DK Street Journal

A guidance raise that shrank the bottom line

Hubbell, the 138-year-old Connecticut maker of the cutouts, connectors, arresters and distribution transformers that electric utilities hang on poles and bolt into substations, told investors on 28 July that its sales would grow 16–18% this year. Its previous forecast was 8–11%. Reported profit went the other way: net income for the June quarter fell 1.6% to $240.4m, on revenue up 15.3% to $1.71bn.

That gap is a balance sheet, not a demand signal. Hubbell closed a $3.0bn purchase of NSI, a connector and fastener maker, in early June. NSI contributed $35m of revenue for the stub month and should add about $0.20 of earnings per share this year, but it brought roughly $170m of additional annual interest expense and took leverage to 2.9x. With restructuring spending on top, gross margin fell 139 basis points year over year to 35.8% and operating margin fell 230 basis points.

Underneath, both halves accelerated. Utility Solutions — the distribution, transmission and substation arm — sold $1.03bn in the quarter, up 6% organically, with grid infrastructure up 7% and a segment book-to-bill of 1.2x that pre-sells much of the second half. Electrical Solutions grew 18% organically, with data-center revenue up about 65%. Management raised adjusted earnings guidance to $20.25–$20.55 a share.

The bottleneck that sets the price

This rung of the build-out earns its margin from scarcity, not from compute. Large power transformers are quoted at about 128 weeks, with medium-voltage 15kV switchgear at 52–80 weeks — two to four times pre-pandemic norms — because grain-oriented electrical steel has roughly doubled since 2020 and copper trades near $9,800 a tonne. The result is pricing power: distribution transformer prices are up 78–95% since 2019, with medium-voltage switchgear up about half. Hubbell has taken roughly a point of price in April and half a point in July, for 3–4 points across the year, and is adding about $25m of capacity a quarter on capital spending of $175–190m against $155m last year. It is quoting a $1.5bn high-voltage transmission pipeline over ten years, with 550kV product shipping in the second half and 765kV work starting in 2027.

Powell has the orders and none of the credit

Powell Industries, the Houston builder of custom switchgear and factory-assembled modular substations, is the same trade at four times the multiple. Its June quarter brought record orders of $934m against $362m a year earlier — three dollars booked for every dollar shipped — and a backlog of $2.4bn, more than twice fiscal 2025 revenue. The mix moved decisively: commercial and other industrial revenue rose 54% and electric utility 18%, while petrochemical fell 49%. Gross margin has climbed for three straight quarters to 30.6%. The shares are down 29% over three months, which has cut the price paid per dollar of trailing gross profit from about 30x in late May to 20.7x.

Eaton, the $163bn power-management group that is the scaled comparator here, supplies the reason to take the order books seriously: announced US data-center projects total 307 gigawatts, equal to fifteen years of construction at 2025 build rates, of which only about a fifth converts near-term. Eaton's own 2030 plan assumes 17% data-center growth while it is currently realizing 65%.

Atkore, the conduit and cable-tray maker, has stopped being a read on any of this. Its gross margin did finally inflect, to 22.2% from 18.6% the prior quarter on 9% organic volume, but Prysmian agreed on 2 August to buy it for $95.00 a share in cash, and the stock has since sat about 1.5% below that price. Conduit prices no longer set it.

The last leg was rates

Between 17 and 21 August the whole layer fell together — Powell 9.3%, Eaton 7.8%, Hubbell 6.9% — as investors rotated out of debt-heavy AI infrastructure names around the 19 August Federal Reserve minutes. Hubbell's 50-day average slipped below its 200-day on 30 July even as the shares rallied from $476 to $518 in the weeks after the print, and it has given that back since. The durable risk is not demand but capacity: Korean transformer makers and Western incumbents are all expanding, and analysts there put margin normalization around 2028–2029, when new lines land.

At 27.7x trailing and 23.0x forward earnings, Hubbell is not cheap against the market. Against its own rung it is the least expensive thing in it.

The setup

Where it stands — Hubbell's sales and orders are accelerating while reported profit is held down by acquisition interest and restructuring. Would confirm — Utility Solutions organic growth at or above 6% with book-to-bill still above 1.0x in the third quarter. Would invalidate — Full-year sales guidance cut back below 16%, or adjusted earnings falling under $20.25 a share. Watch next — Hubbell's third-quarter results in late October; Powell's fiscal fourth quarter in early December. Valuation — Hubbell at 11.3x trailing and 10.3x forward gross profit, against Eaton's 13.8x and Powell's 20.1x.

AXT, a Fremont, California maker of compound-semiconductor substrates — the polished crystal wafers on which laser and radio-frequency chips are grown — reported a June quarter that looked nothing like its own recent history. Revenue reached $47.6m, up 164.8% from a year earlier. Indium phosphide, the material of choice for the high-speed lasers inside optical transceivers, contributed a record $30.7m of that. The company guided the September quarter to roughly $66m.

The margin move is the more striking number. Gross margin was 7.97% in the June 2025 quarter and 44.9% in the one just reported, the sixth consecutive quarter of expansion. Trailing-twelve-month gross profit roughly doubled in a single quarter, from $20.5m to $40.4m — the June quarter alone produced more gross profit than all of 2025.

Three suppliers, and a permit office

What makes the operating leverage durable is scarcity. Sumitomo Electric, JX Metals and AXT together control 80-90% of the indium phosphide substrate market, and both Japanese rivals are spending heavily to expand — JX has flagged up to ¥120bn of capital expenditure. For now AXT's output is fully allocated; management says it is declining orders it cannot serve, and customers are paying to reserve capacity. Lumentum, the optical-component maker, signed a supply agreement running to the end of 2031 with an initial $43.5m deposit and a second $43.5m payment due in 2028. Coherent prepaid $25.4m, Casella $22.3m. Backlog exceeds $100m and is growing faster than shipments, stretching visibility to three or four quarters from the historical one or two.

