DK Street Journal

Prices for LPDDR5X memory rose 89% in the second quarter of 2026 and DDR4 by as much as 51%, as artificial-intelligence data-centre buyers soaked up supply that would otherwise have gone into laptops and general-purpose servers (TweakTown). For the companies that buy finished boxes and move them on, that inflates the invoice, thins the spread and ties up cash — the defining pressure on this group of six infrastructure names this summer.

The box movers

CDW is the clearest case. On its 5 August call management raised full-year guidance to mid-single-digit gross-profit growth and high-end high-single-digit adjusted earnings growth, said written orders exceeded invoicing with backlog "significantly elevated", and still watched the stock drop 12.3% (StockStory). The reason sits below the revenue line: revenue up 10.0%, gross profit up only 6.3%, operating income up 2.0%, and adjusted free cash flow of $278m year-to-date, 42% of non-GAAP net income against an 80-90% target. This is the second such quarter — CDW fell more than 20% in May when Q1 gross margin contracted 60 basis points to 21.0% amid memory-supply constraints (Investing.com). Net leverage is 2.5x within its 2-3x target and $545m of stock has been repurchased this year.

Ingram Micro, the global distributor that sells devices, servers, networking and cloud subscriptions to resellers, took the same cost shock and produced the opposite result: record net sales of $14.53bn up 13.6%, gross profit $959m up 14.2%, operating income up 38.6%, gross margin 6.60% versus 6.56%, and a guide to continued double-digit growth (earnings call). The cost is on the balance sheet: trailing free-cash-flow yield is negative 3.6%, the working capital of financing pricier inventory.

Verdict on the business: SPLIT. Component inflation confirms the tape at CDW and contradicts it at Ingram Micro, whose shares slipped 3% on a record quarter.

Two collapses that skewed the group

IBM lost about $67bn of market value on 14 July, its worst single day on record, after pre-announcing quarterly revenue of $17.2bn against $17.86bn expected (CNBC); chief executive Arvind Krishna said the company "faltered" and "did not adapt and move quickly enough" (Fortune), and the miss dragged the wider software complex with it (Bloomberg). Revenue growth decelerated from 9.5% to 1.1%, gross profit fell 0.7% and operating income 15.3%. Management blamed large licence deals in transaction-processing software as customers redirected budget to supply-constrained servers — the same memory shock, seen from the vendor side. Annual recurring revenue of $24.6bn grew 8% and Red Hat accelerated to 11%.

N-able, which sells remote-monitoring, backup and endpoint-security tools to the managed service providers that look after small businesses, fell 36% on 10 August (Investing.com). Quarterly earnings met consensus; full-year revenue guidance was cut to $539-542m from $554-559m because its largest renewal group came in at mid-80s renewal rates against higher-80s expected, with some endpoint-security customers moving to SentinelOne. Growth has gone 13.1% to 5.3% in two quarters and gross margin from 78.1% to 76.8%. This is competitive loss, not a cost cycle.

The outsourcers, on other news

DXC Technology, a 115,000-person IT outsourcer, rose despite organic revenue down 6.7%, gross profit down 20.8% and non-GAAP earnings per share down 41% (company release) — because a consortium of Apollo Global Management and Kyndryl has approached it with an all-cash proposal discussed at $22-25 a share against a $10.91 close (Nasdaq). Kyndryl, spun out of IBM in 2020 to run mainframes and core platforms for banks and telecoms, reported revenue down 3% but consulting up 10% and hyperscaler-related revenue above $530m growing 34% (Kyndryl), with trailing signings of $14.2bn now exceeding revenue and fiscal-2028 targets of over $1.2bn pretax income against a $3.0bn market value. Kyndryl also quantified the direct-sourcing threat from the other end: client spend routed through IBM has halved to under $2bn as customers buy hardware direct and keep the service.

Verdict on valuation: MIXED. CDW at 12.6x forward against 16.4x trailing with a 6.3% free-cash-flow yield, and Ingram at 8.3x forward and 7.6x enterprise value to EBITDA, sit well below their trailing marks while gross-profit dollars still grow. IBM at 19.2x forward is the most expensive name here and has de-rated only modestly from the roughly 22x it carried in May. DXC at 4.2x forward and 0.58x book is cheap on takeover odds, not on trading.

The tape has already moved ahead of the fundamentals in places: DXC, Kyndryl and N-able all climbed out of sustained downtrends into neutral trend structure during July, and CDW's 50-day average pushed further above its 200-day, while IBM alone broke down. Only IBM's chart agrees with its income statement.

The setup

Where it stands — A memory-cost shock is inflating hardware revenue and compressing reseller margins, while IBM and N-able fell for unrelated, company-specific reasons. Would confirm — CDW's next quarter converting adjusted free cash flow back toward 80% of non-GAAP net income as the inventory build unwinds. Would invalidate — CDW gross margin below 20.1% again with full-year gross-profit guidance reduced from mid-single-digit growth. Watch next — CDW and Ingram Micro report third-quarter results in early November 2026; N-able guidance assumes no execution improvement in the second half. Valuation — CDW 12.6x forward versus 16.4x trailing; Ingram 8.3x versus 15.2x; IBM 19.2x forward against roughly 22x in May.

