DK Street Journal

Targa Resources told investors on 17 August that subsidiaries of ExxonMobil had dedicated acreage in the Delaware and Midland basins to it under fee-based agreements running through 2046. The deal is integrated: Targa gathers the gas at the wellhead, processes it, and takes the resulting natural gas liquids (NGLs) downstream. It comes with three new Permian Delaware processing plants adding roughly 825 million cubic feet a day by 2028 and a 70-mile pipeline, Bull Run II, lifting takeaway to the Waha hub. Full-year net growth spending was raised to about $5.0bn in the same breath.

Targa — a Houston group with roughly 28,400 miles of gathering pipe, 42 processing plants and a Gulf Coast liquids export business — is the dominant processor in the Permian, with about a quarter of regional capacity. Its shares rose 8.4% the session after the announcement, on 2.49m shares against 0.43m the day before. That single day more than accounts for the entire month's gain across the seven largest gathering-and-processing companies.

The advance is earnings, not re-rating

Revenue is a poor read on this business, because most of it is commodity pass-through; in the first quarter, revenue fell 15.6% while gross profit rose 37.9%. Gross profit is the honest line, and in the second quarter it reached $2.14bn, up 139% from a year earlier, on a gross margin of 48.2% against 22.2%. Adjusted earnings before interest, taxes, depreciation and amortization (EBITDA) set a record at $1.603bn, up 38%. Permian plant inlet volumes hit a record 7.2 billion cubic feet a day.

So the price paid per dollar of trailing gross profit is about 10.6 times today — against roughly 11.9 times three months ago and 10.9 times in February. The stock is up 30.9% over six months and is no dearer per dollar of gross profit than when it started. On forward estimates the figure falls to about 9.2 times. Trailing price-to-earnings is 28.4, forward 26.6, priced off consensus for a 32% earnings step-up in 2026 that decelerates to 9.5% in 2027.

Two things temper it. Management said roughly $250m of first-half marketing and optimization margin was not in guidance and should not be assumed to repeat. And the growth is bought, not free: about $5.0bn of net growth capital against $5.7–5.9bn of guided EBITDA, leverage at 3.4 times, a trailing free-cash-flow yield of 1.15%, and a second-quarter buyback of $80m — 308,102 shares, about 0.14% of the count. Management points to a cash-flow inflection only after its export expansion completes in late 2027.

The constraint is takeaway, not load

The mechanism in the Permian this year has been oversupply, not demand pull. Waha hub prices were negative for 118 of the first 131 days of 2026, averaging -$2.19 per million British thermal units, with a record -$7.95 in late April. That forced Targa to accept 200–400 million cubic feet a day of producer shut-ins in the second quarter, resolved by July. The emerging squeeze is geographic: about 5 billion cubic feet a day of new processing capacity is being built in the northern Delaware while nearly all long-haul lines start at Waha. Bull Run II is aimed precisely at that gap.

None of that is artificial intelligence. Targa's gas marketing team is in discussions to supply fuel to behind-the-meter power projects, but no volume is contracted. The only one of these companies with a firm power deal is ONEOK, the Tulsa system that spans gathering, NGLs and refined products: a 1-gigawatt gas supply agreement requiring over $100m of capital, with more than 40 counterparties in discussion. ONEOK's shares are up 1.8% over three months. It trades at 16.1 times trailing earnings against 16.2 times forward — no growth priced — and 6.8 times gross profit, the cheapest of the three.

Antero Midstream: volumes up, earnings flat

Antero Midstream gathers and compresses gas in West Virginia and Ohio almost entirely from one affiliated producer, Antero Resources, and handles its water. Second-quarter gathering volumes rose 20% year over year to 4.1 billion cubic feet a day. Adjusted EBITDA rose 2%. Gross profit fell 1.7%, with margin down from 64.7% to 58.7%, and net income fell 8.8%. The volume came from the acquired HG Midstream assets, not from new demand.

