DK Street Journal

What happened

Over the past month, the companies that make the physical guts of the AI buildout — memory chips, fiber-optic cable, semiconductor substrates and the contract manufacturers that assemble AI servers — lost 10% to 30% of their market value. In the same stretch, the IT-consulting firms and identity-security software makers that had spent the past year being priced as AI's casualties jumped 10% to more than 40%. The trigger was not gradual: a single event, an AI-chip selloff in the week of July 27-29 that wiped more than $1 trillion off chip stocks after Meta signaled excess data-center compute capacity and Alphabet raised 2026 capital-spending guidance to $195-205 billion while posting negative quarterly free cash flow, reviving fears that hyperscalers are overbuilding AI infrastructure. The same week, Jefferies upgraded Indian IT-services stocks from underweight to neutral, citing reduced worry about the AI trade and explicitly framing the move as money rotating out of chip names.

The hardware side: guidance up, stocks down

Micron (MU), the memory-chip maker whose DRAM and high-bandwidth memory go into AI servers, posted revenue up 196% year over year with its entire fiscal-2026 HBM capacity sold out under non-cancellable contracts — yet its stock fell roughly 16-20% in the same window. Corning (GLW), which makes the fiber-optic glass and cable used in AI data-center networks, fell about 30%, but most of that came in a single-day drop after a guidance miss in its own earnings print, not a cohort-wide rotation; it still trades at roughly 52 times forward earnings. Celestica, Jabil and Sanmina, the contract manufacturers that assemble AI server racks, all beat and raised guidance in July even as every name in that group saw its trend band cut. MaxLinear (MXL), a networking-chip maker, LightPath Technologies (LPTH), an optical-components maker, and POET Technologies (POET), a pre-revenue developer of optical chips with just $503,000 in quarterly revenue and negative gross margin, all lost their bullish trend status in the same three-day window — though LPTH and POET were only modestly bullish beforehand, not the strong uptrends the price action implies. AXT Inc (AXTI), which makes semiconductor substrate wafers, was downgraded on the charts even though its stock actually rose 6.7% over the 30 days — a reminder that cohort averages here are pulled by a handful of names, not moved uniformly.

The services side: stocks up, guidance mixed

EPAM Systems (EPAM) and Globant (GLOB), both software-engineering outsourcing firms, jumped roughly 13-20% in July and remain statistically inexpensive even after the bounce — 15.2 and 14.9 times trailing earnings, versus richer peers like Grid Dynamics (GDYN), a smaller AI-consulting firm trading near 230 times trailing earnings on a thin profit base. But EPAM, Globant, Infosys and Cognizant all cut or lowered guidance in their latest prints, and Wipro (WIT), an Indian IT-outsourcing firm, rallied on sector sentiment rather than its own numbers. Okta (OKTA), an identity and access management software maker, is a different story: it never fell out of favor and now trades near 100 times trailing earnings with growth decelerating to 9-11% and a consensus price target sitting about 13% below the current price. SailPoint (SAIL), a rival identity-security firm, only crossed into bullish territory on July 30-31 — two days after this desk's own prior analysis found it had briefly touched bull territory in June before reverting, meaning the current flip is unconfirmed, not an established trend.

Sector versus stock, and the verdict

The hardware decline is genuinely sector-wide — memory, RF, EMS and optical names moved together on one catalyst — but it is multiple compression after a run that saw some names gain 81% to 456% over the trailing year, not a demand slowdown; order books and guidance at Micron, Celestica, Jabil, Sanmina, Texas Instruments, Analog Devices and Microchip all improved in the same window their stocks fell. That is a business/price divergence that CONTRADICTS the bearish framing of the hardware selloff. The services rally is real and broad but not yet backed by bookings: Accenture's bookings fell, and several consultancies cut guidance the same week they rallied, making this a valuation catch-up rather than a confirmed re-acceleration — INCONCLUSIVE on fundamentals, CONFIRMS only on price momentum. Both hardware and services multiples remain elevated relative to their own history in several names (GLW at 52x forward, OKTA at 100x trailing), meaning the give-back on one side and the bounce on the other have not obviously restored value on either.

The setup

Where it stands — Hardware names fell 10-30% in July on a sector-wide AI-capex scare despite raised guidance; services/identity names rose 10-40% despite mixed bookings. Would confirm — Hyperscaler Q3 capex guidance holds or rises and Micron/Celestica book-to-bill stays above 1.0 into September. Would invalidate — Services cRPO or net-new ARR at EPAM, Globant or Okta shrinks in the next reported quarter despite the price rally. Watch next — Next hyperscaler earnings cycle (late October 2026) and EPAM/Globant Q3 reports for bookings confirmation. Valuation — GLW ~52x forward P/E vs its own multi-year range; OKTA ~36x forward EPS with PT 13% below spot; EPAM/GLOB near 15x trailing, below sector norm.

The re-rating already happened

A popular narrative holds that cybersecurity stocks are the one software corner that missed 2025-26's AI-driven rally and are only now stirring. The price record says otherwise. Over the trailing twelve months, Fortinet, which makes network firewalls, gained 66%; Palo Alto Networks, the largest cybersecurity software platform by revenue, gained 92%; Okta, which sells identity and login-security software, gained 49%; and Cloudflare, which operates a content-delivery and "zero trust" internet-security network, gained 39%. Five of the group's seven members — those four plus Rubrik, a recently public cloud data-backup and ransomware-recovery vendor — currently sit in confirmed uptrends, with 50-day moving averages holding above 200-day averages. Only CrowdStrike, the cloud-based endpoint-security vendor, and Zscaler, which sells cloud-delivered internet-security gateways, have rolled over in the past several weeks. This is a cohort that already re-rated hard, not one waiting to start.