Four budgets, one reaffirmation

Between 22 and 31 July the four largest buyers of artificial-intelligence computing hardware each told investors what they intend to spend. Three raised the number. The fourth did not, and it was the one the market paid for.

Microsoft — the software company behind Office, Windows and the Azure cloud platform, run by Satya Nadella — reported fiscal fourth-quarter revenue of $90.0 billion, up 18%, with Azure growth accelerating to 43% from 40% and crossing $100 billion of annual revenue. But the line that moved the stock was finance chief Amy Hood's statement that calendar-2026 capital spending remains approximately $175 billion — unchanged, not raised, a reaffirmation that extended the after-hours gain from 3% to as much as 8%.

Amazon, whose Amazon Web Services division rents computing capacity alongside its retail marketplace, went the other way, lifting 2026 cash capital expenditure to about $220 billion from $200 billion on inflated memory prices while warning capacity will still trail demand. Alphabet, parent of Google Search, YouTube and Google Cloud, raised its range to $195-205 billion from $180-190 billion and saw its shares fall nearly 5% for it. Meta Platforms, owner of Facebook, Instagram and WhatsApp, raised capital spending to $130-145 billion and declined to give any 2027 figure at all.

The businesses: two accelerating, one mispriced, one shrinking

Amazon and Microsoft CONFIRM their re-ratings on operations. AWS revenue reached $42.2 billion, up 36.7% — a fifth consecutive quarter of acceleration, at a $169 billion annualised run rate — with operating margin up 520 basis points to 39% and a $496 billion backlog. Management said most 2027 capacity is already reserved. Its Trainium and Graviton in-house chip business runs above $25 billion annualised. Microsoft's fiscal 2026 revenue was $331.8 billion, up 17.8%, operating margin 46.8%; but gross margin has now compressed three straight years, 69.8% to 68.8% to 67.9%, as AI capacity costs land in cost of revenue. Its record $678 billion of commercial contracted backlog carries a caveat: almost a third traces to a single customer, OpenAI, and excluding it the figure grew 25% rather than 84%.

Alphabet is the divergence, and it runs the wrong way. Google Cloud revenue rose 82% to $24.8 billion, operating margin widened from 20.7% to 35.6%, backlog rose $50 billion sequentially to $514 billion, and group operating income grew 30.4% on 24.2% revenue growth. The shares are down over three months. Meta CONFIRMS its own decline: revenue grew 28% to $60.8 billion while operating income fell 8.2% and margin dropped from 43.0% to 30.9%, on $2.4 billion of legal charges and $1.2 billion of severance. Earnings of $6.18 a share missed the $7.14 consensus and free cash flow collapsed from $8.5 billion to $784 million.

Valuation: the verdicts differ by name

Because huge non-operating gains distort trailing earnings at Alphabet and Amazon, price-to-gross-profit is the cleaner comparison. Microsoft has gone from 13.4 times trailing gross profit on 29 July to 16.7 today — the top of its 2026 range, re-rated in eight sessions, at 25.8 times forward earnings. That CONTRADICTS the case for further runway from here. Amazon at 7.57 times is cheaper than May's 7.75 despite the rally, on the lowest enterprise-value-to-EBITDA of the four at 12.3 — CONFIRMS. Alphabet at 15.8 times gross profit versus 17.6 in May, roughly 23.6 times clean 2027 consensus earnings of $15.01, while cloud compounds at 82% — the clearest CONTRADICTION between business and tape in the group. Meta at 8.10 times gross profit, down from 10.34 in May, is a de-rating its own falling profits justify: INCONCLUSIVE, because earnings are still going down.

Who funds 2027

The financing has changed character. Across the five largest hyperscalers, incremental annual debt has risen from 9% of capital spending in fiscal 2024 to 32% on a trailing-twelve-month basis, with Alphabet pricing an $84.75 billion equity raise in June and aggregate 2026 capex above $690 billion. Alphabet's second-quarter free cash flow was negative $5.9 billion; Amazon's trailing free-cash-flow yield is negative 0.39%. Depreciation assumptions are doing work too: the group extended server lives from three-to-four years toward six during the early 2020s, and Amazon has since cut back, trimming certain servers to five years and reducing 2025 operating income by about $0.7 billion.

The tape agrees only in parts. Microsoft gapped 15.5% on 30 July and Amazon 15.3% on 31 July, while Meta fell 8.0%; the two winners supply essentially the entire group's one-month gain. Microsoft's 50-day average sat below its 200-day every session of 2026 until 6 August, its first bullish crossover of the year. Over twelve months, Alphabet is up 82% and Meta down 22%.

The setup

Where it stands — Three of four raised 2026 capital spending; only Microsoft's reaffirmation was rewarded, and only Alphabet's accelerating cloud is unpriced.

Would confirm — Alphabet delivering the "significant" 2027 capex increase alongside cloud growth holding above 50% next quarter.

Would invalidate — Azure decelerating below 40%, or AWS backlog conversion slipping while Amazon's free cash flow stays negative.

Watch next — Third-quarter results in late October, when Meta must finally supply a 2027 capital-spending figure.

Valuation — Microsoft 16.7x trailing gross profit versus 13.4x on 29 July; Alphabet 15.8x versus 17.6x in May.