It is nonetheless the more expensive stock: 12.7 times trailing gross profit and 12.5 times forward, against Targa's 10.6 and 9.2. The balance sheet is the strong part — leverage of 2.8 times after a $370m litigation settlement received in July, a twelfth straight quarter of free cash flow after dividends, a 9.3% trailing free-cash-flow yield. Its first regional pipeline, Eastside Express, is framed as a bet on West Virginia power and data-center demand at $200–300m of capital — underwritten by an affiliate acreage dedication with no minimum volume commitments.

What the trend does and does not say

Targa's 50-day average has sat above its 200-day continuously since 12 December 2025, the longest such run among these seven; Antero Midstream's ended on 18 August, on a day its shares actually rose 1.91%, after seven weeks stuck below its 26 June high of $23.19. Nor was the group's big session a bond-market bid: the 30-year Treasury yield topped 5.33% that same day, a 19-year high. Twelve-month returns inside the group run from Targa's +86.9% to Hess Midstream's -5.5%, with Kinetik, Western Midstream and MPLX in between. Targa is leading on results; Antero Midstream is diverging, growing throughput without growing earnings; ONEOK is following the group while holding the only contract that matches the story being told about all of them.

The setup

Where it stands — Targa's advance is funded by record Permian volumes and a 20-year Exxon dedication, not by any contracted power demand. Would confirm — Third-quarter adjusted EBITDA at or above $1.5bn with the $25–50m of returning shut-in margin recognized. Would invalidate — Fourth-quarter gross profit falling back toward $1.2bn as the $250m marketing windfall lapses without volume offset. Watch next — Third-quarter results in early November, and the East Driver and Train 11 utilization disclosure. Valuation — 10.6x trailing gross profit, 9.2x forward, against 11.9x three months ago and ONEOK's 6.8x.

GE Vernova, the power-equipment maker spun out of General Electric in 2024, sells heavy-duty gas turbines into a market where the queue has become longer than the product. In the June quarter it contracted 20 gigawatts of gas equipment. Only 2 GW of that was a firm order.

A queue that outgrew the order book

The other 18 GW were slot-reservation agreements — a customer pays to hold a manufacturing position while final terms are still being negotiated. Total gigawatts under contract went from 100 to 116, and the company is targeting at least 125 GW by year-end. But the composition matters more than the total: firm gas equipment backlog is 53 GW, while reservations stand at 63 GW. Management expects firm backlog to exceed reservations in the second half of 2026. That inversion is the single observable on which the equipment-revenue case rests.

The reason customers pay to queue is that there is nowhere else to go. Three firms build heavy-duty turbines — GE Vernova, Germany's Siemens Energy and Japan's Mitsubishi Heavy Industries — and waits now run five to seven years depending on frame class. Scarcity has moved price: GE Vernova guided 2026 gas orders to price 10 to 20 points above the fourth quarter of 2025, and industry turbine prices are projected to approach $600 per kilowatt by end-2027, close to triple 2019 levels.

Siemens Energy runs the same book differently, capping its reservation agreements at 10-20% of contract value and describing them as down payments that lapse if the turbine is never bought. It also deliberately held data centers to 20% of last quarter's gas turbine orders — the same share GE Vernova reported, but reached by rationing rather than by demand.

Scarcity is also being spent. GE Vernova is lifting its Greenville, South Carolina plant from 3 to 5 GW a quarter beginning this quarter, and raising output of its HA-class machines from 37 units in 2025 to 74 in 2027. Every added slot erodes the wait that justifies today's pricing.

The numbers underneath

Second-quarter revenue was $11.1bn, up 21.9% and accelerating from 16.1% the quarter before. Gross margin widened to 21.3% from 20.3%. Operating income rose 73%, more than three times the pace of sales. The Power segment earned an 18.8% EBITDA margin, up 320 basis points, against a group full-year guide of 12-14%. Orders of $24.2bn were up 88%, and equipment backlog — not services — grew 77% to $88bn. Free cash flow guidance was raised from $6.5-7.5bn to $11.5-12.5bn. The drag is Wind, where orders fell 40% and the segment lost $275m of EBITDA in three months.