The binding constraint has not been demand but Beijing. AXT manufactures through its Tongmei subsidiary in China, and gallium- and indium-bearing shipments outside China require export permits. That is why the fourth quarter of 2025 came in below guidance and the first quarter of 2026 beat it — permit issuance moved, not orders. Management now calls permits predictable rather than a showstopper, and the $66m guide counts only volumes already licensed or exempt.

The price has come down and is still demanding

None of this has produced a settled share price. AXT closed at $36.97 on 29 July, ran to $93.72 by 17 August, then fell 24.8% in four sessions to $70.48. Against May it is down 41.8%. Even so, the shares change hands at 87.6 times trailing gross profit and 16.2 times forward sales. Reported earnings only just turned positive, which makes the trailing price-to-earnings ratio useless; on consensus it is 81.5 times 2026 earnings and 31.8 times 2027 — and reaching 2027 requires revenue to more than double again, to $460.6m. The multiple has roughly quartered since May. It still prices near-flawless execution against two well-capitalized Japanese rivals.

Wolfspeed: the AI line is real and too small

The contrast within compound semiconductors is Wolfspeed, the Durham, North Carolina silicon-carbide maker that emerged from a prepackaged Chapter 11 last September with debt cut from about $6.7bn to $2bn. Its case for AI relevance is silicon carbide in 800-volt direct-current rack power supplies. On 19 August it reported June-quarter revenue of $149.6m, down 24.1% and roughly a third short of consensus, with gross margin at -24.6% and operating cash flow of -$54m. AI data-center revenue more than doubled over the fiscal year and grew about 20% sequentially — off a base too small to offset softening automotive demand, which remains roughly 70% of silicon-carbide consumption. Consensus has revenue falling another 15.2% next year, which is why Wolfspeed's forward price-to-sales ratio, at 2.10, sits above its trailing 2.03. No earnings or gross-profit multiple exists: gross profit is negative.

What actually moved the month

Every discrete- and power-semiconductor name in this group fell over three months, from Vishay's 26.4% to Wolfspeed's 61.3%. The common cause is the discount rate: the 30-year Treasury yield touched 5.33% in mid-August, its highest since 2007. The sharpest recent faller, fiber-laser maker nLIGHT, dropped for a different and specific reason: it postponed about $17m of shipments on heightened Chinese inspections of dual-use goods, despite record quarterly revenue — the same export machinery that gates AXT.

Vishay, the broad-line discrete and passive component maker, is the group's quiet diverger. Adjusted revenue of $919m grew 20.5% year on year, adjusted gross margin reached 22.6%, and the September guide of 24% arrives a quarter ahead of plan against a 30.3% margin earned in 2022. Book-to-bill was 1.32 and backlog rose 18% in the quarter to $1.9bn. The shares trade about 38% below the $50 at which the company sold stock in June, at 17.4 times 2027 consensus earnings.

The setup

Where it stands — AXT's numbers have inflected hard; the share price has retraced 41.8% from May yet still embeds years of triple-digit growth. Would confirm — September-quarter revenue lands at or above the $66m guide with gross margin holding above 40%. Would invalidate — Chinese export permits tighten again and AXT guides December revenue below September's. Watch next — AXT's September-quarter report, due late October, plus its stated exit-2026 indium phosphide capacity of about $60m a quarter. Valuation — 87.6x trailing gross profit and 50.5x forward; 81.5x 2026 consensus earnings, 31.8x 2027.

Itron, which sells smart electricity, gas and water meters and the radio networks that carry their readings back to utilities, told investors on 28 July that its gross margin had reached the highest level in the company's history. Revenue fell for the fourth quarter running.

That combination is the whole story. Gross margin hit 41.0% on a reported basis in the second quarter, up from 34.1% two years earlier, so gross profit rose 3.1% to $230.6m even as revenue dropped 7.2% to $562.9m. Adjusted earnings of $1.59 a share beat consensus by 23%, and the shares rose about 20% in the session. Management raised full-year earnings guidance to $6.30–$6.50 a share from $5.75–$6.25 while narrowing the revenue range to $2.37–$2.41bn — roughly 1% growth. It was an earnings raise, not a demand raise.

Where the margin comes from

Itron is the incumbent in advanced metering infrastructure, holding roughly 34% of the North American market against Landis+Gyr's 32%, and about 63% of installed network endpoints. That installed base is the asset. Competition in the industry has moved away from commoditized meter hardware toward cellular connectivity, edge analytics and managed services carrying higher lifetime margins — which is precisely the shape of Itron's quarter. Its Outcomes software segment grew 13% with annual recurring revenue up 21% to $417m, while Networked Solutions revenue fell 17% on deployment timing and meter hardware slipped 3%. A new resiliency unit added $16m at roughly 75% gross margin. Licensed applications reached 28 million endpoints, more than half again the prior year, with platform expansions at the Los Angeles Department of Water and Power and the Sacramento Municipal Utility District.

Management says the 2027 company margin target has already been met and that the gain is structural mix rather than volume.

What the order book does not show

The demand backdrop is not in doubt. S&P Global Market Intelligence forecasts a record $1.295trn of US utility capital spending over 2026-2030, with data centers adding around 125GW of load. Itron's own examples fit: a West Coast utility deferring over $1bn of transformer upgrades through coordinated electric-vehicle charging, a Southeast utility cutting outage duration 12-15% with distributed intelligence.

None of it is in the bookings yet. Total backlog ended the quarter at $4.4bn against $4.5bn a year earlier, and second-quarter bookings of $550m against $562.9m of revenue put book-to-bill at 0.98. Operating income fell 2% and net income fell 22%. Investors are being asked to pay for a margin structure, not a volume inflection.

The price has not yet demanded much for it. Itron trades at 4.74x trailing gross profit and 4.58x forward, against 4.89x in late February and 3.93x in May, with a forward price/earnings ratio of 15.5 below its trailing 16.4, enterprise value at 9.4x EBITDA and a trailing free-cash-flow yield of 8.8%. The stock remains 20% below where it traded a year ago.