Three companies are grouped together as identity-and-access-management vendors — the software that issues, governs and revokes the credentials employees use to log in. Over the past month the group has averaged an 8% gain. Almost all of it belongs to one name, and for once the fundamentals point the same way.

The business that grew

SailPoint sells identity governance: software that decides who inside a company may reach which system, reviews those entitlements, and increasingly extends the same control to machine accounts and autonomous software agents. Its June quarter showed annual recurring revenue of $1,163m, up 26% year on year, on revenue of $280m, up 22%. Revenue growth has held in the low twenties for four straight quarters — 33.1%, 19.8%, 22.7% and 21.6% — roughly double Okta's pace. Gross margin under generally accepted accounting principles (GAAP) rose 9.3 percentage points to 64.7%, as the accounting amortisation left over from Thoma Bravo's 2022 buyout rolls off; gross profit grew 41.9% on 21.6% revenue growth.

It is not yet profitable. The quarter carried an operating loss of $79.8m, a −28.5% margin, worse sequentially than the prior quarter's −13.6% though far better than the −80.3% a year ago. So earnings multiples do not apply: the usable lens is 9.5 times trailing sales and 14.3 times trailing gross profit. Thoma Bravo's February 2025 relisting priced 60 million shares at $23, raising $1.38bn. At $18.76 the stock is about 18% below that price and 20.6% below its twelve-month high, on a revenue base that has grown from $450m in fiscal 2022 to $1.07bn in fiscal 2026. Verdict: the business CONFIRMS the move, and the valuation suggests the re-rating has not happened yet.

The timing is the caveat. The gain came in nine consecutive up sessions into early August, a cumulative 25% that added about $2.0bn of market value, starting the week the Philadelphia Semiconductor Index fell over 5% and 44 of 51 software stocks finished July positive — part of a rotation in which chip stocks excluding Nvidia shed nearly $1.7 trillion of market value in July. SailPoint led its two peers, but it did not move on its own news.

The multiple that grew instead

Okta sells single sign-on, multi-factor authentication and directory services to enterprises and governments, plus the Auth0 customer-login business it bought in 2021. Its profitability turnaround is real and finished: fiscal 2026 delivered its first meaningful GAAP profit — $153m of operating income against a $63m loss the prior year — and last quarter operating income grew 43.6% on 11.2% revenue growth, lifting operating margin to 7.3%. Remaining performance obligations rose 16%, net revenue retention inflected up to 107%, and 570 customers now spend over $1m a year. It has agreed to buy identity-threat-detection startup Permiso Security for close to $200m, aimed at policing AI agents' access.

But growth is decelerating — 12.7%, 11.6%, 11.6%, 11.2% across four quarters, guided to 9–10% this year — while the multiple has run. In early May the shares stood at $13.5bn of market value and 4.6 times sales; today it is $25.1bn and 8.4 times, 39.3x forward earnings for consensus revenue growth of 10.0% this year and 9.5% next. A consensus twelve-month target of $120.45 across 44 analysts sits roughly 13% below the $151 spot price. Verdict: the business is INCONCLUSIVE — margins expanding, growth shrinking — and the valuation CONTRADICTS the advance.

BlackBerry, the third name, no longer sells identity software at all: it divested Cylance, and QNX, its embedded automotive operating system, supplied $72.3m of $152.9m in quarterly revenue. Management raised full-year guidance to $594–621m citing a royalty backlog near $1bn, yet the stock fell 16.8% over the month from 89x trailing earnings after a June melt-up. Its 50-day average slipped below the step it had held since spring on 6 August — a deterioration, while SailPoint's crossed upward on 31 July.

The setup

Where it stands — SailPoint's recurring revenue grew 26% and its shares trade below their February 2025 relisting price; Okta's multiple doubled on decelerating growth. Would confirm — SailPoint's next quarter holds annual recurring revenue growth above 25% with the operating loss narrowing from −28.5%. Would invalidate — SailPoint ARR growth slipping toward Okta's high single digits, or gross margin gains stalling below 65%. Watch next — Okta reports fiscal second-quarter results after the close on 26 August 2026, guiding to $790–794m. Valuation — SailPoint at 9.5x trailing sales, no forward earnings multiple while loss-making, versus roughly 15x at its 2025 listing.

The layer of the semiconductor industry that gets paid per package assembled and per die tested had an unusually good July. ASE Technology Holding, the Taiwanese firm that is the largest chip packager outside Taiwan Semiconductor Manufacturing Company (TSMC) itself, reported July consolidated revenue of NT$73.8bn, up 43.2% year over year and 12.2% on June. ChipMOS Technologies, a Hsinchu house that assembles memory, mixed-signal and display-driver chips, reported July revenue up 43.6% year over year, its highest month since 2014, and attributed it explicitly to an artificial-intelligence-driven imbalance between demand and available capacity. Both prints landed after their shares had already fallen.

What these companies do, and what happened to them

Outsourced assembly and test firms — OSATs — take finished silicon wafers from foundries and turn them into packaged, tested chips. In the AI era that work has moved up the value chain: 2.5D silicon interposers, fan-out wafer-level packaging and the chip-on-wafer-on-substrate (CoWoS) process that binds accelerator dies to high-bandwidth memory.