The shares were up 11.5% over the trailing month as late as 17 August. Then they fell 12.7% in four sessions, 8.6% of it on 18 August alone — the day the 30-year Treasury yield touched 5.323%, a 19-year high. Nothing came out of the company. The whole behind-the-meter power complex went down together that session, alongside an analysis of hyperscaler filings putting roughly $3trn of off-balance-sheet obligations across nine AI-spending companies.

What the slot is worth

Earnings are not a usable anchor here: a one-off gain lifted first-quarter net income to $4.75bn. On price per dollar of trailing gross profit, GE Vernova trades at 30.5x, down from about 36.2x in May and back near its February level. Siemens Energy fetches 15.1x and Mitsubishi Heavy 12.1x on the same measure, with the same order book in front of them. The de-rating has been real and has not closed the gap.

The rest of the prime-mover layer is cheaper and, in places, growing as fast. Cummins, the engine maker whose Power Systems arm builds standby generator sets, posted record segment revenue of $2.3bn, up 19% at a 24.5% EBITDA margin, and is booking 95-litre orders into the second half of 2028; it trades at 9.23x trailing gross profit. Caterpillar's power generation revenue rose 72%, and it is restarting a 10-MW reciprocating engine it discontinued in 2022. Generac, long a residential standby brand, booked $100m of data-center revenue in the quarter against residential sales down 2%. Bloom Energy, which sells solid-oxide fuel cells that make electricity from gas without combustion, delivered its first $1bn quarter at a 37.2% product gross margin — but discloses no shipped data-center megawatts, and its diluted share count rose 37.6% in four quarters.

One constraint sits upstream of all of them. US gas production has a near-term ceiling around 128-132 billion cubic feet a day, and LNG exports are set to more than double by 2030. The levelized-cost models underwriting today's turbine orders assume a flat mid-$3s gas curve. If they are wrong, 63 GW of reservations is where it shows up first.

The setup

Where it stands — GE Vernova's order growth is in equipment, not services, but more than half the gas book is still reservations rather than firm orders. Would confirm — Firm gas equipment backlog exceeding 63 GW of slot reservations when third-quarter results are reported. Would invalidate — Gigawatts under contract stalling below the 125 GW year-end target, or reservations lapsing without conversion. Watch next — Third-quarter results in late October, with Power guided to 17-19% revenue growth at a 17-18% margin. Valuation — 30.5x trailing gross profit and 27.3x forward, against 36.2x in May, Siemens Energy at 15.1x and Mitsubishi Heavy at 12.1x.

ServiceNow sold considerably more software in the June quarter and kept less of each dollar than it used to. Revenue reached $3.99bn, up 24% from a year earlier and accelerating for a second straight quarter. Gross profit grew 13%. The gap is the story: GAAP gross margin fell to 70.68% from 77.48%, as hyperscaler compute for the company's agentic features and the consolidation of three acquisitions — Moveworks, Vesa and Armis — pushed cost of revenue up faster than sales. GAAP operating income fell 55% to $162m, leaving an operating margin of 4.06% against 11.14% a year earlier.

The demand side is not in doubt. ServiceNow's second-quarter release shows subscription revenue of $3.88bn up 24.5%, current remaining performance obligation — contracted revenue due within twelve months — of $13.2bn up 21%, 123 deals above $1m of net new annual contract value, and AI annual contract value crossing $1bn. Renewals ran at 98%. Management told investors roughly half the new business it writes is no longer seat-based, guided full-year subscription gross margin to 81%, and raised full-year subscription guidance by only $15m, calling part of the beat a pull-forward of federal on-premise deals from the third quarter.

