, all flipped from deep downtrends to neutral in the July 27–28 window — meaning the three largest incumbents are lagging the smaller names on the tape even as all eight rally together in price.

The businesses tell three different stories

On fundamentals, the picture CONFIRMS the rally for three names. Cognizant has delivered six straight quarters of margin expansion, raised full-year EPS guidance to 8–10% growth, and grew financial-services bookings 12% for a second straight quarter, with trailing bookings up 5% to roughly $29 billion. Grid Dynamics' AI-related revenue is now 30.7% of the total, up 55% year over year, while headcount actually fell 3% as revenue rose 7% — a genuine productivity tailwind, not a pricing cut, according to management. CGI carries a $31.8 billion contracted backlog, 1.9 times annual revenue, at 108–115% book-to-bill, and management says it sees no discretionary spending pullback despite what peers describe — plausible given CGI's public-sector weighting, including a new $251 million U.S. federal contract.

The picture CONTRADICTS the rally for three others. EPAM's non-GAAP earnings per share grew 22% on just 4.5% revenue growth — margin and buyback arithmetic, not demand — and it cut full-year revenue guidance to 3.2–4.2% the same day its stock jumped, with management saying its North America turnaround is slipping into 2027, a point flagged by Morgan Stanley's price-target cut. Infosys cut fiscal-2027 growth guidance to 1.5–3% (organic growth closer to 0.5–0.6%) and confirmed ongoing price deflation on large renewal deals. Globant is the starkest case: it has not reported earnings in over 45 days, yet rallied 17.8% on no new information, sitting on several quarters of flat-to-negative revenue growth. Wipro is the weak middle ground — flat revenue, margins down 120 basis points, no growth timeline given — and, fittingly, the smallest mover in the cohort.

A sector-wide headwind cuts across all three Asia-anchored names: a new $100,000 fee on H-1B visa petitions filed from outside the U.S. is expected to push more delivery offshore over time, a cost noted to have briefly hit TCS, Infosys and Wipro shares and flagged as a structural exposure for Cognizant and Infosys. Accenture, meanwhile, is absorbing a federal-contract review that has already led to at least 10 terminated government contracts, a headwind it calls a roughly 1-point drag on growth as reported by The Daily Upside — and it was Accenture's June guidance cut that first dragged down the entire Indian IT sector.

Valuation: mostly still cheap, one exception

Forward multiples sit well below trailing multiples across most of the cohort — Cognizant 10.0x forward versus 12.4x trailing, CGI 8.2x versus 12.9x, Globant 6.0x versus 15.0x — implying the market still expects earnings to recover, which CONFIRMS there is room left even after the bounce. Infosys is the exception: its forward P/E of 15.8x sits above its trailing 15.1x, meaning the market is not pricing improvement, which CONTRADICTS the case for further upside from here without a guidance reversal.

The setup

Where it stands — Cognizant, CGI and Grid Dynamics show bookings and margins genuinely improving; EPAM, Infosys and Globant rallied around guidance cuts or a reporting gap. Would confirm — Accenture's federal drag anniversaries in fiscal Q4 as guided, or Infosys's constant-currency organic growth moves back above 1%. Would invalidate — Cognizant's book-to-bill or margin guidance rolls over in its next print, or Globant reports and organic growth is still negative. Watch next — Accenture's fiscal Q4 2026 earnings (period ends August 31, 2026), the first test of its stated federal-drag anniversary. Valuation — Cohort forward P/Es of 6x–15x sit near or below each company's own 5–10 year range, except Infosys, whose forward multiple exceeds its trailing one.

An eight-company group of software makers whose products are used to write code, track projects and design products — including Atlassian, GitLab, Figma, Adobe, Autodesk, PTC, and chip-design software leaders Cadence and Synopsys — has risen roughly 18% over the past 30 days, on the surface suggesting the market's year-long bet that generative AI would erode demand for these tools is reversing. It hasn't reversed evenly. Strip out one company's single trading day and the group's average gain roughly halves, and the two businesses best positioned to benefit from AI — not be hurt by it — actually declined over the same month.