Seven such companies fell together in the 30 days to 10 August: ASE (-9.8%), Amkor Technology, the Arizona-based packager and the only large American one (-22.8%), ChipMOS (-28.0%), Powertech Technology, a Hsinchu memory-packaging specialist (-9.8%), China's JCET Group (-23.3%) and Tongfu Microelectronics (-10.6%), and Korea's Hana Micron (-15.9%). The breadth is real; the path was not gradual. Every one took a 28-47% peak-to-trough drawdown inside three weeks, and every one has since rebounded 20-35% off its low. The trigger was mostly external — chip stocks shed more than $1 trillion in the week to 29 July as investors questioned hyperscaler returns — but these names fell three times as far as Nvidia or TSMC did over the same sessions.

The businesses: CONTRADICTS

Every member with reported numbers accelerated in the window the prices fell. ASE's revenue growth ran 7.8%, 9.2%, 17.4% and then 34.1% year over year across four quarters, with gross margin climbing from 17.1% to 21.0%. Its assembly, test and materials segment — 66% of revenue and 94% of operating profit — set a record, up 36%, and management guided the third quarter up 21-22% sequentially with advanced capacity running at 80-85% utilisation and pricing it called "very friendly." Amkor grew 25.6% with gross margin up 4.75 points and utilisation moving from the low-70s to the high-70s. Powertech grew 28.0% with margin up 5.85 points and net income nearly tripled; it also agreed in July to invest in a joint venture with Broadcom. Taiwan's OSAT industry as a whole grew 23.7% year over year in June.

One genuine company event sits inside the selloff. Amkor's largest division, legacy communications packaging, was guided down high single digits on phone-assembly migration to Vietnam and Android weakness, taking total third-quarter guidance to $1.95-2.05bn against $2.11bn consensus; the stock fell 24.7% in a day despite beating on revenue and earnings. Its AI-exposed computing segment set a record and was guided up roughly 30% sequentially, and it announced a ten-year advanced-packaging agreement with TSMC and a multi-year Nvidia partnership in the same print.

The structural fear — that TSMC keeps the AI packaging work for itself — did not hold up either. TrendForce reported on 5 August that TSMC is expanding outsourcing of the chip-on-wafer front-end step it had previously retained, on top of an estimated 240,000-270,000 CoWoS wafers already slated to outside partners in 2026.

The valuations: split three ways

Amkor is where the arithmetic is cleanest. At $54.43 it trades at 22.0x consensus 2026 earnings of $2.48 a share, against 32.3x for the same estimate at its 1 July peak — a ten-turn de-rating with the number unchanged — on 10.4x enterprise value to EBITDA and a 3.8% free-cash-flow yield, the only positive one here. The consensus target rose to $75.50 after results, in a $60-92 range. Its 2027 estimate implies only 7.9% growth, the Arizona depreciation drag showing through. Powertech sits at 22.7x forward against 28.6x trailing, and targets mass production of through-silicon-via memory packaging in the fourth quarter.

At the other end, JCET trades at 109.7x trailing earnings and Tongfu at 75.4x, both up over three months, after Tongfu's RMB 4.4bn placement to expand memory and computing packaging. ASE is inconclusive: its forward figures are currency-corrupted, and on trailing metrics (45.3x earnings, 19.3x EBITDA) it is the dearest of the Taiwan-US trio. Hana Micron, which supplies packaging to Samsung and SK Hynix, has no usable multiple at all — and faces KRW 120tn of in-house Korean packaging and memory investment announced on 2 July.

The cost is cash. ASE raised 2026 capital spending by $2bn to $10.5bn and expects negative free cash flow into 2027; trailing free-cash-flow yields are -1.1% at ASE and -6.0% at Powertech. Verdict: the business CONTRADICTS the month's tape; the valuation CONFIRMS only at Amkor and Powertech, and CONTRADICTS at the two Chinese listings.

The setup

Where it stands — Monthly revenue at the Taiwanese packagers accelerated through July while their shares fell, with only Amkor's phone division actually guided lower.

Would confirm — August monthly revenue at ASE and ChipMOS holding above 40% year-over-year growth, and ASE's third-quarter assembly gross margin printing 28-29%.

Would invalidate — ASE deferring any part of the $10.5bn 2026 capital plan, or advanced-capacity utilisation slipping below the 80-85% it reported.

Watch next — September monthly revenue releases in the first ten days of the month; Amkor's third-quarter results against $1.95-2.05bn guidance.

Valuation — Amkor 22.0x forward earnings versus 24.3x trailing and 32.3x at its 1 July peak; ASE 45.3x trailing, JCET 109.7x.

Between 17 June and 30 July, all six of the publicly traded North American contract electronics manufacturers that assemble artificial-intelligence server racks reported quarters that beat estimates and raised full-year guidance. This is the layer beneath the chip designers: firms paid per box built, on gross margins of 8.8% to 12.3%, that screw together the racks, power shelves, sheet metal and interconnects the cloud operators order.