Salesforce grew slower and kept more

Salesforce, the customer-relationship-management platform founded and run by Marc Benioff, is the mirror image. Growth is far lower but rising — 8.6%, then 12.1%, then 13.3% year over year in the April quarter, on revenue of $11.13bn. Gross margin was 76.92% against 76.96% a year earlier, essentially unchanged, which is unusual for a company selling inference-heavy features. Operating income grew 25% on that 13% revenue line. In its first-quarter release the company put Agentforce annual recurring revenue at $1.2bn, up 205%, with combined AI and data revenue above $3.4bn, and reported a record 34.8% non-GAAP operating margin.

The share count did the rest. Salesforce returned $27.5bn to shareholders in the quarter, including a $25bn accelerated repurchase that delivered 103m shares upfront. Diluted shares went from 962m in October to 871m in April.

That is why a 22% month left the valuation almost where it started. Measured against reported gross profit per diluted share, Salesforce cost about 5.68 times trailing gross profit in February, 5.14 in May and 5.45 now. ServiceNow went 10.60, then 9.70, then 12.14 — a 25% re-rating in three months on 13% gross-profit growth. The premium the workflow incumbent commands over the seat-priced one widened from 1.89 times in May to 2.23 times. On forward earnings the two sit at 31.6 times and 14.8 times; on trailing free cash flow, ServiceNow yields 3.45% and Salesforce 8.56%.

The consumption-pricing test failed

If this month were the market finally paying for usage-based billing over licensed seats, Twilio should have led it. Twilio sells voice, messaging and email programming interfaces that developers embed in their own products, billed per unit of usage, with 5,587 employees against Salesforce's 83,334. It was the weakest of the three, up 11.3%. Its business is fine — four quarters of accelerating growth to 22%, dollar-based net expansion of 116%, record free cash flow of $353m, full-year organic guidance raised to 13-13.5% — and its reported margin decline is a pass-through, about $71m of US carrier fees that never touch operating income. But it already costs 12.64 times trailing gross profit against 6.95 in February, and third-quarter organic growth is guided to 11-12% from 17%.

What actually moved was the whole category. Atlassian rose 87.5% over the thirty days, Workday 42.6%, Veeva 31.0%, Microsoft 21.5%, while HubSpot managed 7.9% and JFrog 1.2%. Morningstar's application-software index fell roughly 27% from October to mid-July and has bounced about 15% since Microsoft reported on 22 July, with software rising as semiconductors fall. The second leg came on 13 August, when Silver Lake was reported in talks to take Workday private at a value above $51bn — a private-equity bid read as a vote of confidence in seat-priced software. Salesforce has disclosed no new financial number since 27 May; two windows, 22-28 July and 19 August, account for 17 points of its 22-point month.

What each side is defending

ServiceNow's durability rests on its configuration management database, which encodes the dependency state of a customer's IT estate; replacing it means re-architecting operations, which is what a 98% renewal rate looks like. Salesforce's risk is subtler and runs the other way: rival platforms are indexing customer-relationship and Slack data into their own context graphs, and Atlassian's latest shareholder letter treats Salesforce as an ingested data source rather than a partner, with an agent that turns Slack threads into work items. Coordination value migrates to whoever holds the cross-application context, not to the application that generated it.

The price history frames both. Salesforce fell into a downtrend on 18 June and only regained a rising 50-day average above its 200-day on 19 August; ServiceNow got there a week earlier. Both remain well below a year ago — Salesforce by 15%, ServiceNow by 26%.

The setup

Where it stands — The market is paying a widening premium for ServiceNow's growth while its gross profit decelerates, and almost nothing extra for Salesforce's. Would confirm — ServiceNow subscription gross margin recovering toward the guided 81% while cRPO growth holds above 20%. Would invalidate — Salesforce cRPO growth falling below its total revenue growth, or Agentforce ARR growth slowing sharply from 205%. Watch next — Salesforce reports second-quarter fiscal 2027 results on 26 August 2026, after the close. Valuation — Salesforce 5.45x trailing and 4.78x forward gross profit; ServiceNow 12.14x and 10.96x, against 9.70x in May.