One earnings day did most of the work. Atlassian, which sells the Jira and Confluence project-tracking software used by engineering and business teams, jumped about 37% on August 6 after reporting fiscal fourth-quarter revenue up 28% year-over-year to $1.8 billion, cloud revenue up 31% to $1.2 billion, and remaining performance obligations — contracted future revenue — up 44% to $4.8 billion. Management said the beat came from cross-selling and seat expansion, not migration accounting, and disclosed that customers using its Rovo AI assistant grew annual recurring revenue more than twice as fast as non-adopters, with Rovo-assisted actions up 50% quarter-over-quarter and adoption above 80% of the Fortune 500. That is a genuine beat-and-raise, though fiscal 2027 subscription-revenue guidance of 18% marks a deceleration from 23% in fiscal 2026. Remove Atlassian's outsized move from the group average and the remaining seven names' 30-day gain falls from roughly 18% to about 11% — real, but far less dramatic than the headline number implies.

The AI beneficiaries were a drag, not a source of strength. Cadence and Synopsys, the two dominant electronic design automation vendors whose software chipmakers use to design semiconductors, both beat second-quarter estimates and raised guidance — Cadence on 24% revenue growth and a record $8.1 billion backlog, Synopsys on 42% growth following its Ansys acquisition. Yet both stocks fell over the past 30 days (Cadence -10.1%, Synopsys -5.5%) after a new U.S. rule took effect July 1 restricting cloud-based access to their design software for customers in China, a regulatory headwind unrelated to the AI-disruption thesis. Cadence trades at 41.3x forward earnings and Synopsys at 27.8x — both still elevated versus the group, meaning their fundamentals confirm demand strength even as export policy, not AI substitution, drives the tape.

GitLab and Figma sit on opposite sides of the same divergence. GitLab, which sells a unified software-development platform including its Duo AI assistant, has seen quarterly revenue growth decelerate for four straight quarters (29% to 23%), cut full-year guidance from 26% to 16-17%, and disclosed that AI-driven seat contraction is affecting a fifth of its accounts — yet its stock rallied on an unconfirmed takeover rumor involving Datadog and Morgan Stanley, pushing its forward price-to-sales multiple from 4x to 5.4x on a guidance cut. That is the tape confirming while the business contradicts. Figma, the browser-based design platform whose $10,000-plus customers are now heavy AI users (roughly 80% consuming AI credits weekly), reported the opposite problem: second-quarter revenue grew 48%, its third straight quarter of acceleration, net dollar retention hit 136%, and management raised full-year guidance — yet the stock fell about 15% the next session, weighed down by an approaching 77.7-million-share lock-up expiration that lets early investors sell roughly $6-7 billion of stock. That supply overhang, not the numbers, is what moved the price, leaving Figma still 69-87% below its August 2025 high — the business CONFIRMS a trough, the price action does not.

Adobe, Autodesk and PTC look like the group's steadier core. Adobe, the Creative Cloud and Firefly maker facing generative-image competition, trades at a historically depressed 10.8x forward earnings despite growth mildly accelerating to 12.7% and mid-30s operating margins holding. Autodesk, which sells AutoCAD and Fusion 360 to architecture and manufacturing customers, is growing 18.4% with margins expanding and trades at 19.7x forward earnings versus 35.9x trailing. PTC, whose Creo and Windchill software serve industrial engineers, beat on annual-recurring-revenue growth (+9.1%), raised guidance, and repurchased $525 million of stock in one quarter — more than double its target, citing a "compressed" share price — while trading at just 18x forward earnings, below Autodesk's multiple despite comparable growth.

On the tape, none of the eight names appears in violent-mover territory; the moves are gradual by the classification used here, with Atlassian's band having drifted to a mildly bullish reading three sessions before its earnings jump and PTC only flipping out of a bearish trend the day after. That alignment between price and disclosed fundamentals holds for Atlassian, Cadence, Synopsys, Autodesk and PTC. It breaks down for GitLab, where the tape is ahead of the business, and for Figma, where the tape is behind it.

The setup

Where it stands — Atlassian's beat drives most of an 18% group gain; GitLab's rally outruns cut guidance while Figma's beat is buried under lock-up supply. Would confirm — GitLab's next-quarter revenue growth stabilizes above 20% without a completed takeover, or Figma's stock recovers once the August 31 lock-up clears. Would invalidate — GitLab's guidance is cut again, or Figma's net dollar retention (currently 136%) declines sequentially next quarter. Watch next — Figma's 77.7-million-share lock-up expiration on August 31, 2026, and GitLab's next quarterly print for confirmed M&A or renewed deceleration. Valuation — GitLab: 5.81x forward sales vs 6.45x trailing, expensive against cut guidance; PTC: 18x forward P/E vs Autodesk's 19.7x despite similar growth.