The two companies filed alongside it

Digi International, which sells cellular routers, embedded modules and console servers for connecting industrial machines, is the reverse case: revenue growth accelerated for three straight quarters to 29.0%, gross margin reached 64.8%, operating income rose 53%, and recurring revenue hit a record $191m. But its own target of $200m of that revenue by 2028 implies growth collapsing to about 2% a year after this one, much of the recent gain came from buying Jolt Software and Particle, and its data-center exposure — Opengear console servers sold to specialist cloud operators — is excluded from guidance. At 8.72x trailing gross profit, up from 6.70x six months ago against 14% gross-profit growth, and a trailing price/earnings ratio of 57.5, the multiple has run ahead of the numbers.

Montrose Environmental, an air, water and soil testing and remediation firm, renamed itself Onterris in April and now trades under the ticker ONT; price histories filed under the old symbol stop on 18 June. On the live listing the stock fell 33.8% on 6 August after second-quarter revenue dropped 20.4% to $186.7m and full-year guidance was cut roughly $80m — about $45m of pass-through work, $40m of emergency-response revenue with environmental event activity at historic lows, and $20m deferred by temporary air-permitting waivers. Margins actually improved, with the full-year EBITDA margin midpoint up 150 basis points. The board has launched a strategic review covering acquisition interest and adopted a shareholder rights plan triggering at 15% ownership.

One shared pressure links the first two. DRAM prices rose 80-90% in the first quarter as manufacturers shifted capacity to high-bandwidth memory for AI servers, and Samsung has warned the shortage will spread through mature nodes used in industrial devices. Itron says it is managing memory pricing proactively; Digi says the shortage is pushing customers to place orders early. Both statements describe the same cost arriving.

The setup

Where it stands — Itron is earning record margins on shrinking volume, with backlog and book-to-bill both pointing the other way. Would confirm — Third-quarter bookings above revenue, lifting book-to-bill over 1.0 and backlog back above $4.5bn. Would invalidate — Gross margin slipping below the roughly 40% full-year target as memory costs pass into meter builds. Watch next — Itron's third-quarter results, due late October, and Digi's fiscal fourth quarter in November. Valuation — Itron at 4.74x trailing gross profit, 4.58x forward, versus 4.89x in February; forward P/E 15.5 against trailing 16.4.

Five American makers of heating, cooling and water equipment reported summer quarters within nine days of one another, and every one of them spent part of the call talking about data centers. What separated them afterward was not how much artificial-intelligence work they had won. It was whether that work was showing up in gross profit yet.

Carrier won the orders and lost the margin

Carrier Global, the Palm Beach Gardens maker of residential and commercial air conditioning, transport refrigeration and Kidde fire safety, spun out of United Technologies in 2019, delivered the loudest demand quarter in the group. Total orders rose 40%. Backlog passed $8bn, and roughly 40% of it is data-center work. Commercial heating-and-cooling orders rose about 65%, and data-center orders came in at four times the prior year. Management raised its disclosed 2026 data-center revenue to about $2bn from $1.5bn — a second consecutive doubling, and near 9% of guided sales — with a 2.5-megawatt coolant distribution unit launching this quarter and a 5-megawatt version due at year-end.

And reported gross profit fell 2.3% year over year, on revenue up 3.9%. Gross margin compressed 173 basis points to 27.2%. The cause is specific and mechanical: expanded Section 232 tariffs of up to 50% now apply to the full value of imported equipment rather than only its metal content, and Carrier's pricing lagged the April tariff date by weeks. New-construction work also carries thinner margin than replacement. The shares fell 5.1% the session after the print despite beats across the board and a raised guide, and have drifted since. At 9.30x trailing gross profit against roughly 10.0x six months ago, part of that de-rating is honest — the denominator is shrinking.

Johnson Controls converted it

Johnson Controls, the Cork-domiciled building-systems group that installs and services chillers, controls and fire-detection systems, is doing what Carrier has not yet done: turning campus construction into margin. Organic revenue rose 10% in the June quarter, adjusted operating margin widened 260 basis points to 17%, and backlog reached a record $21.0bn, up 32% organically. Reported revenue growth has accelerated four quarters running, from 3.1% to 9.3%. Data-center revenue is tracking to a high-teens share of the fiscal year, and management expects roughly a third of the company within three to five years. It holds over 20.9% of the data-center chiller market, the largest single share, and its coolant-distribution pipeline exceeds $1bn with Nvidia certification in hand. The shares are up 37% over twelve months, yet price per dollar of trailing gross profit is 9.50x, below the 9.98x of six months ago. Earnings did the work.

SPX is the divergence

SPX Technologies is the small one: a $10.3bn maker of Marley cooling towers, industrial air movement and boilers, plus a detection-and-measurement arm that locates buried pipe. It does not sell transformers — that business went in 2021. Second-quarter revenue rose 22.9% and operating income 34.6%. Cooling backlog reached $919m, up 59% organically, on a 1.4x book-to-bill. Management raised 2026 data-center revenue to $430m from $300m, against $150m two years ago, and lifted eventual capacity to $1.1bn.

The shares are down 15.0% over six months. Price per dollar of trailing gross profit has compressed from about 13.3x to 10.4x while trailing gross profit grew 8.6% — a 22% de-rating paid for by nothing in the numbers. Forward earnings are priced at 24.6x against 36.6x trailing, a 33% expected step-up. The stock jumped 10.2% on the Q2 release and has since given the entire move back.

The controls weren't controls

Watts Water Technologies, a maker of valves, backflow preventers and flow control for plumbing systems, was supposed to have no compute exposure. On 6 August it said data-center sales tripled year over year, now 8% of year-to-date revenue versus 3% for all of last year, and doubled its addressable estimate to $2bn. Content runs $25,000 to $100,000 per megawatt. It raised organic growth guidance to 8–11%. It is the best performer in the group — and the only one whose multiple genuinely expanded, to 9.55x gross profit from 8.23x in May.