What's driving it, name by name

The growth behind these moves is mostly genuine. Palo Alto Networks' next-generation-security annual recurring revenue (ARR) — its cloud and AI-security subscription line — rose 60% year over year to $8.1 billion in its fiscal third quarter, and total revenue grew 31%, according to its own earnings release. That figure includes $1.6 billion pulled in from its recently closed $25 billion acquisition of identity-security vendor CyberArk, a deal that added roughly 13.5% share dilution and pushed net debt to about 2x EBITDA from near zero, prompting KeyBanc to cut the stock to Sector Weight on leverage concerns even as the growth story holds. Fortinet's product revenue jumped 52% and billings rose 33% in its most recent quarter, evidence of a genuine hardware refresh cycle as customers upgrade firewalls, per its Q2 2026 results. CrowdStrike posted a record $256 million in net-new ARR, up 32%, as customer-concession packages tied to its 2024 global-outage incident finally rolled off, per its Q1 FY2027 release. Rubrik grew subscription ARR 34% to $1.46 billion with net revenue retention near 120%, while stock-based compensation fell from $914 million to $329 million and free cash flow turned positive, per its fourth-quarter results — the cleanest confirmation in the group that price and business are moving together.

Okta is the exception that contradicts the group's pattern: growth decelerated from roughly 13% to 9-10% year over year even as the stock climbed 49-79% over recent months, and net revenue retention has been flat near 107%. The stock's rise appears to be pricing an AI-driven identity-security re-acceleration that hasn't shown up in the numbers yet. Cloudflare's 30-34% revenue growth is real and its "zero trust" security bundle is a plausible AI-traffic beneficiary — more autonomous software agents means more network traffic to secure — but at roughly 40 times sales, it now carries the richest multiple in the group. Zscaler is a different story altogether: its 31% single-day crash in May came after management cut fiscal 2027 growth guidance to 16-17% from a roughly 20% Street expectation and disclosed sales-leadership turnover, a real fundamental reset rather than tape noise, as reported by CNBC. Its recent stabilization looks like a bounce off a justified bottom, not a fresh re-rating.

Valuation: from cheap to stretched in three months

Multiples have followed the price moves. Palo Alto Networks' price-to-sales ratio rose from 14.8x to 24.5x since May; Fortinet's trailing price-to-earnings jumped from 35x to 59x; CrowdStrike remains the group's most expensive on a price-to-sales basis at roughly 36x, with no positive GAAP earnings yet to compute a P/E. Against that, Rubrik trades at 10.3x sales on 35-39% growth, and Zscaler — post-crash — trades at 7.7x sales, the cheapest in the group. A long-flagged structural risk sits underneath all seven: Microsoft has bundled its Defender, Entra and Purview security tools into a unified detection platform explicitly marketed to reduce reliance on standalone point solutions, per Microsoft's own security-cloud materials, though no measurable seat or price losses at any of these seven vendors have been reported to date.

The setup

Where it stands — Five of seven cybersecurity names sit in confirmed uptrends after 12-month gains of 39%-92%, with fundamentals mostly supporting the moves except at Okta. Would confirm — Okta's next-reported quarter shows revenue growth reaccelerating above 10% and net revenue retention rising above 107%. Would invalidate — Palo Alto Networks' next-generation-security ARR growth decelerates below 50% or net debt/EBITDA exceeds 2.0x. Watch next — Palo Alto Networks and CrowdStrike both report fiscal quarters in late August 2026. Valuation — Rubrik trades at 10.3x forward sales against 35-39% growth, the only cohort member below its recent-history multiple for its growth rate.

What happened

Eight companies that sell IT consulting, software engineering and outsourcing services to large corporations — Accenture (ACN), Cognizant (CTSH), EPAM Systems (EPAM), Grid Dynamics (GDYN), CGI (GIB), Globant (GLOB), Infosys (INFY) and Wipro (WIT) — fell together in mid-June and then rallied together in July, a round trip that shows how a single earnings report can move an entire industry regardless of what each company is actually doing. Accenture, the largest technology-consulting and outsourcing firm in the group with more than 700,000 employees, reported fiscal third-quarter results on June 18 that included a 2% year-over-year drop in new bookings and a 14.7% decline in its order book, and cut its full-year revenue growth guidance to 3-4% from 4-5%. The stock fell roughly 18% in two sessions, and every other name in the group — including Indian outsourcers Infosys and Wipro that had nothing to do with Accenture's report — fell alongside it on fear that artificial intelligence is shrinking the market for human consulting labor.

By late July the group had round-tripped: all eight stocks gained an average of 17.8% over 30 days, matching the +17.6% figure in the original thesis. But the rebound's real driver was not company news. A more than $1 trillion selloff in AI-chip stocks in July rotated capital into IT-services names seen as having no chip exposure, and on July 27 — one day before five of the eight names had their trend status upgraded from a downtrend to neutral — Jefferies upgraded Indian IT services stocks purely on positioning grounds after the sector's roughly 25% fall, not on any change in earnings expectations.

The band upgrades didn't track the earnings, or the bounce

EPAM (a software-engineering firm with large delivery centers in Ukraine and Eastern Europe), Grid Dynamics (a small digital-engineering firm serving retail and technology clients), Globant (a Latin America-based digital-transformation consultancy) and Wipro (an Indian IT-outsourcing firm) all flipped from a persistent downtrend to neutral on the same day, July 28, alongside CGI (a Canadian IT and business-consulting firm known for steady government contracts). But EPAM had not yet reported second-quarter results — its release is scheduled for August 6 — and Globant's results aren't out until August 13, so neither company's own trend upgrade can be explained by its own fundamentals. Grid Dynamics' July 30 report, two days after its upgrade, did show real strength — AI-related revenue crossed 30% of sales, up 54.6% year-over-year, with margins improving — but that print came after, not before, the signal changed. Wipro's own July 16 report showed only 1% revenue growth and cautious guidance, weaker than the trend upgrade implied.

Meanwhile Accenture, Cognizant and Infosys stayed pinned in a persistent downtrend through July 31 even though Infosys beat estimates and held its full-year guidance and Cognizant posted 4.1% constant-currency revenue growth. And the two biggest 30-day price gainers in the whole group were Cognizant (+34.8%) and Accenture (+26.5%) — the very names still flagged as in a downtrend — while Wipro, one of the upgraded names, gained only 5.9%. Trend status and price performance moved in opposite directions.