Eight companies that build the software running behind payments and banking — merchant card processors, core banking systems for community banks, and back-office infrastructure for brokerages — have risen an average of 10.6% over the past month, clawing back part of a brutal year in which the group fell 17% on average. But the rebound is not a broad re-rating of the sector. It is concentrated in one name.

A rally led by one earnings beat

Paymentus (PAY), a cloud billing-and-payment platform that lets utilities, insurers and government agencies collect payments online — not to be confused with Paysafe (PSFE), a separate, similarly named payments company — is up 43.2% in 30 days and 50.6% over 90 days after reporting Q2 2026 revenue up 28.8% year-over-year and raising full-year guidance. Strip Paymentus out and the remaining seven names — Broadridge (BR), Fiserv (FISV), FIS (FIS), Flywire (FLYW), Jack Henry & Associates (JKHY), NCR Voyix (VYX) and Waystar (WAY) — average a 5.9% gain over the same month, a far more modest bounce than the headline number implies. Paymentus's own operating margin expanded from 5.7% to 9.0% year-over-year, and net income grew 74%, consistent with a genuine growth inflection rather than a distressed name simply stopping its fall — the business explains this stock's move (CONFIRMS), even if its 41.8x forward earnings multiple has already re-rated to reflect it.

Fiserv's guidance cut is still unfolding

Fiserv, which runs the Clover payment terminals and processing rails used by millions of small merchants, cut its full-year 2026 organic revenue guidance for the second consecutive quarter on August 6, to a range of flat to down 1%, from a prior forecast of 1%-3% growth. Quarterly revenue actually turned negative and operating margin fell from about 27% to 17% over the past year — real, ongoing deterioration, not merely a tough comparison being lapped. Clover, the company's merchant-payments platform, still grew transaction volume 9%, but currency swings in Argentina and delayed client rollouts overwhelmed that growth. Fiserv's stock is up 6.9% in 30 days even as the business kept decelerating — the tape and the fundamentals CONTRADICT each other here, and Fiserv's 6.7x forward earnings multiple looks cheap mainly because the earnings base was just cut, not because confidence is returning.

FIS's headline cheapness is an accounting mirage

FIS, which sells core processing software banks use to run checking accounts and payments, swapped its stake in payment processor Worldpay for Global Payments' card-issuing software unit plus $7.7 billion cash, a deal that closed in January. The transaction produced a one-time accounting gain that pushed net margin to 72% last quarter, distorting FIS's trailing price-to-earnings ratio down to 8.6x and enterprise-value-to-EBITDA to 4.65x. Forward revenue estimates jump roughly 30% for 2026, but that reflects folding in the acquired business, not organic acceleration — the low multiples are not a clean signal of undervaluation (INCONCLUSIVE).

The defensive anchors: Broadridge and Jack Henry

Broadridge, which processes proxy votes and post-trade paperwork for banks and asset managers, has posted mid-to-high single-digit revenue growth for four straight quarters and has 93% of its full-year proxy business already locked in, yet its shares still fell 38% over the past year and dropped further on its latest earnings print even as recurring revenue grew 6% and adjusted earnings per share grew 11%. That gap between a stable, contracted business and a battered stock is the clearest dislocation in this group (business CONTRADICTS the prior sell-off). Jack Henry, which supplies core operating software to community banks and credit unions, has accelerated revenue growth for three straight quarters, with its forward P/E of 21.6x sitting almost exactly at its trailing multiple — the market is pricing continuation, not a turn either way (CONFIRMS current trend).

Growth stories misread as turnarounds

Flywire, which processes cross-border tuition, medical and travel payments, grew revenue 41% last quarter, its fastest pace in over a year, and trades at 37.7x forward earnings on that growth. Waystar, whose software helps hospitals bill insurers and collect from patients, grew subscription revenue 34% and raised full-year guidance after its July 29 earnings call, with net revenue retention of 108% and leverage of a manageable 2.5 times earnings before interest, taxes, depreciation and amortization. Neither fits a profile of a levered, structurally challenged business merely bottoming — both are re-rating on real, accelerating fundamentals (CONTRADICTS the "stopping falling" framing).

NCR Voyix: the actual value-trap risk

NCR Voyix, which sells point-of-sale and self-checkout systems to retailers and restaurants after selling its digital-banking unit for $2.45 billion in 2024, still shows revenue declining, with consensus estimates projecting a 17% drop for 2026 and trailing free-cash-flow yield deeply negative at -31%. Its forward P/E of 9.5x versus a 22x trailing multiple implies the market expects a sharp earnings recovery that the current numbers don't yet support (CONTRADICTS).