That leaves A.O. Smith, the 1874 Milwaukee water-heater maker, as the sole business here with no compute load. Its revenue fell 0.7%, operating income fell 18.6%, Chinese sales fell 28% in local currency, and it narrowed guidance on soft residential demand. The stock rose 5.9% on the month, at 5.97x gross profit and 16.9x forward earnings — the cheapest in the group by a distance.

The backdrop explains the shape. Money rotated violently out of the thermal pure-plays over three months — Vertiv down 23.7%, Modine 21.0%, Lennox 17.4% — and into diversified books, before the whole complex sold off together over four sessions to 21 August as investors reassessed whether near-term revenue justifies AI capital spending. Flat is not calm here.

The setup

Where it stands — The two largest disclosed data-center books in commercial HVAC, at Carrier and SPX, carry the group's weakest six-month share performance. Would confirm — SPX's third-quarter cooling backlog holding above $900m with organic growth above 15%. Would invalidate — Carrier's gross margin failing to recover from 27.2% as tariff pricing catches up in the second half. Watch next — Johnson Controls' fiscal fourth quarter in early November, the first with coolant-distribution units shipping. Valuation — SPX at 10.4x trailing gross profit versus 13.3x six months ago; forward earnings 24.6x against 36.6x trailing.

Three of the four listed companies that sell software for running corporate customer relationships have now put a number on their artificial-intelligence business. All three also reported a lower gross margin than a year earlier. That is the whole argument in one line: an agent answering a customer question consumes computing power, that cost sits in cost of revenue, and none of these vendors has yet priced it back out.

The exception is the largest. Salesforce, whose Customer 360 suite plus Slack, Tableau and MuleSoft runs sales pipelines and service desks for 83,000-employee-scale enterprises, earned a gross margin of 76.92% in its April quarter against 76.96% a year before. It did that while disclosing $1.2bn of annualized Agentforce revenue, up 205%, and 28.6 trillion tokens processed.

What the month actually was

The group's roughly 23% equal-weight gain over 30 days is two things stacked. About half arrived in one session on 28 July, when money rotated out of semiconductors into oversold enterprise software — Salesforce up about 7%, ServiceNow 8%, Workday 10%, on no company news. The rest is a grind: across the 19 sessions since, Salesforce rose 11 times and fell 8, with no gain above 5.1%. On market value the group is Salesforce — $170.5bn against $12.3bn, $5.9bn and $3.5bn — so its 27.7% month effectively is the group's. The backdrop is a valuation unwind, not new demand: the Morningstar US Software Application Index fell about 27% from October to mid-July while the broad market rose, and roughly $2 trillion of software value has been removed on the argument that agents do the work seat licences pay for.

The seat, the meter and the outcome

Salesforce is trying to graft consumption onto a subscription base and is running three prices at once for the same product: $2 per conversation at launch, Flex Credits at $500 per 100,000 credits, and a flat per-user digital-labour licence from $125 a month. That is a company that has not settled on a billing unit. The underlying business is nonetheless accelerating — 8.6%, then 12.1%, then 13.3% revenue growth — with operating margin at 21.80% from 19.76% and current remaining performance obligation of $33.6bn, up 14%. Agentforce is still only about 3% of a $41.5bn revenue base, so the flat gross margin proves less than it looks: inference cost has not landed because the volume has not arrived.

HubSpot, which sells an all-in-one marketing, sales and service platform to mid-market companies on seats and contact tiers, is the contradiction. It added 7,000 net customers in the June quarter against 9,000-10,000 expected and guided the second half down to 5,000-6,000 a quarter. Net revenue retention was 102%, down a point, with average subscription revenue per customer of $11,800 — up only 2% in constant currency. Gross margin fell 159 basis points. Its agents are being priced on outcomes — $0.50 per resolved conversation, $1 per qualified lead — and adoption is far ahead of billing. The shares fell 18% on 6 August, the worst session in the company's history, and had recovered the entire loss by 19 August.

Braze, which orchestrates cross-channel messaging for consumer apps and bills on messages and data points rather than seats, is the meter in its purest form. Revenue grew 30.2% in a fourth straight quarter of acceleration, with retention at 110% and customers above $500,000 of annual revenue up 33%. Gross profit grew only 24.7%: the meter fills at 65.72% gross margin against 68.62%. Its own guidance implies about 22% growth next quarter.

NICE, which sells the CXone contact-center platform and Actimize financial-crime software and is not a customer-relationship vendor at all, is the control that decides the question. It has the most AI attach of any of them — $362m of AI and self-service annualized revenue, up 52%, now 15% of cloud — and it is the most damaged by it: revenue grew 9.5%, gross profit 5.1%, and GAAP operating income fell 31.5%. It reported a beat on 5 August and the shares fell 6% in two days.

What the prices assume

On price per dollar of trailing gross profit — the lens that normalizes a 77%/82%/66%/64% margin spread — Salesforce is at 5.13x trailing and 4.76x forward, above 4.62x three months ago but below 5.68x in February, on a 8.6% free-cash-flow yield and 14.70x forward earnings against 24.01x trailing. Consensus still models fiscal 2028 revenue growth of 9.4%, slower than the company just delivered; JPMorgan initiated at Overweight on 13 August with a $250 target. HubSpot at 4.28x has expanded 13% since May in the window its guidance came down, yet sits 8.5% below February. Braze is the stretch: 3.94x in February, 5.23x in May, 6.72x today — a 71% re-rating on no new disclosure since 27 May, at a 1.9% free-cash-flow yield. NICE, the cheapest at 2.89x and 9.03x forward earnings, is the weakest performer.

The durable risk sits behind all of it. Klarna publicly abandoned Salesforce for an assembled AI stack, the second such defection after Workday, and rival platforms are now indexing Salesforce and Slack data into their own cross-application context graphs — Atlassian's latest shareholder letter treats Salesforce data as an ingested source rather than a partner. Coordination value migrates to whoever holds the graph, not whoever holds the record.