Fundamentals: cheap across the board, no clear winner

On valuation, there's little to separate the two groups. EPAM trades at roughly 11-15x trailing earnings versus its own five-year average near 30.7x and ten-year average near 41.9x — genuinely cheap by its own history. Accenture, at about 13.2x trailing earnings, sits near half its historical 25-30x premium multiple, even as its bookings and order book keep shrinking. CGI's July 30 quarter was the cohort's steadiest: 2.5% revenue growth, a 108.1% trailing book-to-bill and a backlog 1.9 times annual revenue, consistent with its role as the group's defensive anchor rather than a growth story. Cognizant's own attrition ticked up to 13.0% from 12.3%, an ambiguous demand signal. None of this reads as a fundamentals-driven leadership rotation — it reads as a cheap sector where the trend measure is still catching up to a June crash that was itself driven by one company's report, Accenture's, rather than an industry-wide deterioration.

Verdict: CONTRADICTS the mid-tier-inflection thesis on causality — the trend upgrades preceded, rather than followed, company-specific catalysts for three of five upgraded names. INCONCLUSIVE on valuation — the group is broadly cheap versus its own history, but that cheapness applies to laggards and upgraded names alike, so it does not by itself validate a rotation into the mid-tier.

The setup

Where it stands — Eight consulting and IT-outsourcing stocks rebounded together in July after crashing together in June on Accenture's bookings decline. Would confirm — EPAM's August 6 and Globant's August 13 reports show organic constant-currency growth accelerating from their own May guidance (2.7% and roughly flat, respectively). Would invalidate — Accenture's bookings and order book keep declining next quarter while its trend status stays a downtrend despite the July bounce. Watch next — EPAM reports Q2 2026 results August 6; Globant reports August 13. Valuation — EPAM trades near 11-15x trailing earnings versus a 30.7x five-year average; Accenture near 13.2x versus a historical 25-30x premium.

These ten companies sell software billed per employee seat: payroll systems, IT-ticketing platforms, e-signature tools, project trackers, tax software. Over the past year investors bet that artificial intelligence would let companies do the same work with fewer paid seats, and the group lost roughly a fifth of its combined market value — enterprise software overall shed about $1 trillion in 2026 on that fear, with every one of these ten stocks down between 11% and 65% over the trailing year. In July, all ten rallied together, gaining 4.6% to 20.5% for the month, averaging about 14%. The question is whether that reflects businesses proving the AI-erosion fear wrong, or a crowded short position unwinding. The answer splits three ways.\n\nGrowth is reaccelerating at three names. ServiceNow, which sells software that manages internal IT tickets and workflow automation for large corporations, posted subscription revenue up 24.5% and a renewal-committed backlog (cRPO) up 21% last quarter, while its AI add-on, Now Assist, crossed $1 billion in annual bookings with new AI revenue accelerating more than 40% sequentially — evidence the AI product is additive rather than a seat replacement, per the earnings call. Atlassian, maker of the Jira and Confluence tools software teams use to track projects, guided cloud revenue growth to roughly 25.5% after cloud growth actually accelerated to 29% last quarter on "continued strong seat expansion in Jira" — directly contradicting an earlier report of Atlassian's first-ever enterprise seat decline, per Atlassian's results. Veeva Systems, which sells regulatory and customer-management software to pharmaceutical companies, raised its full-year revenue guidance to $3.635-3.645 billion (16% growth) and continues migrating customers to its newer Vault CRM product with an 80%-plus win rate against Salesforce, according to guidance detail. All three saw their trend bands upgrade out of bear territory in the past two weeks — ServiceNow and Atlassian both flipped from a severe downtrend to a milder one within the last seven trading sessions.\n\nGrowth is decelerating at three more, even as the stock price rises. Workday, which sells human-resources and payroll software billed per employee, saw subscription backlog growth slow to a 13-14% guide for the coming year from roughly 16% previously, and activist investor Elliott Management has taken a $2 billion stake pushing cost cuts and buybacks rather than a growth turnaround, per Elliott's filing coverage — even as the stock gained 18% in the past month. ADP, the payroll and HR-outsourcing giant, guided its cleanest seat-count proxy, pays-per-control, to flat-to-1% growth for the coming fiscal year after posting just 1% growth last quarter, per its earnings call — a genuine plateau in the number that would show whether AI is actually eroding payroll headcount, even as the stock rallied 10% in July and now sits in a strong uptrend. monday.com, a per-seat work-management and project-tracking tool, guided revenue growth down to 18-19% from roughly 24-27% previously, with net retention flat at 110%, per its earnings coverage, despite a 9% one-month price gain.\n\nThe clearest divergence: Intuit. Intuit, maker of QuickBooks accounting software and TurboTax tax-prep software, is the only one of the ten whose trend band never left its severe downtrend during the entire period, even after its stock rose 14.8% in the past month. That's because the underlying business moved the other way: Intuit cut its TurboTax revenue forecast, announced a 17% workforce reduction, and Morningstar downgraded its competitive moat from wide to narrow, citing AI-native bookkeeping tools that can automate up to 98% of routine tax and bookkeeping tasks. Intuit's stock remains down 20.8% over the past three months even after the July bounce — a textbook case of the tape and the business pointing in opposite directions.\n\nThe rest is unsettled. Paycom, a smaller payroll and HR software provider, has no fresh quarterly data — its results are due August 5 — leaving its own seat-count metric untested even though its stock has already rallied into a strong uptrend. Bill.com, which processes business-to-business payments for small companies, posted 16% core revenue growth and its first GAAP profit alongside a 30% headcount cut, per its results, but its trend band was still mild-bear even after an 11.5% one-month gain. DocuSign, maker of e-signature and contract software, posted the group's weakest growth at 9%, offset by a 35% free-cash-flow margin and an expanding contract-management add-on now at 12.6% of revenue.\n\nValuation: cheap on a forward basis, but the recent move is mostly multiple expansion. ServiceNow trades near 22x 2027 consensus earnings versus a historical forward multiple above 40x, and monday.com near 13x 2027 earnings — both well below their own five-year norms, consistent with room left to re-rate. But the trailing multiples at ServiceNow and Workday expanded almost exactly in step with their price gains over the past three months, meaning much of the July move is unaccompanied multiple expansion rather than fresh earnings beats — a fragile foundation until the next quarter's numbers confirm it.\n\n### The setup\nWhere it stands — Eight of ten names have moved off severe downtrends since late July, but the underlying growth split confirms only NOW, TEAM and VEEV, while WDAY, ADP and MNDY show deceleration.\nWould confirm — ADP's pays-per-control growth exceeds its flat-to-1% FY2027 guide, or Paycom's August 5 report shows employees-on-platform still expanding.\nWould invalidate — Intuit's trend band fails to exit its severe downtrend even as TurboTax filing-season data comes in, or Workday's subscription backlog guide is cut again next quarter.\nWatch next — Paycom reports Q2 results on August 5, 2026; Atlassian reports Q4 fiscal 2026 results on August 6, 2026.\nValuation — ServiceNow trades near 22x forward 2027 EPS versus a historical forward multiple above 40x; Workday's trailing price-to-sales rose roughly in line with its price over 90 days.