Celestica, the Toronto assembler of switches, routers and custom racks for hyperscalers, grew June-quarter revenue 62.4% to $4.70bn — accelerating from 52.8% — and raised 2026 guidance to $20.5bn in revenue and $11.30 in earnings per share, up 87%. It disclosed custom-rack work for OpenAI reaching mass production in 2027 and a design-and-manufacturing role on Advanced Micro Devices' Helios platform, plus ten 1.6-terabit networking programmes ramping this half.

Sanmina, the San Jose builder of printed circuit boards, backplanes and enclosures, grew revenue 69.7% and expanded operating margin 173 basis points to 6.43%, the widest expansion in the group. Its cloud and AI end-market hit $2.15bn, 62% of the company, up 173%, with the ZT Systems manufacturing business bought from AMD contributing $1.1bn. Flex, the Singapore-founded maker of switchgear, busway and power-distribution gear for data centres, grew 20.6% — a fourth straight quarterly acceleration — lifted gross margin 72 basis points and guided its cloud-and-power segment to 65–75% growth on booked visibility covering more than 90% of the next three quarters. Benchmark Electronics, the smallest at $2.96bn, grew 17.7% with its AI programmes up 71% and guided to a record $3.0bn year.

The inversion

The two exceptions went the other way in the accounts — and the right way on the tape. Jabil, the Florida manufacturer whose Intelligent Infrastructure segment is now 49% of revenue, saw growth decelerate from 23.1% in February to 11.8% in May, with operating margin slipping to 5.09%; it nonetheless raised full-year revenue to roughly $35bn with AI-related sales of $13.6bn. Plexus, the Wisconsin builder for medical, defence and industrial customers, grew 28.1% but saw operating margin fall to 4.70% from 5.26% and net income decline 4.7%, even as its opportunity funnel reached a record $4.5bn.

Jabil and Plexus are the only two names up over the past 30 days. The four that accelerated and expanded margins are the four that fell. On business momentum, the move CONTRADICTS the fundamentals for Celestica, Flex, Sanmina and Benchmark, and CONFIRMS only for Jabil and Plexus — in reverse.

What actually moved them

The repricing was sector-wide and preceded the results. All six lost their strongest uptrend rating in a rolling sequence between 30 June and 22 July — Celestica first, Benchmark last — every one of them before the 27–30 July reporting cluster. That window covers the global technology selloff of 16–17 July and the late-July session in which Meta said it had built enough capacity to become a net seller of AI compute, erasing roughly $200bn from chip and cloud values. Company-specific damage came after: Sanmina fell 20.5% on 27 July because next-quarter guidance landed 2% below consensus; Flex fell on 29 July despite record adjusted earnings as four banks cut price targets on valuation; Celestica dropped 14.8% on 6 August after pricing a $3bn equity offering, roughly 8–10% dilution for working capital and capital expenditure.

Valuation: two cells, not one

Multiples compressed because estimates rose faster than prices fell. Celestica's trailing price-to-earnings went from 50.2x in early May to 32.7x, its price-to-gross-profit from 28.7x to 20.2x; Flex from 62.2x to 46.9x; Sanmina from 44.2x to 35.6x. Forward multiples now span nearly twofold: Sanmina 16.8x, Flex 25.8x, Jabil 26.8x, Benchmark 27.9x, Celestica 28.0x, Plexus 31.5x. Trailing free-cash-flow yields span more than tenfold — Sanmina 9.46%, Jabil 4.21%, Celestica 1.42%, Plexus 0.86%.

On valuation the verdict is INCONCLUSIVE as a group and splits by name. Plexus carries the highest forward multiple on the lowest consensus earnings growth (19.5%) with margins going backwards. Sanmina carries the lowest on 103% expected earnings growth — but consensus already models deceleration to 15.8% revenue growth in 2027, and management warned working capital will build as the accelerated-compute programme ramps.

The live risks are supply, not demand. Benchmark said complex chip lead times went from 3–5 months to 7–12; Celestica cited 52-week-plus waits and called materials, not factory space, the bottleneck. Server memory contract prices are rising 13–18% quarter over quarter with lead times at 58 weeks — a direct cost pass-through problem for assemblers with no pricing power. Against that, the four largest cloud operators guide to roughly $725bn of 2026 capital spending, up about 77%.

The setup

Where it stands — Six assemblers raised guidance in July; four accelerating names fell hard, the two decelerating names rose. Would confirm — Celestica delivering the guided $20.5bn 2026 revenue and Flex's cloud-and-power segment printing 65%+ growth next quarter. Would invalidate — Sanmina's next quarter landing below its $3.30–3.60bn guide, or any hyperscaler trimming 2027 capital spending. Watch next — Jabil reports fiscal fourth-quarter results in September, its first commentary since 17 June. Valuation — Forward price-to-earnings spans 16.8x (Sanmina) to 31.5x (Plexus); Celestica at 32.7x trailing versus 50.2x in May.

Texas has more electricity demand queued up than it can plausibly serve. The state's grid operator, the Electric Reliability Council of Texas (ERCOT), is processing roughly 474 gigawatts of interconnection requests — more than five times the state's record peak demand, about 90% of it data centers. On 3 August, Governor Greg Abbott directed ERCOT and the Public Utility Commission to audit every one of those projects before any further connections advance, effectively pausing the pipeline that the state's regulated wires companies have been building capital plans around.