The setup

Where it stands — Every vendor here that monetizes AI has lost gross margin doing it; the one that hasn't has barely started. Would confirm — Salesforce reporting Q2 constant-currency cRPO growth at or above the 14% Q1 pace with gross margin near 77%. Would invalidate — Salesforce gross margin falling below 76% as Agentforce volume scales, matching the pattern at NICE and Braze. Watch next — Salesforce fiscal Q2 on 26 August 2026; Braze fiscal Q2 in early September. Valuation — Salesforce 5.13x trailing and 4.76x forward gross profit, versus 5.68x in February; Braze 6.72x, versus 3.94x.

The mix, not the boom

Hewlett Packard Enterprise sells servers, storage and high-performance computing to enterprises, and — since closing its $14bn all-cash purchase of Juniper Networks — one of the largest enterprise networking franchises in the industry. In its quarter to April it converted $10.7bn of revenue into gross profit at a rate of 36.5 cents on the dollar, against 27.6 cents a year earlier. Revenue grew 41%. Gross profit grew 86%.

The reason is largely what HPE is not selling. AI systems orders were $1.8bn in the quarter, a rounding error beside Dell's book. Networking revenue was $2.7bn at a 13.3% operating margin, and management raised full-year networking growth guidance to 72-75%, lifted non-GAAP earnings guidance to $3.35-$3.45 a share, and guided to at least $3.5bn of free cash flow. Free cash flow in the quarter was a record $915m. The company said Juniper integration is running ahead of plan and that it sees no sign customers are pulling AI orders forward.

Dell is the AI-server business, and pays for it

Dell Technologies, the Round Rock, Texas group run by Michael Dell that sells servers and storage to enterprises and PCs to everyone else, is the volume leader in branded AI servers with roughly a fifth of the market. Its April-quarter revenue reached $43.8bn, up 87.5%. AI-optimized server revenue was $16.1bn, up 757%; Dell booked $24.4bn of AI orders and closed the quarter with $51.3bn of backlog, lifting full-year guidance by about $27bn.

Gross margin fell to 17.75% from 21.11%. The named mechanism is memory. Server DRAM contract prices rose 90-95% quarter on quarter in the first quarter of 2026 and a further 58-63% in the second, with NAND up 70-75%, because the three memory makers diverted capacity to high-bandwidth memory for AI accelerators. Dell raised list prices about 17% across its line on 30 March, after Lenovo reissued quotes 10-15% higher on New Year's Day. The squeeze is easing rather than ending: TrendForce expects server DRAM up 13-18% in the third quarter, partly because US cloud providers have signed multi-year agreements capping supplier increases.

Below the gross line Dell held together. Infrastructure Solutions operating margin rose 80 basis points to 10.5% even as AI servers grew nearly eightfold, and group operating income nearly tripled to $3.66bn. What did not happen is the attach story: storage revenue was $4.33bn, up only 8%, far slower than the racks it is supposed to ride alongside. The PC business, at $14.61bn and up 17%, is a passenger, not the driver.

What the share price has already paid

Dell's shares closed at $122.27 on 20 February and $434.78 on 20 August. Trailing gross profit over the same stretch grew roughly 19%, to $25.6bn. The price per dollar of trailing gross profit therefore went from about 3.9x in February to 7.0x in May to 11.4x now — four fifths of the move was re-rating. Consensus has run past the company: analysts model $173.1bn of revenue and $18.80 of earnings per share this fiscal year, against the $138-142bn Dell itself guided in May. On those estimates the shares are 23.3x forward earnings versus 34.3x trailing.

HPE's advance is smaller and better funded by its own numbers. The shares roughly doubled since February; the multiple of trailing gross profit went from about 2.9x to 5.5x while gross profit itself grew 30%, to $12.8bn. HPE trades at 15.6x forward earnings against 49.0x trailing, on a 5.6% trailing free-cash-flow yield. The debt is the offset: gross borrowings above $18bn after the Juniper deal, including $1.35bn of 7.625% mandatory convertible preferred, and a 2.0x net-debt-to-EBITDA target management is not expected to reach by 2027.

The two controls

Celestica, the Toronto contract manufacturer that designs and builds full AI racks, switches and storage for hyperscalers, is the test of whether cloud buyers are bypassing the brands. It is not losing: second-quarter revenue was $4.70bn, up 62%, full-year guidance went to $20.5bn, and OpenAI custom racks and AMD's Helios platform were named as multi-billion-dollar 2027 programs. Yet its multiple of trailing gross profit compressed from 26.3x in May to 18.9x, because on 5 August it sold 9.68m shares at $310, 14.5% below the prior close, raising about $3bn for capex and working capital. Its trailing free-cash-flow yield is 1.52%; its top three customers are roughly 65% of revenue. The de-rating is the cost of funding growth, not lost share.

Super Micro is the sharper control. The San Jose builder of high-density GPU systems — the whitebox alternative to a branded rack — reported June-quarter gross margin of 17.5%, up from 9.9%, on $11.1bn of revenue. That is now within a rounding error of Dell's 17.75%. Super Micro trades at 6.1x trailing gross profit and 10.9x trailing earnings, and its shares are up 12.6% over six months and down 14.3% over twelve. The branded socket and the unbranded one earn the same gross margin; only one of them carries an 11.4x multiple.

The bear case is not absent. At Barron's midyear roundtable, Scott Black judged Dell's hyperscaler servers to beat HPE's hands down on total cost of ownership and ease of maintenance — HPE's structural disadvantage in the very business Dell dominates. David Giroux has argued the opposite risk for both: hyperscaler in-sourcing plus rising memory costs squeezes every server OEM. What the April quarters showed is that those pressures land on the gross line of whoever ships the most accelerators, and that is Dell.

The setup

Where it stands — HPE's margin is expanding on networking mix while Dell's contracts on AI-server mix, and Dell's gross-profit multiple is twice HPE's.