Two Michigan utilities are financing the largest construction programs in their corporate histories for customers whose buildings are not finished. DTE Energy, which delivers electricity to about 2.3 million customers in southeastern Michigan and gas to 1.3 million statewide, has signed 2.4 gigawatts of data-center agreements — 1.4 GW with Oracle already under construction, and 1 GW with Google awaiting regulatory approval expected in September. CMS Energy, whose Consumers Energy unit serves 1.9 million electric customers from Jackson, has agreed to supply roughly 1 GW to a single hyperscale site, with local zoning the remaining gate.

Both reported falling operating income in the June quarter. CMS revenue slipped 0.5% to $1.83bn while operating income dropped 16.7% and net income fell 40.3% to $120m. DTE's revenue fell 2.4% and operating income 7.3%; in the March quarter its operating income fell 34% even as revenue grew 15.8%. This is what a rate-base build looks like from the inside: depreciation, operating costs and interest arrive on the income statement before the regulator lets the meter reflect them.

The gap between what capital costs and what it earns

The Michigan Public Service Commission held Consumers Energy's authorized return on common equity at 9.9% with a 50/50 capital structure, rejecting the 10.25% the utility asked for. Weeks earlier it made the same call on DTE Electric, refusing a request for 10.75%.

Meanwhile the price of new capital has climbed toward that ceiling. NiSource sold $750m of junior subordinated notes at 6.25% on 13 August; NextEra placed $3.75bn of hybrid debentures in June at 6.00%-6.625%, and Emera issued at 6.65% and 6.85% in March. Duke's mid-August equity units cost 7.75% all-in. Against a 9.9% allowed return, incremental funding now carries two to three points of headroom — on five-year plans of $24bn at CMS, driving 10.5% compound rate-base growth, and $36.5bn at DTE.

Neither company can fund that internally. Both run negative trailing free-cash-flow yields, minus 8.9% at CMS and minus 6.9% at DTE. CMS filed a $3.0bn at-the-market and forward equity program in May; DTE targets $500m-600m of equity a year through 2030. Diluted share counts are up only about 0.5% so far, so most of the dilution is still ahead.

Who is actually ahead

The two compete for the same siting decisions under the same regulator, and the durable advantage is contract execution, not technology. DTE is ahead on executed load, with 2 GW more in advanced talks and what it describes as an 8.4 GW opportunity in Michigan. It told investors Oracle should deliver about $300m a year of bill benefit to existing customers and Google roughly $1.7bn over the contract's life, possibly deferring its next rate case to 2028. It also flagged Oracle's credit downgrade as a live risk on its largest signed contract, mitigated by collateral triggers.

CMS's protection is written into the tariff. Michigan's approved large-load terms require a 15-year minimum contract and a minimum billing demand of 80% payable whether the power is used or not. CMS is also selling NorthStar's non-utility renewable development arm, redirecting $1.7bn of capital to the utility and cutting planned equity issuance by at least $350m.

The politics are unresolved. Michigan's attorney general called DTE's offer of a two-year rate freeze conditioned on a data center opening on time a "ransom note."

The fall bought no cheapening

Since early May, CMS's market value is down 7.3% and DTE's 8.0% — and the multiples did not follow. CMS went from 20.6x trailing earnings to 20.2x; DTE's expanded, from 20.9x to 21.2x, because profit fell as fast as the price. Both sit near 17.5x forward against a long-run sector median closer to 16.8x. Consensus still carries CMS earnings up 7.8% this year and DTE up 6.6%, and both managements reaffirmed 6%-8% growth in July.