The tape versus the story

By early August, six of the eight names had climbed out of a steep multi-month downtrend into neutral territory; only Broadridge remained in a mild decline. Twelve-month returns still ranged from Flywire's +47.6% to Fiserv's -59.2%, underscoring how differently the market has treated names inside a group that looks similar on paper. A broader industry risk sits in the background: a July consortium of more than 140 firms, including Visa and Mastercard, launched a stablecoin payment network called Open USD that could eventually route transactions around traditional card rails — Fiserv and FIS have each launched their own stablecoin initiatives in response, but neither company's current numbers show volume actually migrating away yet.

The setup

Where it stands — The rebound is real but concentrated in Paymentus; the rest of the group splits between a decelerating Fiserv and steady, undervalued anchors in Broadridge and Jack Henry. Would confirm — Fiserv's Merchant Solutions organic revenue returns positive in its next quarterly report without a third guidance cut. Would invalidate — Broadridge's recurring-revenue growth slips below its recent 6% pace or NCR Voyix's revenue decline widens beyond the projected 17% for 2026. Watch next — Fiserv's next quarterly earnings report, due within the current quarter, for whether the guidance cut holds. Valuation — Fiserv trades at 6.7x forward earnings versus a historically higher trailing multiple; Broadridge trades at 15.1x forward versus 16.7x trailing, near the low end of its own recent range.

The financial firms that actually write the checks behind the AI data-center boom — private-equity shops structuring loans against Nvidia chips, banks arranging multibillion-dollar leases, and infrastructure funds building the power and fiber underneath — just finished reporting second-quarter earnings, and the numbers largely back up this month's rally rather than explaining it away as chart noise. Six of the group's largest names posted record or near-record fee income and fundraising in late July and early August, with named, dated financing deals tied directly to the AI buildout. But one member, Brookfield Corporation, moved the opposite direction on the chart even as its own asset-management subsidiary reported a record quarter — a split that looks more like a timing gap than a credit problem.

The private-credit and buyout core. Blackstone (BX), the world's largest alternative-asset manager, reported $1.8 billion of fee-related earnings (FRE) — the recurring management fees that anchor its valuation — up 22% year over year, on record assets under management (AUM) of $1.35 trillion. Its infrastructure arm grew AUM 40% to $90 billion, and its data-center platform, built around the QTS operator it owns, was marked at $185 billion, up from $130 billion at the start of the year; in July it also detailed a $35 billion financing platform with Broadcom for a gigawatt of AI compute. KKR (KKR) posted FRE of $1.32 per share, up 34%, with a 70% margin for a tenth straight quarter above 65%, and raised a record $133 billion over the trailing year; in June it launched Helix, a $10 billion-plus AI-infrastructure vehicle backed by Nvidia and power producer Vistra, though management flagged that financing spreads on the largest hyperscaler deals have begun widening. Apollo Global Management (APO), which pairs private credit with an insurance balance sheet, grew FRE 25% to $785 million on record quarterly origination of roughly $74 billion, and is one of two lenders, with Blackstone, behind a $35 billion loan to buy custom AI chips for Anthropic and lease them back — the largest private-credit financing ever completed. Ares Management (ARES) grew FRE 20% to $491 million on record fundraising of $36 billion, with direct-lending non-accruals still under 2%, while building Ada Infrastructure, a seven-campus, roughly one-gigawatt data-center platform. Carlyle Group (CG) reported record FRE of $358 million, up 11%, with capital-markets fees more than doubling to $111 million.

The banks. Goldman Sachs (GS) and Morgan Stanley (MS) both posted record revenue, $20.3 billion and $21.3 billion, with Goldman's equities revenue up 86% on Asia financing activity and Morgan Stanley's investment-banking revenue up 58%. Morgan Stanley raised its 2026 data-center capex forecast to $850 billion from $575 billion and 2027 to $1.3 trillion from $700 billion, positioning itself as an intermediary rather than a direct lender. Goldman flagged its supplementary leverage ratio, a regulatory capital limit, as the binding constraint on how much more financing it can carry.