Would confirm — HPE's next quarter holding networking operating margin in the low teens with full-year free cash flow tracking above $3.5bn.

Would invalidate — HPE gross margin falling back toward 30% as AI systems grow into a larger share of shipments.

Watch next — Dell reports second-quarter results on 1 September; HPE's fiscal third quarter follows in early September.

Valuation — HPE at 5.5x trailing gross profit and 4.8x forward, against Dell's 11.4x, Celestica's 18.9x and Super Micro's 6.1x.

Fair Isaac, whose FICO credit scores are pulled on essentially every American mortgage application, told investors in late July that its Scores division had grown 41% in the June quarter, to $458.9m. Almost none of that came from more lending.

The mortgage line inside Scores nearly doubled — up 97% — while mortgage origination volumes grew in the low single digits. The gap is a price list. Fair Isaac charges lenders a royalty each time a score is pulled, and it raised the wholesale mortgage price to $10 from $4.95. That is the fourth such change in three decades of selling into the mortgage market, following a tiered system in 2023 and a $3.50 step before it.

A meter with a cheaper meter next to it

The reason that arithmetic makes people nervous is that a substitute now exists at a hundredth of the price. The Federal Housing Finance Agency (FHFA), which oversees Fannie Mae and Freddie Mac, validated new credit models in April and put 21 large lenders into a first adoption wave, with Fannie permitting VantageScore 4.0 immediately. The bureaus have priced that product at roughly $1 a score.

Management's answer is that VantageScore's conforming share, concentrated at Rocket and UWM, sits near 20% and is a ceiling: lenders pull both scores and cherry-pick, and no FICO volume loss has yet appeared. That is a claim about lender behavior, and it is the single thing this equity turns on.

What the selloff obscured is that the non-royalty half of Fair Isaac is compounding faster than the royalty. Platform annual recurring revenue reached $413m, up 62%, overtaking the legacy licence book to become 51% of the $816m total; dollar-based net retention on the platform ran 148% against 82% off it. Group operating margin was 53.8%.

Fair Isaac still beat and raised — fiscal-year guidance went to $2.53bn of revenue and $42.43 of adjusted earnings per share. The market read it as a miss because revenue landed just under the $677m consensus and the guide under $2.56bn, and the stock fell 17.0% on 30 July. At $1,170.72 it is 50.9% below its November 2024 close of $2,382.40. The company bought back $1.96bn of stock in the quarter — three times its previous record — partly with a $1.5bn June term loan, leaving $5.58bn of debt at 5.64%.

The seat sellers rallied instead

The eight listed companies selling software into banking, insurance and government finance — nCino, Q2 Holdings, Alkami, BlackLine, Guidewire, SS&C, Tyler Technologies and Fair Isaac — returned about 19.8% over the past 30 days, equal-weighted. Fair Isaac was the only faller, at -4.2%. Most of the rest landed in two sessions on 27-28 July, when investors rotated out of chip stocks into beaten-down software and Guidewire, Tyler and SS&C each jumped double digits. This was participation in a broad unwind, not a banking-technology turn — and a modest one, against Atlassian's 99% and Wix's 68% over the same window.

Underneath it, two businesses diverge. Q2 Holdings, which runs digital banking, account opening and fraud tools for regional and community lenders, grew June-quarter revenue 12.6% to $219.8m while gross profit rose 24.5%; gross margin reached 59.2% from 53.6%, the sixth straight quarterly expansion as its cloud migration finished. Subscription ARR of $826m rose 15%, backlog of $2.8bn rose 17%, and full-year guidance went up. Its new artificial-intelligence products are sold as separate line items, not folded free into the seat price. At 8.2x trailing gross profit it is above the 6.9x of three months ago but well under 13.5x a year ago.

nCino, which sells banks a cloud operating system for onboarding, lending and compliance built atop Salesforce, is the weaker leg. Quarterly revenue growth ran 12.4%, then 9.6%, then 5.9%, then 10.6%; its own guidance for the quarter reporting in early September is $157.75m-$159.75m, or 6-7% growth. The real change is profitability: operating income of $22.1m, a 13.8% margin, against a small loss a year earlier. Its AI monetization is a pricing change rather than a usage meter — about 21% of contract value moved to one-time 10% renewal increases plus asset-growth escalators. At 6.1x trailing gross profit it sits below 9.3x a year ago.

The fashionable worry is that AI deletes the software seat. In this corner of the market it is the per-transaction toll, not the seat, carrying the visible risk — and the risk is a regulator, not a model.

The setup

Where it stands — Fair Isaac's Scores growth is price, not volume, and the shares sit at half their 2024 peak while earnings guidance rises. Would confirm — September-quarter mortgage Scores revenue growth holds above 40% with no disclosed decline in FICO pull volumes. Would invalidate — Fair Isaac discloses lost mortgage volume, or VantageScore conforming share moves durably above the 20% management calls a ceiling. Watch next — nCino's fiscal second quarter in early September; Fair Isaac's fiscal fourth quarter in early November. Valuation — Fair Isaac at 33.7x trailing and 27.3x forward earnings against guided fiscal-2026 earnings growth of roughly 45%.

Palo Alto Networks sells network firewalls and a widening stack of security subscriptions — threat prevention, web and domain filtering, cloud and data protection — to large enterprises, service providers and governments. Its pitch to customers is consolidation: replace a dozen point vendors with one contract, one console, one renewal. Over the past year it has been buying the pieces to make that pitch bigger.

What the headline number contains

The metric management steers by is next-generation security ARR, the recurring subscription base outside legacy hardware. In the April quarter it reached $8.1bn, up 60% year over year, and the company guided the full year to $8.90-8.95bn, growth of 59-60%.

About $1.6bn of that balance came in through acquisition. That leaves organic growth near 28% — still fast for a business of this size, but less than half the headline rate. The contracted backlog reads the same way: remaining performance obligations, the value of signed business not yet recognized as revenue, stood at $18.4bn, up 36%, and 22% excluding roughly $1.8bn from the two deals. Reported revenue of $3.0bn grew 31.1%, of which $388m was contributed by the acquired businesses.