That matters because regulated utilities are not paid on power prices. They earn an allowed return on capital they put in the ground, so a load pipeline is a capital-spending pipeline, and a pause on the first is a pause on the second.

The two Texas names

Sempra, a San Diego holding company that owns the Oncor transmission network serving 3.8 million Texas customers, gas and electric utilities in California, and liquefied natural gas (LNG) export terminals on the Gulf Coast, is the most exposed. The freeze clouds Oncor's roughly 44 gigawatt large-load pipeline, though management publicly supported the pause and said KKR's purchase of 45% of Sempra Infrastructure remains on track to close this quarter, deconsolidating about $9bn of debt.

The business is not deteriorating. Second-quarter revenue was flat at $2.997bn but operating income rose 42.5% to $832m and net income 68.5% to $797m, with margin widening from 19.5% to 27.8%. Adjusted earnings of $1.16 a share beat the $1.07 consensus and guidance was affirmed. Consensus has 2026 earnings at $5.11, up 11.5% — the fastest in this group — yet Sempra carries its cheapest forward multiple, 16.4 times, against 23.0 times trailing. This desk's own read in May had it at 33.8 times trailing. That is a genuine dislocation, on a business whose Texas growth is delayed rather than cancelled.

CenterPoint Energy, the Houston electric and gas distributor, is the opposite case. Revenue rose 10.7% to $2.152bn, operating income 28.1%, and it lifted its ten-year capital plan to $66.7bn with no new equity, guiding Houston Electric rate-base growth above 18% a year. But $800m of that increase funds system upgrades for ERCOT's "batch zero" large-load approval process — the process ERCOT has now delayed — and CenterPoint's 14 gigawatts of eligible projects sit inside it. At 21.3 times forward earnings, the most expensive here, the multiple was underwriting a queue the state has just stopped.

Where the gas generation actually is

Only three of these seven own material gas-fired plant. Southern Company, which serves 8.7 million customers across Georgia and Alabama, raised 2026 guidance to the top of its $4.50-$4.60 range as data-center usage jumped 55% and contracted large-load agreements passed 17 gigawatts, backed by roughly $21bn of collateral and minimum bills covering the full incremental cost to serve — the cost-allocation question settled in shareholders' favour. Georgia Power has signed a 25-year agreement with OpenAI for 3.2 gigawatts at the $20bn-plus Project Camellia campus, powered mainly by new gas, and the Department of Energy closed $26.5bn of loans to Southern's utilities financing 5 gigawatts of new gas generation. Southern trades at 20.2 times forward earnings.

Duke Energy, with 50,259 megawatts across six states, settled its North Carolina rate case at a 9.8% allowed return, reported adjusted earnings up 14%, and has raised its GE Vernova turbine order to 26 units. It trades at 18.6 times forward against 18.7 trailing — no de-rating priced at all — and the group's lowest enterprise value to EBITDA, 11.6 times. OGE Energy, an Oklahoma utility with 7,207 megawatts, is the smallest and carries the highest trailing free cash flow yield, 11.5%.

The other two are not gas generators. Dominion Energy's headline project is $11.65bn of offshore wind, 81% complete, with 53 gigawatts of data-center capacity in contracting. Its tape and its accounts disagree: revenue grew 19.6% while operating income fell 2.4% and net income 55%, and the trailing multiple expanded from 18.1 to 23.3 times on the slowest consensus growth here, 4.9%. Public Service Enterprise Group's generation arm is nuclear, and its New Jersey utility faces a governor who declared a utility affordability emergency; GAAP operating income fell 43.6%.

Verdicts

Does the business explain the selloff? CONTRADICTS at Sempra, CenterPoint and Southern, where results and guidance improved; CONFIRMS at PSEG. The rate explanation fails too: the 10-year Treasury yield fell 7 basis points to 4.6% on 7 August after payrolls shrank 23,000, during the week these stocks fell. Does valuation justify it? INCONCLUSIVE across the group — Sempra's compression is real, CenterPoint's is deserved, Dominion's expansion is not earned. Merchant generators Vistra and NRG, paid on power prices rather than allowed returns, fell far harder over the same month, and Fitch Ratings moved its North American utility outlook to "deteriorating" on affordability grounds in June.

The setup

Where it stands — A Texas regulatory freeze, not interest rates, repriced the two Texas-exposed utilities while their reported earnings and capital plans rose.

Would confirm — ERCOT restarts batch-zero approvals with CenterPoint's 14 gigawatts and Oncor's queue largely intact after the audit.

Would invalidate — The audit strikes a large share of the 474-gigawatt queue, or Sempra cuts its $4.80-$5.30 guidance.

Watch next — Third-quarter results in late October, and the KKR sale of 45% of Sempra Infrastructure closing this quarter.

Valuation — Sempra 16.4x forward against 23.0x trailing and 33.8x trailing in May; CenterPoint 21.3x forward, the group's highest.

The two companies that make the power electronics inside a rooftop solar system — the boxes that turn direct current from the panels into usable alternating current — are being priced as one story, and they are not one story.