Over the past month the common shares have fallen faster than the companies' own listed hybrid notes — CMS down 5.7% against roughly 3.5% for CMSA, CMSC and CMSD; DTE down 7.0% against about 4.7% for DTB, DTG and DTW — inverting the twelve-month pattern, when the bonds fell 9%-16% and the equities held. That bond decline is duration, not doubt: the 30-year Treasury reached 5.34% on 18 August, a 19-year high, while utility investment-grade spreads sit at 95-105bp, inside their five-year average. The equity leg is the one repricing now.

The setup

Where it stands — Both utilities are growing rate base into contracted AI load while reported operating income falls and the shares de-rate without getting cheaper. Would confirm — Third-quarter operating income declining year on year again at either company, a third straight quarter. Would invalidate — Operating margin recovering toward 17% at CMS as new Michigan rates land, with 2026 guidance held. Watch next — The Michigan Public Service Commission's decision on DTE's 1 GW Google contract, expected September 2026. Valuation — CMS 20.2x trailing and 17.6x forward; DTE 21.2x and 17.5x, against a 16.8x long-run sector median.

Two of the companies whose machines must sign off on a stack of high-bandwidth memory before it can be sold into an artificial-intelligence server told investors this month that their order books have never been fuller. Then, in the four sessions after 17 August, their shareholders lost more than the entire month's gains.

This is the rung of the memory supply chain nobody names: process control. Every 2.5D package and every stack of high-bandwidth memory (HBM) has to be inspected, measured and electrically probed before it ships, and the intensity of that work rises faster than wafer volume as stacks get taller. Micron has said its HBM capacity is fully booked through 2026, with meaningful new capacity arriving only in 2027, and HBM4 sells for roughly ten times conventional DDR5 memory per gigabyte — so the test burden is being pulled forward onto lines that already exist rather than deferred to new fabs.

Onto raised the number, then handed back the rally

Onto Innovation, a Wilmington, Massachusetts maker of defect-inspection, optical-metrology and packaging-lithography tools, reported June-quarter revenue of $343m, up 35.3% from a year earlier. That is a sharp break from the 9.5% growth of the prior quarter and the 13.5% decline of a year ago. Advanced packaging and specialty devices — 2.5D logic, HBM and silicon photonics — hit an all-time high at close to half of the quarter, about $170m. Gross margin recovered to 53.4%, from 46.4% two quarters earlier, and operating profit nearly doubled on a third more revenue.

Management then raised full-year advanced-packaging growth to about 80%, from about 50%, and disclosed backlog above $1.1bn. Between 30% and 40% of that backlog delivers in 2027. The shares gapped 14.7% higher the next session. All of it is gone. The competitive claim underneath is real: Onto has been taking inspection and metrology share from KLA and is qualified at TSMC, Samsung and Intel for gate-all-around transistor metrology.

Camtek's paradox

Camtek, an Israeli inspection specialist in Migdal HaEmek with 709 employees, is the awkward case. June-quarter revenue of $133.2m grew only 8.0%, and GAAP operating income fell 14.8% year on year — a second consecutive decline — with operating margin compressing to 20.4% from 25.9%.

Against that, the disclosure: more than $600m of orders year to date, 80% of it advanced packaging, over half from outsourced assembly houses and more than 20% from HBM manufacturers, delivering through 2026 and into 2027. Advanced packaging was 75% of the quarter and is guided to 80% by the fourth. Third-quarter revenue is guided to $158-160m, about 20% above the second. Chief executive Rafi Amit attributed the demand to the industry's transition to HBM4 and continuing 2.5D and 3D packaging capacity. The gap between an order book and a profit-and-loss statement is a delivery schedule — but for two quarters now, the schedule has been costing Camtek margin.

FormFactor is the third leg and the cleanest. Its probe cards are consumed per die tested, and DRAM probe-card revenue rose 48.9% to a record $85m, with HBM about two-thirds of it and two customers already adopting its full-wafer contactor for testing above 10 gigabits per second. Gross margin widened 13.5 points in a year to 50.7%. The one genuine slip is its hyperscaler custom-chip revenue, now expected from 2027.