Six companies that physically assemble the servers, power shelves and networking gear behind the artificial-intelligence buildout all reported stronger results and raised their outlooks in the last week of July — and every one of their stocks fell anyway. Celestica, Jabil, Flex, Sanmina, Plexus and Benchmark Electronics are contract manufacturers, or electronic manufacturing services (EMS) providers: businesses that build hardware designed by their customers, historically on thin margins. After a year in which the group gained 41% to 128%, all six gave back 8% to 26% over the past 30 days, and each stock's price trend was downgraded from a strong to a mild uptrend within a three-week window between June 30 and July 22 — a synchronized step-down across an entire industry, not one name breaking.

What each company does. Celestica builds data-center switches and server racks mainly for cloud giants, with roughly 80% of revenue tied to that cloud/communications business. Jabil, larger and more diversified, runs an Intelligent Infrastructure segment serving AI customers alongside legacy consumer-electronics and automotive lines. Flex makes data-center power and cooling equipment through its Anord Mardix unit, which it is spinning off as a standalone company targeted for early 2027. Sanmina, after acquiring ZT Systems' server-assembly business, now gets 62% of revenue from communications, cloud and AI customers. Plexus and Benchmark Electronics are the outliers: both build circuit boards and finished devices mainly for medical, industrial and aerospace-defense customers, with little to no disclosed AI revenue.

The results say the business is fine. Celestica's second-quarter revenue rose 62% to $4.70 billion and it raised its full-year revenue guide to $20.5 billion from $19 billion. Jabil raised its AI-related revenue outlook to $13.6 billion from $13.1 billion and said it had won a third hyperscale customer. Flex posted record adjusted earnings per share of $1.00 on 21% revenue growth and confirmed its power-infrastructure spinoff is still on track. Sanmina grew revenue 70% and beat earnings estimates, though its stock still fell about 5% because next-quarter revenue guidance came in slightly below Wall Street's forecast and management flagged rising inventory needed to support the AI ramp. Benchmark Electronics raised its 2026 revenue guide to $3.0 billion, a company record, despite having almost no AI exposure. Plexus raised its guide to more than 20% revenue growth on record quarterly revenue — again, with minimal AI content.

So the sector isn't the reason these stocks fell together — the tape above them is. In the same window, Meta's announcement of a commercial cloud-computing service on July 17 triggered a roughly 35% single-day drop in CoreWeave on fears a top AI-infrastructure customer would compete with its own suppliers. Days later, Alphabet's raised capital-spending guidance triggered broad 'hyperscaler capex scrutiny' across Amazon, Meta and Microsoft shares. By late July, credit-default-swap spreads on CoreWeave and Nebius spiked on fears about how AI infrastructure is financed. That Celestica's own trend downgrade came on June 30 — a full month before its earnings beat — shows the rotation started ahead of, not because of, company news. That Plexus and Benchmark, with almost no AI revenue, fell nearly as hard as the AI-heavy names confirms this is a sector-wide de-rating in high-multiple tech-adjacent stocks, not a verdict on AI-hardware demand specifically.

Valuation: the pullback looks earned, not excessive. Trailing price-to-earnings ratios across the group — 98x for Benchmark, 55x Flex, 50x Sanmina, 45x Jabil, 42x Plexus, 41x Celestica — remain well above the sector's historical 10x-15x commodity-assembly range, even after the correction. Celestica's own gross margin sits near 8.6%, in line with the industry's traditionally thin structure — a reminder that most of this year's gain has come from a valuation re-rating, not a change in what these businesses actually keep from each dollar of sales. Rising memory-chip costs add a real risk: DRAM and NAND contract prices have risen 58%-75% quarter-over-quarter through 2026 on AI-driven memory demand, a cost EMS firms often pass through with little markup — margin pressure on the same GPU-rack revenue that has driven the AI story. On the numbers: the business does not explain the pullback — it CONTRADICTS a demand-driven selloff, since every name accelerated. But the valuation reset looks justified rather than a market mispricing: multiples remain rich versus history, so further room from today's price depends on continued earnings growth rather than another leg of multiple expansion.

The chart move. All six names cut their sma-crossover trend classification from a strong to a mild uptrend inside a three-week span (Celestica June 30, Sanmina July 8, Jabil July 14, Flex July 17, Plexus July 21, Benchmark July 22), a genuinely synchronized signal that lines up with the AI-financing scare rather than any single earnings date — confirming the business and the tape are telling different stories right now.

The setup

Where it stands — All six EMS names beat and raised guidance in July yet fell 8%-26% in 30 days as trend signals downgraded together. Would confirm — Data-center segment revenue (Celestica CCS, Jabil Intelligent Infrastructure) keeps growing above 40% year-over-year next quarter without margin compression. Would invalidate — Any of the six discloses a hyperscaler pulling back orders, or gross margins compress as memory costs rise. Watch next — Sanmina's Q4 FY2026 print (guided $14.0-14.3B revenue) and whether its inventory build resolves without a margin hit. Valuation — Trailing P/Es of 41x-98x across the group remain well above the sector's historical 10x-15x range, even after the pullback.