SolarEdge, which sells DC-optimised inverter systems to installers, distributors and engineering contractors from Herzliya, Israel, delivered on 5 August the best quarter it has reported in three years. Revenue of $346.2m rose 19.6% year on year and 11.5% sequentially. Gross profit nearly tripled. Reported gross margin went from 11.1% a year ago to 27.5%, a sixth consecutive quarter of expansion, and operating margin narrowed from -39.9% to -4.6%. Non-GAAP operating income of $10.2m was its first profit in nearly three years.

The stock fell 28.1% that day on roughly 13.3m shares, about five times normal volume. The reason was the outlook: third-quarter revenue of $310–340m, whose $325m midpoint sat roughly 12% below the analyst consensus of about $368–372m. Management attributed the sequential decline to European seasonality of about $15m, continued softness in US residential, and the absence of the usual third-quarter uptick. Two further details matter: reported gross margin included $13.3m of tariff refunds under the International Emergency Economic Powers Act, with another $11.5m received in July that is not assumed in the guide — flattering the margin by roughly four points — and US residential revenue of $154.9m actually fell 2% sequentially while European revenue of $154.4m rose 36% and more than doubled year on year.

Verdict on SolarEdge: the business CONTRADICTS the move. Revenue is accelerating, margins are expanding, and consensus 2027 revenue of $1.495bn (+11.7%) is rising, not falling. A 12% cut to one quarter's revenue path is real, but it is one quarter.

Enphase is the mirror image

Enphase, the Fremont, California maker of semiconductor-based microinverters that convert power at each individual solar module, plus batteries and monitoring software, has the demand problem SolarEdge only partly has. Revenue fell 19.6% year on year to $291.9m in the second quarter, after 20.6% and 10.3% declines in the two prior quarters. The cause is policy: the Section 25D federal residential solar credit, worth 30% of system cost to homeowners who buy their own systems, expired outright at the end of 2025 rather than stepping down, and BloombergNEF data projects 2026 US rooftop installations at their lowest since 2020 — though California is still set to grow 17% and Florida 62%.

Yet Enphase's gross profit rose 2.7% and operating income 39.2% year on year. Non-GAAP gross margin reached 46.8%, up from 43.9%, and third-quarter revenue is guided to $290–320m — of which roughly $75m is safe-harbour shipments, buyers pulling forward purchases to lock in credits, not end demand. Verdict: INCONCLUSIVE. Shrinking sales, expanding profits, and a quarter of the guide borrowed from the future.

What you pay now

SolarEdge has no earnings anchor — trailing price-to-earnings is -7.0x. On sales, the right lens for a loss-maker, it trades at 1.45x trailing and 1.44x forward, against 2.94x when this desk last examined it in May at a $3.76bn market value versus $1.93bn today. It holds $601.6m of cash against convertible notes due 2029 carried at $331.6m, generated $3.1m of free cash flow in the quarter and expects positive free cash flow for the year.

Enphase, at a $5.53bn market value, trades at 41.0x trailing and 20.6x forward earnings — but its forward price-to-sales of 4.67x sits above trailing 4.16x, because forward revenue is smaller. The earnings multiple compresses on cost cuts and tax credits, not growth, on a 2.8% free cash flow yield. Brokers cut targets after the print — Susquehanna to $39, Citi to $40, Wells Fargo to $44 — bracketing the current price.

Valuation verdict: SolarEdge, a possible dislocation; Enphase, a justified de-rating.

On the tape, the pair did not move together at all. Over the 30 days to 7 August Enphase fell 2.7% while SolarEdge fell 39.4%, and Enphase was flat across the three sessions that took SolarEdge down 34.9%. Enphase's 50-day average crossed below its 200-day on 30 July, the session after its own results; SolarEdge's had not yet registered its collapse a week later, a lag in the averages rather than evidence of a broadening decline.

The setup

Where it stands — SolarEdge's fundamentals improved while its shares halved their sales multiple; Enphase's shares held while its revenue shrank a third straight quarter.

Would confirm — SolarEdge third-quarter revenue printing at or above the $340m top of guidance with non-GAAP gross margin inside 22–26%.

Would invalidate — SolarEdge free cash flow turning negative for the full year, or European revenue falling back below $120m.

Watch next — Third-quarter results from both companies in late October 2026, and Enphase's fourth-quarter guide net of safe-harbour shipments.

Valuation — SolarEdge 1.45x trailing / 1.44x forward sales versus 2.94x in May; Enphase 41.0x trailing / 20.6x forward earnings.

Three of these nine companies raised their full-year revenue outlook within thirty-six hours of each other, and a fourth reported the first profitable year in its history while telling investors that next year would be slower. That is the actual news inside a month that also happened to include a violent rotation out of semiconductor stocks and into anything software-shaped.

What the four that reported actually said

Dynatrace, which sells software that watches applications, servers and cloud infrastructure for large enterprises and flags failures before users notice, opened the week on 5 August. Annual recurring revenue (ARR) — the run-rate value of subscriptions in force — reached $2.136bn, up 17%, with organic net-new ARR up 41% and new-logo ARR growth above 160%, exceeding the high end of guidance on every metric. It landed 122 new customers at roughly $285,000 each, and log management now runs at $200m of annualised consumption, double two quarters ago. The company raised fiscal-2027 revenue guidance to $2.306-2.320bn and adjusted earnings to $1.97-1.99 a share. The blemish: reported revenue growth has decelerated three quarters running, to 16.2%, and net revenue retention sits at 110%.