What actually broke

From 17 to 21 August, Onto fell 18.2%, Camtek 18.4% and FormFactor 19.9%. Over the identical four sessions ASML fell 7.2%, Applied Materials 9.0% and Lam Research 10.5%. Before 17 August all three smaller names were up over the trailing month. The 30-year Treasury yield topped 5.33% on 18 August, a 19-year high, and earnings sitting in 2027 order books discount harder than earnings arriving next quarter.

The rest is an industry argument. Applied Materials reported a record quarter of $9.115bn and guided above consensus, and fell almost 6%, because it declined to lift full-year equipment-market growth beyond 30% while its customers' own capital plans imply nearer 40%. Equipment shares kept falling on 19 August after yields retreated. The market is pricing the gap between what chipmakers say they will spend and what toolmakers will underwrite.

Where the price sits

On price per dollar of trailing gross profit — the fairer lens here, since one-off losses distort Camtek's trailing earnings to 183x — FormFactor is at 21.0x, against 28.2x three months ago and 23.7x six months ago, while its trailing gross profit grew 80% over the same six months. Camtek is at 26.4x against 29.9x in May, though two quarters of falling operating income mean part of that de-rating is earned. Onto is the weakest case for dislocation: 25.3x now, 26.0x in May, but 21.4x in February, having peaked near 30.9x on 17 August. Its forward price/earnings ratio of 36.2x against a trailing 109.5x is the measure of the growth being underwritten — consensus revenue of $1.43bn for the year to January 2027, up 42%.

All three remain among the market's largest twelve-month winners: FormFactor up 299%, Onto 172%, Camtek 77%. What happened last week did not reverse that; it removed August.

The setup

Where it stands — Order books at Onto and Camtek are at records, but four sessions of duration repricing took roughly a fifth off all three names. Would confirm — Camtek delivering third-quarter revenue of $158-160m with operating margin recovering toward its 30-32% year-end target. Would invalidate — Onto's second-half revenue failing to exceed the first half by the guided 25%, or backlog falling below $1.1bn. Watch next — Third-quarter results in late October; FormFactor guided to $270m, plus or minus $10m. Valuation — Price per dollar of trailing gross profit: FormFactor 21.0x, Onto 25.3x, Camtek 26.4x, against roughly 34x for all three in early May.

The world's three largest memory makers spent this year diverting capacity toward high-bandwidth memory for AI accelerators, where a gigabyte of output earns three to four times what ordinary server memory does. The result reached everyone who builds hardware around it: DRAM contract prices rose 90-95% quarter on quarter in January, and server DRAM rose a further 58-63% in the second quarter, according to TrendForce. Dell raised list prices across its product line roughly 17% at the end of March to cover it.

All three of the big enterprise hardware names have now told investors the same thing: gross margin is going down. Only one of them was marked down for saying it.

Dell: the margin that was never the argument

Dell Technologies sells servers, storage and networking to corporations and governments through its Infrastructure Solutions Group, and PCs through its Client Solutions Group. In the quarter to 1 May, revenue grew 87.5% to $43.84bn — the third straight acceleration, from 10.8% two quarters earlier. Gross margin fell to 17.75%, so gross profit rose only 57.6%. Operating income rose 197%, to $3.66bn.

That gap is the whole Dell case. The infrastructure segment produced $29bn of revenue, up 181%, with operating margin up 80 basis points to 10.5%, while the PC business improved its margin faster still, by 280 basis points to 8.0%. The company booked $24.4bn of AI orders and recognised $16.1bn of AI server revenue, raising its expectation for full-year AI server revenue to $60bn, and exited the quarter with a record $51.3bn AI backlog. Dell is the largest AI server builder by volume, and scale plus supplier relationships give it allocation priority on Nvidia GPUs — a position rivals have to buy their way into.