The yield and infrastructure names. HA Sustainable Infrastructure Capital (HASI), a specialty lender to renewable and efficiency projects, and Brookfield Infrastructure Partners (BIP), which owns toll-like assets including data centers and fiber, sit at the more bond-like end of the group. BIP's data-segment funds from operations grew 36% on a bulk fiber network and an Intel Arizona chip-plant partnership, and it spun off a U.S. data-center unit in an IPO that raised roughly $1.2 billion while retaining 64% ownership.

The Brookfield divergence. Brookfield Corporation (BN), the holding company atop Brookfield's fund complex, has broken from a mild uptrend into a strong downtrend across every horizon out to a year, and its renewable arms, Brookfield Renewable Corp (BEPC) and Brookfield Renewable Partners (BEP), have stepped down from strong uptrends over the same period. That looks like sequencing rather than deterioration: BN does not report second-quarter results until August 13, after Morgan Stanley cut its price target to $59 from $61 in July citing real-estate-heavy exposure, while every other name in the group had already reported and re-rated. Brookfield Asset Management (BAM), BN's own fee-generating subsidiary, grew FRE 20% to $808 million with record $77 billion fundraising in the same window, including a new $10 billion AI infrastructure fund — evidence against the idea that Brookfield's AI-financing book itself is weakening.

None of this is a rate-cut trade. The Federal Reserve held its benchmark rate at 3.50%-3.75% at its July 29 meeting, and futures on August 4 priced better-than-60% odds of a hike, not a cut, at the September meeting.

What's unresolved. Moody's has flagged roughly $662 billion of hyperscaler data-center lease commitments signed but not yet begun paying on. Apollo's own economists note that $5 trillion of planned AI capex would require $1.5-2 trillion of annual AI revenue by 2030, against only $40-60 billion generated today. And GPU-backed loans such as CoreWeave's facility reportedly lean on informal residual-value support from Nvidia rather than independently appraised collateral — an assumption untested by a downturn. None of this shows up yet in reported credit metrics, but it's the risk this group is being paid to carry.

On valuation, most alt managers show large gaps between trailing and forward multiples — KKR 30.8x trailing versus 16.6x forward, Carlyle 48.3x versus 13.6x, Ares 60.3x versus 23.5x — implying consensus expects current fee growth to hold rather than a fully priced move. Goldman and Morgan Stanley show little compression (14.8x and 16.6x forward) because both already re-rated over the past year, up 42% and 51%. Brookfield Infrastructure remains the least compressed, at 35.9x forward funds from operations.

On the tape itself, the group's one-month move is not evenly shared: 30-day returns range from -8.9% in Brookfield Renewable Corp to +13.4% in Ares, averaging closer to 3-4%. The alt managers and banks are rallying despite a rough trailing year for several (Ares -27%, KKR -28.5%, Carlyle -22.3%), a stretch Goldman, Morgan Stanley and HASI outran (+42.5%, +51.3%, +56.5%). Only Brookfield's holding company and its renewable arms are breaking down, on earnings that haven't printed yet.

The setup

Where it stands — Alt-managers and banks rallied on record fee income and named AI-financing deals; Brookfield's parent diverged into a downtrend ahead of its own earnings. Would confirm — BN's August 13 results show fee-related and distributable-earnings growth near BAM's 15-20% pace, not a real-estate-driven NAV markdown. Would invalidate — BN discloses leverage or NAV deterioration tied specifically to its data-center or AI-financing exposure, not just real estate or sector rotation. Watch next — Brookfield Corporation's Q2 2026 earnings call, August 13, 2026. Valuation — Alt managers trade near 15-24x forward earnings versus 30-60x trailing; BN's forward P/E is 16x versus an 83x trailing figure distorted by thin GAAP margins.

Seven semiconductor companies that supply the timing, radio-frequency and optical components used to move signals inside and between artificial-intelligence datacenter racks reported earnings over the past two weeks, and the results split the group in two: four posted genuine beat-and-raise quarters tied to datacenter demand, while two of the seven diverged for reasons that have nothing to do with datacenters. That matters because a simple twelve-month average return for the group — roughly +187% — hides that the gain was carried almost entirely by four names, and the past month's average pullback of about 4% conceals a spread from +12% to -22% across members.

This is a story split by business model, not a uniform sector move. SiTime (SITM), which makes silicon oscillators and clock chips that synchronize signals inside servers and networking gear, jumped 26.6% on August 6 after second-quarter revenue of $157.4 million, up 127% year over year, with non-GAAP gross margin rising 890 basis points to 67.1%. Its communications-and-datacenter segment grew 181% year over year to $101 million — a ninth straight quarter of triple-digit growth, on clock adoption inside 1.6-terabit optical modules — and third-quarter guidance of $285-295 million came in well above the $219.7 million Wall Street expected. CONFIRMS: the pop tracks disclosed, dated numbers.