In early 2026, investors punished enterprise software broadly on fear that AI agents would let companies do more with fewer paid software seats, sending a widely tracked software index down more than 21% and, for the first time ever, below the S&P 500's valuation. That fear didn't materialize for eight companies that supply the internet's underlying infrastructure — content delivery, cloud databases, identity checks and system monitoring — which have gained an average 30% over the trailing year. The group's last holdout, ServiceNow, which sells workflow-automation software large enterprises use to manage IT tickets and other back-office processes, just reported subscription revenue up 24.5% and a bookings measure (contracted-but-not-yet-billed revenue) up more than 21%, both beating its own guidance, and raised its full-year outlook. The trend model tracking its stock moved from deeply bearish to mildly bearish within three trading days of that report — a sign the rally's final laggard is turning on new bookings, not just a bounce in a beaten-down stock.

Broad participation, not two names carrying it. Six of the eight members — Datadog (+92.6% over the year), MongoDB (+49.0%), Akamai (+52.6%), Okta (+45.3%), Snowflake (+40.7%) and Cloudflare (+33.6%) — are all up double digits over the trailing twelve months; only ServiceNow (-39.7%) and Toast (-34.5%) are negative for the year. But those two laggards are now the fastest-moving names over the past 90 days (ServiceNow +20.9%, Toast +13.1%), meaning the group's weakest members are inflecting hardest, not lagging behind a two-name rally. As of July 31, trend patterns confirm the breadth: Cloudflare, Datadog, Okta and Snowflake trade above both their short- and long-term moving averages in a strong uptrend; Akamai, MongoDB and Toast sit in milder uptrends; only ServiceNow remains in a downtrend, though a less severe one than a week earlier.

Does the business support the move? Mostly, yes. Datadog, whose software watches over the health of clients' cloud servers and applications, posted its first-ever $1 billion revenue quarter, growth of 32%, and raised full-year guidance by $240 million after a single quarter, adding customers paying more than $100,000 a year at a 21% clip. Snowflake, which rents out cloud storage and analytics tools that let companies query massive datasets, raised its full fiscal-year revenue-growth guidance from 27% to 31%, helped by an AI coding tool, Cortex Code, that gained more customers than management had modeled. Okta, which verifies employee and customer logins for other companies' software, saw a retention metric inflect for the first time in years, from 106% to 107% of prior-year revenue retained, in a quarter management linked partly to new AI-agent-identity demand. CONFIRMS: the earnings evidence supports the price move for these three, and for ServiceNow above.

The picture is more mixed for the rest. Cloudflare, which runs a global network delivering websites, blocking cyberattacks and offering a computing platform called Workers, grew revenue 28% to $738 million last quarter and guided to more than $5 billion in revenue before 2028, but the stock also fell 24% on its own earnings day in May after the company cut roughly a fifth of its staff to reshape its cost structure around AI — a genuinely two-sided quarter. MongoDB, whose Atlas cloud database is used by developers building AI applications, grew Atlas revenue 29% and saw vector-search adoption nearly double, even as total-company revenue growth is guided to slow to 16%-18% next fiscal year — a real divergence between an exciting AI-adjacent metric and slowing total growth (INCONCLUSIVE). Akamai, the original content-delivery-network operator now pivoting toward security and cloud computing, signed a $1.8 billion, seven-year cloud contract with AI company Anthropic — its largest customer deal in 28 years — but that new business is still under 9% of revenue while its legacy delivery business keeps shrinking about 7% a year. Toast, which sells point-of-sale and payment software to restaurants, added 7,000 net new locations last quarter and raised full-year profit guidance, even as it was one of the year's two decliners.

Does the valuation justify it? Increasingly, no, for the biggest gainers. Datadog's trailing price-to-sales ratio nearly doubled in three months, from 13.6x to 24.3x; Okta's rose from 4.6x to 8.0x while its revenue grows only 9%-11% a year; Snowflake's climbed from 10.5x to 18.6x. That is CONTRADICTS territory — the re-rating from depressed multiples that started this move has largely already happened for these three. ServiceNow's price-to-sales ratio, by contrast, rose a milder 16% (6.7x to 7.8x) even after its earnings beat, and Toast's trailing price-to-earnings ratio actually fell, from 51.8x to 46.2x, because its profit grew faster than its stock — both still look like unwound de-rates with more room, a genuine CONFIRMS for the hypothesis's question about what's left from current prices.

The setup

Where it stands — Six of eight members show earnings-confirmed growth and bullish trend patterns; three of the biggest gainers already re-rated to stretched multiples. Would confirm — ServiceNow's next-quarter bookings growth holds above 20% and Okta's retention metric climbs past 107%. Would invalidate — Datadog, Snowflake or Okta's price-to-sales ratio retraces most of its 90-day gain without a matching growth slowdown to explain it. Watch next — ServiceNow, Datadog, Snowflake and Okta report fiscal third-quarter results between late September and December 2026. Valuation — Datadog trades at 24.3x trailing sales, up from 13.6x three months ago; ServiceNow at 7.8x, up a milder 16% over the same span.

The two publicly traded companies in the "memory chip" corner of the semiconductor market just delivered a textbook case of a stock price moving faster than the business behind it — but only for one of them. Micron Technology, which makes the DRAM and NAND memory chips found in every server and PC and, increasingly, the High Bandwidth Memory (HBM) chips that feed AI processors, told investors its current quarter will be a record on every measure: roughly $50 billion in revenue, gross margins near 86%, earnings near $31 a share, and capital spending raised to about $27 billion for the fiscal year. Its stock still fell about 20% in July, and as much as 39% peak-to-trough. Everspin Technologies, a far smaller company that makes MRAM (magnetoresistive memory) chips mostly for defense, aerospace and industrial customers, fell nearly as hard over the same stretch — despite reporting 13% revenue growth and landing a new $40 million defense subcontract. The two stocks are grouped together as "memory" chipmakers, but the evidence shows only one of them is actually riding the AI memory cycle.