What has run ahead is the price. Six months ago the shares changed hands at roughly 3.8 times trailing gross profit; in May, about 7.4 times; today, 11.49 times. Trailing gross profit grew 19% over that stretch. Dell now trades at 34.5x trailing earnings and 23.5x forward, against the roughly 19x Barron's cited in July and the 14-17x range that made it, a few months ago, the cheapest large-cap way to own AI hardware. The company repurchased 11m shares at an average $147 in the May quarter; the stock is $435 now.

NetApp: treated like Dell, priced like a re-rating

NetApp sells flash storage arrays and the ONTAP software that runs them, plus cloud storage services resold through Microsoft Azure and Amazon Web Services. Fiscal 2026 revenue was $6.93bn, up 5.4%, but the fourth quarter accelerated to 12.5% and operating margin expanded 715 basis points to 27.3%. All-flash array revenue hit a record $1.2bn, up 18%, and public cloud revenue reached $182m. The company logged more than 1,100 AI and data-preparation wins in the year but discloses no AI revenue line in dollars.

It also guided fiscal 2027 gross margin down to 68.5-69.5% from 71.3%, citing memory and component costs. That is the same disclosure Cisco made. NetApp's price per dollar of trailing gross profit went from 4.55x on 3 May to 7.70x today — a 69% expansion against annual gross profit growth of 6.2%. The forward multiple of 7.08x sits barely below the trailing one, which is the market saying it expects almost no gross-profit growth at all.

Cisco: fastest AI growth, only de-rating

Cisco Systems makes the switches, routers and optics that wire data centers together. Revenue accelerated four quarters running, to $17.25bn in the July quarter, up 17.6%, with operating income up 38.1%. Its AI infrastructure orders reached $9.3bn for the year, roughly 4.5 times the prior year, with fiscal 2027 AI revenue guided to about $7.5bn against the "at least $6bn" previously signalled. On the fastest AI growth of the three, the shares fell 7.6% on 13 August, because those hardware-heavy orders dragged gross-margin guidance to 65-66%.

That is the asymmetry. Cisco's valuation is built on in-house Silicon One switching chips and Acacia coherent optics — more than 40,000 800-gigabit pluggables shipped, an integration Arista and the Juniper business now inside Hewlett Packard Enterprise cannot match — and on the software and services attached to them. Diluting that mix costs it multiple. Dell's gross margin was 17.75% to begin with; there is no premium there to lose. Cisco's price per dollar of trailing gross profit compressed from about 11.9x in May to 10.71x, while its trailing gross profit rose from $39.1bn to $40.9bn.

The trend readings follow the same split. Dell's shares have been in a confirmed uptrend since 31 March, with the 50-day average above the 200-day, having been in the opposite condition as late as 11 March; NetApp crossed over on 21 May; Cisco's uptrend was downgraded on 20 July and has not recovered. Neither of the winners ground higher: Dell gapped 21.9% on 27 February and 32.8% on 29 May, NetApp 12.4% and 22.4% in the same May week. Almost all of it arrived on earnings.

One caveat cuts against the cheap-hardware-rotation reading. Long rates are not helping: a 10-year Treasury auction on 12 August cleared at 4.683%, the highest since 2007. Multiples expanded anyway at two of these three, which points at AI content rather than a defensive rotation — except that the company with the most AI content is the one that got cheaper.

The setup

Where it stands — The same memory-driven gross-margin guide-down re-rated Dell and NetApp and de-rated Cisco, on Cisco's faster AI order growth. Would confirm — Dell holds infrastructure operating margin at or above 10.5% on 1 September despite a further gross-margin fall. Would invalidate — Dell's operating margin contracts alongside gross margin, making the dilution a profit problem rather than a mix effect. Watch next — Dell reports fiscal Q2 on 1 September; NetApp reports fiscal Q1 after the close on 2 September. Valuation — Dell 34.5x trailing and 23.5x forward earnings, against roughly 19x cited in July and 14-17x in May.