What they cost

CyberArk, which secures privileged credentials and machine identities, closed on 11 February for $21.1bn of total consideration — about $2.3bn in cash and 112m Palo Alto shares. Chronosphere, an observability company, was bought for $3.35bn in cash and replacement equity on ARR above $160m as of September 2025, and is being folded into Palo Alto's agent-security platform.

The accounting shows up immediately. Gross margin fell to 67.6% from 72.9%, so gross profit grew 21.5% against revenue growth of 31.1% — purchase-accounting amortization of acquired intangibles sits in cost of revenue. The quarter carried a GAAP operating loss of $183m and a net loss of $177m. Cash is unaffected: adjusted free cash flow was $910m against $578m a year earlier.

The strategic logic is coherent. As enterprises deploy software agents that need credentials to act and telemetry to be audited, identity and observability stop being adjacent categories and become part of the security contract. That is the durable basis on which Palo Alto takes revenue from point vendors — and the reason Microsoft's bundled security suite is the competitor that matters. But a $15bn recurring-revenue target for 2030 requires sustained 20%-plus compounding from here, and the organic line is the one that has to deliver it.

The price got there first

The shares have risen about 38% in three months and 135% in six. In that entire three-month stretch the company published no financial statements; the last were filed on 2 June. Price per dollar of trailing gross profit went from roughly 16.7x in February to 28.3x in May to 38.1x now, a 128% expansion — flattered somewhat by the depressed acquired-margin denominator, though 29.2x forward gross profit and 86.9x forward earnings are unflattered. Consensus has revenue growing 24.3% this fiscal year and 21.2% next, with earnings per share of $3.77 and $4.10.

So the question the 1 September print answers is narrow: whether the recurring base grows fast enough without help to justify a multiple that more than doubled on no new information.

The setup

Where it stands — Headline recurring-revenue growth of 60% is roughly 28% organic, while the multiple has more than doubled since February on no new disclosure.

Would confirm — Fiscal fourth-quarter next-generation security ARR at or above $8.95bn with organic growth holding near 30%.

Would invalidate — Organic ARR growth slipping toward 20% or gross margin falling further below 67.6%.

Watch next — Fiscal fourth-quarter and full-year results on 1 September 2026.

Valuation — 38.1x trailing and 29.2x forward gross profit, against 16.7x trailing in February; 86.9x forward earnings.

A Chinese sensor maker sold more than 628,000 automotive laser scanners in the three months to June, close to twice the year-earlier count. The revenue it collected for them rose by roughly a fifth. That gap — units compounding, price collapsing — is the single most important fact in autonomous-driving hardware right now, and the market spent the past month paying up for it.

Hesai Group, a Shanghai-based maker of three-dimensional lidar sold to carmakers, robotaxi fleets and warehouse-robot builders, is the reason its corner of the auto-parts market rose over the past 30 days. Equal-weighting the four listed suppliers in advanced safety and autonomy gives a gain of 10.7% from 22 July; Hesai's 26.7% supplies about 6.7 points of it. Autoliv, the Stockholm airbag and seatbelt maker that sells no perception content at all, contributed roughly one. The move is not a passive-safety artifact. It is one company.

The price cut is the strategy

Hesai's June-quarter shipments split into more than 485,000 units for driver-assistance systems, up about 60%, and 142,000 for robotics, up 193%. Revenue rose 21.9% to CN¥860.8m with gross margin down to 40.1% from 42.5%. Gross profit grew 14.6%, well behind sales. Operating income has been negative for two consecutive quarters after running at a 9.7% margin last September; net income stayed positive on non-operating items, which is what the fifth-straight-profitable-quarter framing rests on.

None of this is accidental. Hesai plans to roughly halve the price of its main sensor, pricing the next-generation ATX below $200 against about $400 for today's AT128, betting in-house chips and factory automation make the cut pay. Management's counter-argument is content per car rather than price per unit: Li Auto fits four Hesai units to its L6, L8 and L9, taking sensor content toward $500-$1,000 on a vehicle listing near RMB 250,000. Hesai held 44% of China's long-range driver-assistance lidar market in June and has led for 17 straight months — but RoboSense has been added to Xiaomi's supply base, and its shipments show the same tilt away from cars toward robots. A May supply agreement with Mercedes-Benz, served from Thailand, is the one visible route out of the domestic price war.

On valuation the compression is real and it is not the market's doing this month: price per dollar of trailing gross profit has fallen from roughly 26.9x six months ago to 17.6x three months ago to about 13.9x now, while that gross profit still grew. Trailing earnings are 37.6x, and consensus for this year puts the forward figure near 38x once the renminbi estimates are converted. The shares fell 5.8% on the 18 August print — cost pressure, not upside.

Mobileye is the same trade in reverse

Mobileye, the Jerusalem-based, Intel-controlled supplier of EyeQ perception chips and the SuperVision and Chauffeur driving stacks, shipped 10m EyeQ units in the quarter, up 3% while its ten largest customers' own vehicle production fell 3%. It out-shipped its end market by eight points and still posted revenue of $508m, up 0.4%. Gross margin fell to 46.3% from 49.8%; gross profit declined 6.7%. Per the company's quarterly filing, EyeQ chips were about 90% of revenue and average selling prices fell on higher low-price China and Chinese-export volume. Third-quarter units are guided down 5-6%.

The stock fell about 15% on 23 July despite beating and raising, because founder-chief executive Amnon Shashua said he will step down once a successor is named; Mizuho cut its target to $8, below today's price. The raised profit outlook leans on a new research tax credit worth $180-200m this year, a benefit that roughly halves if Intel ceases to control the company. Against that, the shares sit at 7.75x trailing gross profit and 7.81x forward — no gross-profit growth priced — at 0.90x book after a $3.79bn goodwill write-down and a 5.3% free-cash-flow yield. A 2027 Stellantis program for cloud-enhanced highway assist carries more than twice base-driver-assistance gross profit per unit, but mix-driven price gains are a 2028 story.