Micron: the business kept accelerating into the drop. Micron's entire 2026 HBM output is sold out under non-cancellable multi-year contracts, and analyst Vijay Rakesh at Mizuho estimates 2026 contract pricing — not just spot prices — could rise as much as 355% for DRAM and 510% for NAND. Management told investors on its latest earnings call that memory-market tightness should persist beyond 2027, and new capacity — a fab in Idaho and an expansion in Hiroshima, Japan — won't produce its first wafers until mid-2027 at the earliest, meaning the near-term supply picture stays tight regardless of the stock's July slide. No guidance was cut ahead of the drop; instead, the decline tracked a broader AI-chip reset that also hit rival SK Hynix, down about 47% from its June peak after a Chinese competitor, ChangXin Memory Technologies, went public and stirred future-competition worries. On the fundamentals, this desk's own numbers CONTRADICT the price action: nothing in Micron's operating results argues for a 20-39% haircut.

Valuation is murkier. Micron's headline forward price-to-earnings ratio, often cited near 7-8x, looks statistically cheap, but this desk's own prior research on the name flagged that figure as "cyclically misleading at peak margins" — memory earnings run hottest right before they roll over. More tellingly, Wall Street's own 12-month consensus price target sat near $483 when the stock traded near $720 in mid-May, already implying the Street saw the shares ahead of its models; at Micron's end-July close of $823, the stock remains well above even that target. That combination — accelerating fundamentals, but a price the Street's own estimates already viewed as extended — reads as a justified de-rating of an ahead-of-itself multiple, not a business breaking down. Call the valuation verdict INCONCLUSIVE, tilted toward "expensive on consensus, cheap on trailing peak margins."

Everspin: riding the label, not the cycle. Everspin's MRAM chips serve data centers, industrial automation, medical, automotive and aerospace customers — a persistent-memory niche structurally separate from the DRAM, NAND and HBM chips that define the AI memory story. Its latest quarter showed $14.87 million in revenue, up 13% year over year, with gross margin near 53% and a near-breakeven net loss — no deterioration whatsoever. Its roughly 35-40% July decline came with no negative company news; if anything, it followed a fresh $40 million, 2.5-year defense subcontract announcement. This desk's own research on Everspin states plainly that AI-related demand for MRAM "remains speculative" and that the real AI memory stack is HBM plus conventional DRAM — not MRAM. With Micron up roughly 654% over twelve months against Everspin's roughly 140%, the category's eye-catching +401% average return is a Micron story wearing a segment label.

The tape confirms an orderly, not violent, pullback. Trend signals for both stocks cut from a strong uptrend to a milder one across every trailing window at once — Micron's on July 30-31 across the 7-, 30-, 90- and 365-day views simultaneously, Everspin's a month earlier, on June 30, across the 90-, 180- and 365-day views. Neither stock appeared among the market's most extreme 30-day movers, consistent with a broad, gradual reset rather than a single-name shock.

The setup

Where it stands — Micron's fundamentals are accelerating into record guidance while its stock and trend signal have downgraded together; Everspin's decline shares no business link to Micron's. Would confirm — DRAM/NAND contract ASPs keep rising and Micron's next quarterly report shows gross margin still near or above 86%. Would invalidate — Customer inventory days rise or Micron guides capital spending up while contract pricing flattens or falls. Watch next — Micron's fiscal Q4 2026 earnings report, due around late September 2026, against its $50B revenue and 86% margin guide. Valuation — MU trades near 7-8x forward earnings, below its own ten-year range, but above the Street's ~$483 consensus target set at a lower price.

Three of the world's largest technology-outsourcing firms — the companies that send armies of consultants and, increasingly, AI-driven delivery teams into corporate back offices — have been stuck in the market's worst-performing downtrend on this desk's watchlist for four straight months. Accenture and Infosys have not logged a single up-day on the trend signal since March 6; Cognizant since March 19. Yet neither company's revenue, bookings nor margins has actually broken down over that stretch, and two comparable peers have already begun to recover on the charts while the three core names have not.

Accenture, the Dublin-based consulting and IT-outsourcing giant that advises companies on everything from cloud migration to AI deployment, posted its worst single trading day ever on June 18, when it narrowed full-year revenue growth guidance to 3%-4% (from 3%-5%) and reported new bookings down 2%-3% year over year, partly blamed on roughly $400 million of Middle East-related revenue drag (Bloomberg). The stock fell from $156 to $128 in a single session on 41.7 million shares — roughly six times normal volume. Cognizant, a mid-cap IT-services and outsourcing firm best known for serving banks and insurers, and Infosys, India's second-largest IT-services exporter, both sold off the same day in sympathy, with Infosys down more than 7% on no company-specific news of its own (CNBC).

What the numbers actually show. Despite the guidance cut, Accenture's quarterly revenue still rose 6% and earnings per share rose 9%, with operating margin expanding to 17% and book-to-bill holding at 1.0 (Investing.com); its generative-AI bookings share has more than doubled to 7.6% of the total, evidence the AI work is additive rather than purely cannibalizing legacy contracts (Luminix). Cognizant beat second-quarter revenue estimates with 4.5% year-over-year growth, expanded margin 40 basis points, and raised full-year earnings guidance — even as voluntary attrition ticked up to 13.0% and headcount stayed roughly flat near 357,000 (PR Newswire), with large-deal bookings still described as intact (DQIndia). Infosys narrowed rather than slashed its constant-currency guidance to 1.5%-3.0%, kept its 20%-22% margin target, and closed $3.6 billion of large deals — 61% of it net-new business rather than renewals — with AI-linked revenue now 8.2% of the total (Upstox; StockTitan). None of the three shows a bookings or margin collapse consistent with the size of the drawdown — the business record here CONTRADICTS the severity of the stock reaction.