Twilio, whose programming interfaces let developers embed text messages, voice calls and email inside their own apps and which bills by the message rather than the seat, reported record quarterly revenue of $1.5bn, up 22%, and raised full-year revenue growth guidance to 18-18.5% from 14-15%. Organic growth accelerated to 17%, dollar-based net expansion improved to 116% from about 110%, and free cash flow was $353m. Messaging revenue grew 28%, voice 20%. This is the cleanest business acceleration in the group — four straight quarters of faster growth, from 14.3% to 22.0%.

JFrog, which sells Artifactory, the repository where software teams store and version the compiled components of their applications, beat estimates by 12.5% on earnings with revenue up 28.7% to $163.8m and cloud revenue up 53%. Customers paying more than $1m a year rose 59% to 97, net dollar retention hit 121%, and non-GAAP operating margin improved 470 basis points to 19.9%. The unfinished part: 47% of revenue is still self-managed software installed in customers' own data centres, and it grew 9%.

Atlassian, the Sydney-based maker of Jira and Confluence project-tracking tools sold per user seat, beat with adjusted earnings of $1.87 against $1.50 expected and cloud revenue growth accelerating to 31%. Fiscal-2026 revenue grew 26% to $6.572bn with a first positive full-year operating income of $10.4m. But it guided fiscal 2027 to 18% subscription ARR growth, down from 23%, citing macroeconomic and geopolitical uncertainty.

Verdict A on the business: CONFIRMS for these four. Growth accelerated at three of them and guidance rose at three.

The five that have said nothing since spring

Nutanix, whose software bundles virtualisation, storage and networking so companies can run private clouds on ordinary servers, last spoke on 27 May; its April quarter grew revenue just 10%, decelerating from 13.5% two quarters earlier, though operating income rose 51%. Samsara, which bolts dash cameras and sensors onto customer truck fleets and sells the data back as software, last reported 4 June: revenue up 30.5%, but a GAAP operating margin of 1.5%. BILL, which automates bill payment for small businesses and their accountants, last reported 7 May with revenue up 13.5% and its first positive operating income, $4.6m. Veeva, which sells document and trial software exclusively to drugmakers, grew 16.3% with a 30.9% operating margin — the best in the group — and spoke on 3 June. Salesforce, the largest name here at $158bn, grew 13.3% with operating income up 25%, its third straight quarter of acceleration.

For these five, Verdict A is INCONCLUSIVE: the last data is two to three months old, and none of it landed inside the price window.

Valuation splits the group in three

Verdict B: CONTRADICTS at the top end, CONFIRMS at the bottom. JFrog trades at 93.9x forward earnings, 17.1x forward sales and 22.0x forward gross profit on a 1.6% free-cash-flow yield — up from roughly 16x sales nine sessions earlier. Samsara is at 57.4x forward earnings and 11.7x forward sales while consensus has growth slowing from 25.9% to 19.7%. Twilio's forward multiple expanded from about 33x to 42.2x on the guidance raise. At the other end, Salesforce is 13.6x forward and 22.2x trailing earnings with a 9.3% free-cash-flow yield, sitting at only the 38th percentile of its own year's range; BILL is 14.3x forward with an 8.0% yield. In between: Dynatrace 24.7x forward, Veeva 25.4x, Atlassian 24.5x, and Nutanix at 28.6x forward against 61.4x trailing — with consensus modelling 12.8% then 12.5% revenue growth, meaning no acceleration from Broadcom's VMware repricing is priced in. That is consistent with independent work finding Nutanix wins displacement deals at 40-48% discounts and no vendor demonstrating mass displacement of regulated production workloads.

The tape, and when it moved

Eight of nine names rose over the past month and all nine over three months, so this is not two earnings pops carrying an average. But the dates are unforgiving: the four sessions to 28 July supplied 8.8 percentage points of the average and 7 August supplied 10.2 — together essentially the whole month. On 28 July the SOXX semiconductor fund fell 4.80% while enterprise software rallied, one day after the Dow rose more than 500 points as investors rotated out of chips. Across July, Workday rose 31% and Adobe 22% on rotation rather than company news. Nor is this a rate story: the Fed has cut nothing in 2026 and the target range is 3.5-3.75%, with markets pricing possible hikes, which removes the float-income tailwind from BILL's case.

One durable pattern survives the decomposition. Over twelve months the four billed by usage — Twilio, Dynatrace, JFrog, Nutanix — averaged +44.7% while the five billed by seat or subscription averaged -8.1%. Over the past month that ordering reverses. The pricing-model thesis is a year-long fact, not a July one.

The setup

Where it stands — Four of the nine reported strong quarters into the move; five rose on rotation with data last updated in May.

Would confirm — Nutanix reporting revenue growth above 12% and Samsara holding ARR growth near 30% when they next report.

Would invalidate — Dynatrace's ARR growth slipping below 15% or JFrog's cloud growth falling under its raised 41-43% baseline.

Watch next — Nutanix, Samsara, BILL, Veeva and Salesforce all report fiscal quarters between late August and early September.

Valuation — JFrog 93.9x forward earnings and Samsara 57.4x against Salesforce 13.6x forward, 22.2x trailing.