The other two are not evidence

Aeva Technologies, a Mountain View developer of frequency-modulated lidar chips, jumped 44.6% over five sessions in early August because it launched an optical-connectivity business selling into AI data centers. The entire move reversed, 24.7% off between 17 and 21 August, in an AI-hardware selloff that also hit Ouster and Symbotic. On $6.1m of quarterly revenue it trades at 57.4x trailing sales with share count up 17.7% year on year. Autoliv, meanwhile, is the honest read on the end market: it cut its light-vehicle production assumption to -2.5% and China domestic to -5%, consistent with S&P Global Mobility's expectation that global output edges lower this year.

The month's one policy event was China's approval on 30 July of GB 44721-2026, its first mandatory safety standard for Level 3 and Level 4 automation, binding from 1 July 2027. The widely repeated claim that it mandates lidar is wrong — the text never names the technology, though its fallback-manoeuvre requirements push toward redundant sensing.

One technical note: Hesai's 50-day average only crossed above its 200-day on 28 July after months below it, and all three autonomy names remain lower over three months. This is a one-month turn inside a longer decline, and at Hesai the operating line is going the other way.

The setup

Where it stands — Hesai's shipments are compounding while price, margin and operating income fall; its shares have risen 26.7% in a month regardless. Would confirm — Q3 revenue of CN¥1.1-1.15bn on 800,000-850,000 units with gross margin holding near 40%. Would invalidate — A third straight quarter of negative operating income, or full-year shipments tracking below the 3m floor. Watch next — Hesai's third-quarter report, when management expects non-driver-assistance lidar to approach half of revenue. Valuation — Hesai at 13.9x trailing gross profit against 26.9x six months ago; Mobileye at 7.75x trailing versus 7.81x forward.

Clearway Energy did something in June that the market largely ignored. The owner of roughly 5,000 net megawatts (MW) of US wind and solar plus 2,500 MW of gas-fired plants tore up the commodity contracts at all three of its Texas wind farms — more than 600 MW — and replaced them with long-dated deals running past 2040. The Langford facility went to a 15-year PPA with an investment-grade counterparty at more favorable pricing on 10 June; Elbow Creek followed on 25 June. Both are accretive to earnings and cash available for distribution (CAFD) from the first month.

That is the whole thesis for owning contracted renewables into the data-center build-out, demonstrated rather than argued: a plant with no fuel cost, whose legacy contract rolls onto a market price set by gas.

The repricing is real, and it is not only demand

The average North American solar PPA reached $64.49/MWh in the first quarter and wind $79.40/MWh, the highest since LevelTen Energy began indexing in 2018, up 13% and 24% year on year. Labor shortages, tariffs and permitting delays are lifting the cost of new supply at the same time hyperscalers bid for it. Underneath, wholesale power rose 62% in New York and 45% in PJM during 2025, with prices near large data-center clusters far higher.

Clearway: the business improved, the stock did not

Second-quarter revenue was $481m, up 22.7%, with operating income up 36.5%. The blemish was weather: an El Niño-driven wind shortfall in the first half cut full-year CAFD guidance to $430–470m from $470–510m, about 8% at the midpoint. Management called it transitory and reaffirmed the 2027 target of $2.70 or better of CAFD per share, now guiding to the top of its 2030 range.

At $32.63 the shares sit at 12.1 times that reaffirmed 2027 figure, 0.71 times book, on a 10.1% trailing free cash flow yield. The moving averages turned against Clearway on 20 July, two weeks before the guidance cut — the market marked it down first. Cost of capital is now the binding constraint the company manages around: of the $500m–$1bn of external equity in its 2026–2029 plan, only $50m has been raised, and third-party acquisitions are deprioritized until the stock recovers. Its debt tells a different story — $600m of senior notes priced in January at 5.750%, barely above where the 30-year Treasury traded in August.

Brookfield: paid for the same thing

Brookfield Renewable, the Toronto-based owner of hydro, wind, solar and storage across the Americas and Europe, reported record funds from operations of $421m, $0.62 per unit, up 11% per unit. It signed 2.6 gigawatts of PPAs from its advanced pipeline and recycled $2.2bn of assets at or above target returns — the funding model still clears. It also anchors Microsoft's $10bn framework for more than 10.5 gigawatts through 2030, the largest corporate renewable agreement signed. Units are up 33% over twelve months at 9.9 times trailing EV/EBITDA — the cheapest of the three on that measure despite the richest book multiple, 2.25x.

XPLR: a different accident

XPLR Infrastructure, the former NextEra drop-down vehicle that suspended its distribution to self-fund buyouts of its convertible equity portfolios, is down 12.7% over 30 days. Almost all of it landed on 28–29 July, when revenue of $363m missed a $376.5m consensus despite an earnings beat and reaffirmed guidance. On 18 August, when the 30-year Treasury topped 5.33%, a 19-year high, XPLR fell 1.4% while Brookfield fell 4.8%. The most levered name was the least rate-sensitive that session. Its problem is operating leverage — operating income down 33% on maintenance cost creep — priced at 0.32 times book.

Demand is not the variable. Constellation Energy, the nuclear operator selling into the same load, is up 7.7% over 30 days.

The setup

Where it stands — Contract prices for renewable power are at record highs while Clearway trades below book and Brookfield does not. Would confirm — Clearway's third-quarter CAFD recovering toward the $430–470m full-year range as wind resource normalizes. Would invalidate — A cut to the 2027 CAFD-per-share target of $2.70, or equity issued at current prices. Watch next — Brookfield's unitholder vote on collapsing BEP and BEPC into one corporation, scheduled for October 2026. Valuation — Clearway: 0.71x book, 12.1x its 2027 CAFD target; Brookfield 9.9x trailing EV/EBITDA; XPLR 8.2x forward earnings against 16.6x trailing.