Valuation has moved further than the businesses have. Accenture's forward price-to-earnings ratio fell to roughly 11x-12.5x after the crash, below its own 10-year trading range of 12.9x to 37.6x (Macrotrends); Cognizant trades near 12x trailing earnings and Infosys near 15x, both well under IBM's 19.2x reference multiple. Rather than pulling back capital return, Accenture raised its buyback authorization by $2 billion to $7.5 billion — a 62% increase — while yielding 5.2% in dividends (Accenture newsroom). On valuation, this CONFIRMS a group pricing in more damage than its numbers currently justify.

Sector context matters here. Two adjacent names — Wipro and EPAM — already flipped from a strong downtrend to neutral on July 28, and both rallied further alongside Infosys and other Indian IT names after Cognizant's July 29 results beat expectations (Business Today), underscoring that this group trades as a correlated basket, not on isolated company news. IBM's own downtrend has been milder and less consistent, flipping between mild and strong bearish readings rather than holding one signal for months. Demand for AI spending broadly is not collapsing either: Gartner projects enterprise AI spending will rise 34.8% in 2026 to $407 billion, with financial services — the largest client vertical for all three outsourcers — the biggest spender (ValueAddVC).

Since late June, all three have logged higher lows and staged a sharp late-July rally — Cognizant up roughly 40%, Infosys 13%, Accenture 12% over three sessions — without yet earning a band upgrade the way Wipro and EPAM have.

The setup

Where it stands — ACN, CTSH and INFY remain in a strong downtrend on the charts despite bookings, revenue and margins that have decelerated but not broken. Would confirm — Bookings or book-to-bill turn negative for two consecutive quarters at two or more of the three names. Would invalidate — Any of ACN, CTSH or INFY upgrades off the strong-downtrend band, matching what EPAM and Wipro already did July 28. Watch next — Accenture's next quarterly report (expected ~September 2026) for whether bookings growth turns positive again. Valuation — ACN forward P/E ~11x-12.5x vs a 10-year range of 12.9x-37.6x; CTSH ~12x trailing vs a typical 14-15x; INFY ~15x trailing, all below IBM's 19.2x.

Two messaging-infrastructure stocks, one diverging story

Twilio and Bandwidth both sell the plumbing that lets apps send text messages, place phone calls, and now route AI voice agents to actual customers — a category known as CPaaS, or communications platform as a service. Over the past year both stocks rallied hard on hopes that AI agents would need to talk to people through exactly this kind of infrastructure. But when both companies reported quarterly results in the past week, the rallies split apart: one company's growth is real, and the other's was partly an accounting artifact that the market has now repriced.

Twilio (TWLO), the largest programmable-messaging and voice-API provider, processes more than 2.5 trillion customer interactions a year for apps like Uber and Airbnb. Its Q1 fiscal 2026 report, released April 30, showed organic revenue growth accelerating to 16% — the best pace in three years — with dollar-based net expansion (a measure of how much more existing customers spend year over year) rising to 114% from 107%, and non-GAAP operating margin hitting a record 19.8%. Free-cash-flow guidance was raised to $1.08–1.10 billion, and stock-based compensation fell below 10% of revenue for the first time since Twilio's IPO. Its new voice-AI product, ConversationRelay, has logged "millions of minutes" of usage this year, though Twilio hasn't disclosed specific incremental contract value tied to named customers.

Bandwidth (BAND), a much smaller rival that owns its own telephone network rather than reselling carrier capacity, is Salesforce's chosen infrastructure partner for its new Agentforce Contact Center, embedding Bandwidth's voice and messaging software into AI-driven customer-service workflows — a real, named win announced in March, though again without disclosed dollar figures. Its Q2 fiscal 2026 report, released July 29, headlined 22% revenue growth. But analysts flagged that $68 million of that revenue was messaging-surcharge pass-through — fees Bandwidth collects from carriers and passes along — leaving true organic growth in its core Cloud Communications business at just 12%. Non-GAAP gross profit grew only 14%, slower than revenue, and management guided second-half growth below the first half. The stock fell 28–36% over the following days despite an in-line beat and raised full-year guidance — a textbook reaction to a "quality of growth" disappointment rather than a headline miss.

The two verdicts diverge

For Twilio, the business explains the stock: CONFIRMS. Twilio trades around 4.7x forward enterprise value to revenue and roughly 25x EV to free cash flow, both well below its own 10-year median EV/revenue of about 7.2x and far cheaper than peers Datadog (~15x), CrowdStrike (~17x) and Cloudflare (~22x). Twilio's July slide — from about $236 on June 4 to a trough near $183 on July 22, recovering to $197 by month-end — coincided with roughly $342 million in insider selling over three months and pre-earnings positioning ahead of its August 6 report, not a fundamental reversal; Stifel raised its price target to $260 on July 10.

For Bandwidth, the tape now agrees with a business the market had been mispricing: CONTRADICTS the year-long rally's premise. Its price-to-sales multiple compressed from 2.49x in May to about 1.52x post-crash — a cheaper stock, not an inflated one, as the market absorbed the fee-inflation disclosure. Notably, the desk's own trend-following signal still classified Bandwidth as being in a sustained uptrend through July 31, even after a roughly 43% two-day collapse (from $64.37 on July 22 to $36.95 on July 29) — a reminder that price-trend indicators lag fundamental catalysts and can stay stale through a genuine break. Bandwidth did use the period productively on the balance sheet, retiring its 2026 convertible notes and issuing new 2032 converts, pushing out its refinancing risk.

Both companies face the same structural headwind and the same open competitive question. A2P/10DLC carrier surcharges — regulatory fees tied to business texting — are a rising, sector-wide drag inflating reported revenue industry-wide, a ~$190 million hit to Twilio's own margins this year. And hyperscalers — Amazon's Chime SDK, Microsoft's Azure Communication Services, Google Cloud — are expanding into CPaaS, pushing smaller CPaaS vendors like Sinch and Vonage into consolidation. So far AI-agent traffic is described industry-wide as adding message and minute volume rather than displacing the routing layer — a tailwind, not yet a threat, for both names.