MACOM (MTSI), which designs radio-frequency-to-optical chips for telecom, defense and datacenter networking equipment, rose 14.5% the same day, extending a rally begun by its own August 4 report — revenue of $342.2 million (+35.8% year over year), gross margin of 58.3%, and operating income more than doubling to $77.1 million. CONFIRMS on the business, but forward earnings multiples near 60x are well above the roughly 30x level this desk's own prior research flagged as already pricing an optimistic optical-networking ramp.

MaxLinear (MXL), which makes RF and mixed-signal chips for broadband and optical datacenter links, returned to GAAP profitability in its July 23 report, with revenue up 55% to $168.8 million and its now-largest infrastructure segment up 145% to about $85 million on datacenter demand; management raised its 2026 optical-datacenter revenue guide to $210-230 million. Yet the stock is down 18.4% over the trailing month even as it remains up 362% over the past year — the business accelerated while the shares digested a year of gains, a positive divergence rather than a warning sign.

Applied Optoelectronics (AAOI), which manufactures the optical transceivers and lasers that carry data between servers, posted revenue of $191.9 million (+86% year over year), returned to non-GAAP profitability, and raised its full-year guide to more than $1.1 billion from $455.7 million in 2025, backed by a disclosed hyperscaler order for more than $200 million of 1.6-terabit transceivers and a follow-on $53 million 800-gigabit order. But gross margin has compressed for three straight quarters, from 31.2% to 27.7%, the company remains loss-making at the operating line, and diluted share count is up 44% year over year. Shares fell 3.4% the day after the beat, but the stock's current downtrend reading largely reflects a June-July collapse from $171 to $76 that preceded the print. Industry data support the demand side of the story: 800-gigabit shipments are set to more than double in 2026 as 1.6-terabit modules enter volume production, arguing against a broader hyperscaler order pushout. MIXED: demand confirms, margin and dilution do not.

Semtech (SMTC), which makes analog chips including optical signal-integrity components for datacenters, rose 11.2% alongside the group, but its own next earnings report isn't due until August 25 — this move is sympathy with SiTime and MACOM, not confirmed by fresh Semtech numbers. INCONCLUSIVE.

Qualcomm (QCOM), the mobile-chip and patent-licensing giant that also sells automotive and datacenter processors, fell 14% over the past month after disclosing that Apple's iPhone-modem share for the coming launch is running well below the roughly 20% previously assumed, implying a roughly 50% sequential drop in handset revenue into the December quarter. Automotive revenue hit a record $1.6 billion (+61%), and management raised its 2029 non-handset target to $40 billion, including $15 billion of new datacenter custom-silicon business — years out. CONTRADICTS the datacenter-AI framing: Qualcomm's near-term move is a handset story, not a chip-demand one, and at 15.1x forward earnings and 3.9x forward sales it is priced almost entirely for that risk.

indie Semiconductor (INDI), which supplies chips and software for automotive driver-assistance and connectivity systems, posted essentially flat revenue near $217 million versus a year ago with a negative gross-margin quarter, and fell 21.5% over the trailing month, the worst in the group. CONTRADICTS.

On the dashboard's daily trend measure, Applied Optoelectronics and indie Semiconductor both flipped from a mild uptrend to a mild downtrend in early August, while SiTime, MACOM, MaxLinear, Qualcomm and Semtech remain in a mild uptrend after stepping down one notch from a strong uptrend in July. None of the seven appears among the market's most extreme movers over the past month — this reads as rotation among names with different stories, not a sector-wide break.

The setup

Where it stands — Four datacenter-facing suppliers (SITM, MTSI, MXL, AAOI) beat and raised; QCOM and INDI diverge on unrelated handset and auto-chip weakness. Would confirm — Semtech's August 25 report shows datacenter/optical revenue growth and gross-margin expansion in line with SiTime and MACOM. Would invalidate — AAOI's gross margin resumes its decline below 27% next quarter, or MaxLinear's optical-datacenter revenue guide is missed. Watch next — Semtech's fiscal Q2 earnings report, scheduled for August 25, 2026. Valuation — MACOM trades at 60x forward earnings versus its own ~30x prior anchor; Qualcomm at 15.1x forward earnings, cheapest in the group.