The setup

Where it stands — Twilio's rally tracks accelerating usage and margins; Bandwidth's cracked once fee-inflated revenue was disclosed, even as its trend signal stayed bullish. Would confirm — Twilio's August 6 report shows organic growth holding near 14%+ and net expansion above 110%; Bandwidth's core Cloud Communications growth stabilizes near 12%. Would invalidate — Twilio's organic growth falls below 8% for two straight quarters or net expansion drops under 108%. Watch next — Twilio reports fiscal second-quarter results on August 6, 2026. Valuation — TWLO ~4.7x forward EV/revenue vs. its own 7.2x 10-year median; BAND ~1.5x trailing price-to-sales, down from 2.5x in May.

A sector-wide selloff splits into two very different stories

Nine stocks tied to the idea that nuclear power will feed the electricity needs of AI data centers have fallen sharply since mid-July — but the group is not falling for one reason. Cameco Corp (CCJ), the Canadian uranium miner that also owns 49% of reactor-builder Westinghouse, and BWX Technologies (BWXT), which builds naval nuclear reactors for the US Navy and is expanding into commercial small modular reactors, both saw their stock-price trend flip from a strong uptrend to a strong downtrend on both the 90-day and 365-day view in early July — even as their order backlogs and realized prices kept growing. Meanwhile pre-revenue reactor developers Oklo (OKLO), NuScale Power (SMR), NANO Nuclear Energy (NNE) and Lightbridge Corp (LTBR) are falling for a more straightforward reason: cash burn is accelerating, share counts are rising, and commercial revenue keeps getting pushed further into the future.

What actually happened

Mid-July brought a one-day drop of 8-9% across Oklo, X-energy, NuScale, NANO Nuclear and Uranium Energy Corp, which commentary at the time attributed to investors rotating out of "the most expensive AI-linked trades" and demanding proof of commercial delivery over speculative growth narratives — not a change in the physical uranium market, which held broadly steady, with spot ending August at $75.13/lb versus a January peak above $101/lb. By July 31, all nine names — Centrus Energy (LEU), the only US-owned commercial uranium enrichment company; Uranium Energy Corp (UEC), the largest US uranium miner by licensed capacity; and Energy Fuels (UUUU), operator of the only conventional uranium mill in the US, among them — were sitting in a downtrend on the chart. The 30-day declines ranged from BWXT's 12.2% to Oklo's 25.8%; Centrus was the lone exception, up 9.1%.

Fundamentals: CONTRADICTS for the established names, CONFIRMS for the startups

Cameco's fundamentals argue against the selloff. Its second-quarter realized uranium price rose 15% year-over-year to $93.13 a pound, and the company holds contracts for average annual deliveries above 28 million pounds over the next five years, while Westinghouse carries a pipeline of 91 AP1000 reactor opportunities and has confidentially filed for an IPO. Yet Cameco's trailing price-to-earnings ratio compressed from 104-117x in May to roughly 80x by late July, and its market value fell from near $50 billion to $36.8 billion — a valuation reset, not a business one. BWXT tells a similar story: backlog reached $8.65 billion, up 77% year-over-year, and JPMorgan initiated coverage at Overweight in late July, projecting double-digit revenue growth through 2028, even as its own P/E fell from the mid-50s to 42x. That divergence — backlogs and prices rising while multiples fall — is a CONTRADICTS reading on the sell-off's premise for these two names; the business hasn't slowed, the stock got expensive and is now less so.

The pre-revenue reactor group tells the opposite story: CONFIRMS. Oklo's trailing free cash flow is negative $154 million, its worst on record, funded by a $1.18 billion share sale that has more than doubled its share count, with no reactor revenue expected before late 2027 and no NRC design approval yet. NuScale's quarterly revenue collapsed to $0.6 million from $13.4 million a year earlier as one-time licensing fees rolled off, and its price-to-sales ratio actually rose, from 121.6x to 161.5x, because revenue fell faster than the stock price — a sign the de-rating is deserved, not excessive. Lightbridge, which develops nuclear fuel rather than reactors, has reported zero revenue every quarter through the first quarter of 2026, funding itself with a $176 million share sale. Centrus sits in between: it holds a $2.3 billion commercial backlog and a $900 million Department of Energy award for enrichment expansion, but the DOE has deferred a decision on Centrus's waiver to keep importing Russian-enriched uranium, an unresolved policy risk. Uranium Energy Corp's production costs rose 25% sequentially to $54.61 a pound as new mines ramped, while its stock still trades at 225 times sales — among the highest multiples in the group. Hyperscaler nuclear deals — Google's 500-megawatt Kairos Power agreement, Amazon's $700 million X-energy investment, Meta's commitments across four reactor developers — are real but dated 2030 to 2035, supporting the case that reactor-name revenue is still years away regardless of how the stocks trade this month.

The one-paragraph technical read

The band break is real and multi-horizon: Cameco and BWXT both flipped from a strong uptrend to a strong downtrend on the 90-day and 365-day charts in early July, not just the 30-day. But the twelve-month picture still separates the two stories — Cameco, BWXT, Uranium Energy and Energy Fuels are each still up 11% to 29% over the past year, while NuScale (-81%), Oklo (-46%), NANO Nuclear (-53%) and Lightbridge (-37%) are down sharply, confirming the reactor micro-caps are dragging the group average, not the fuel-cycle names.

The setup

Where it stands — Fuel-cycle names (Cameco, BWXT) de-rated on multiples despite growing backlogs; pre-revenue reactor names de-rated on deteriorating cash burn and dilution. Would confirm — Cameco's realized price or contracted volume declines in its next quarterly report, extending the de-rating into the business itself. Would invalidate — Oklo or NuScale posts sequential revenue growth without a new share issuance, breaking the dilution-and-burn pattern. Watch next — NuScale's RoPower final investment decision, previously targeted for 2027, and Centrus's pending third Russian-uranium waiver filing. Valuation — Cameco trades near 80x trailing earnings, down from 104-117x in May but still above its multi-year norm; Oklo and NuScale carry no meaningful P/E given losses.