DK Street Journal

Monolithic Power Systems, which designs the direct-current voltage regulators that turn rack current into the specific volts an artificial-intelligence accelerator runs on, has agreed to put its proprietary process technology inside somebody else's fab. The long-term manufacturing agreement with GlobalFoundries, announced September 9, opens capacity at a 300mm Singapore plant for automotive, industrial-robotics and AI power stages from early 2027. Eight days earlier the same foundry delivered Navitas Semiconductor's first American-made Gen 5 gallium-nitride parts from its Vermont plant.

Neither company's story advertises the overlap, and it matters because the two sit at opposite ends of the same few inches of copper under a server board: Monolithic Power converts the last step down to the processor and already gets paid for it; Navitas, a 190-employee Torrance designer of gallium-nitride and silicon-carbide power devices, is priced on an architecture that has not shipped. Both now need one supplier's 2027 capacity. Three dated things have happened since: a research note, a rate decision, and a reversal in both share prices.

The half that is accelerating

Monolithic Power's June-quarter revenue was $980.6m, up 47.6% year on year and the fourth consecutive quarter of faster growth. Enterprise data — the accelerator and server business — reached $380.6m, up 164.3%, taking that segment from roughly a fifth of company revenue to 38.8% in a year, and the company lifted its full-year growth floor for the segment from 85% to 130%. Gross margin was 55.2%, inside the same narrow band it has held for eight quarters, so the mix shift has cost nothing at the gross line. Operating margin widened to 31.0% from 24.8%. Third-quarter revenue is guided to $1,140m–$1,160m, a midpoint 56% above the year-ago quarter.

Management said on the July 30 call that channel inventory was "very low," that book-to-bill sat well above 1.0, and that enterprise data has "no concentrated customers." Founder-chief executive Michael Hsing told investors on that call that the company "relies on its own silicon carbide devices for the 800-volt solution" — an internal answer to the architecture change, sampling rather than shipping.

Against that, the valuation has come down hard. Priced against trailing gross profit, Monolithic Power has fallen from 49.8x in early May to 33.1x, 26.1x on forward estimates, while gross profit itself grew 47.8%. Market capitalization is $59.8bn, against $81.2bn in May. The de-rating is not estimates falling: consensus has revenue up 49.2% this year and earnings per share up 54.8%.

The complex fell in order of price, not of sockets

On September 14 the stock lost 7.4% in one session on a research note questioning its allocation on Nvidia's next accelerator platform — the same category of event that took it down 24.7% in five sessions in November 2024, when Renesas and Infineon were reported to have taken Blackwell regulator work. The threat is real and named: Analog Devices paid $1.5bn for Empower Semiconductor, whose vertical regulators sit in exactly that position, and Nvidia's own 800-volt supplier list runs to ten silicon vendors, Monolithic Power and Navitas among them. A place on a list is not a socket.

But the three-month cross-section does not read like a socket being reassigned. Every power and analog name fell: Navitas 52.4%, Vishay 49.9%, onsemi 42.5%, Power Integrations 42.2%, STMicroelectronics 36.0%, Monolithic Power 22.1%, Texas Instruments 17.4%, Analog Devices 13.5% — while an S&P 500 fund rose 2.0% and Nvidia itself rose 5.5%. Monolithic Power was the fourth-mildest faller of twelve. The ordering tracks starting valuation and how far out the cash flows sit. The Federal Reserve raised its benchmark rate a quarter point to 3.75%–4% on September 16, its first increase since 2023, with the ten-year Treasury yield above 5% for the first time since 2007 — which is a direct tax on 2027 revenue. Both stocks bottomed within a session of the decision and turned up; Monolithic Power rose 6.6% off its September 15 low on the heaviest volume in a fortnight.

Navitas: the date moved, the revenue did not

Navitas booked $10.5m of revenue in the June quarter, down 27.3%, at a gross margin of minus 9.5%. Across four quarters it produced cumulative gross profit of minus $0.6m on $36.5m of sales — a year of shipping product at no gross profit. September revenue is guided to $13.5m, which would be the first year-on-year growth in five quarters. Solvency is not the issue: $557.4m of cash and no debt against roughly $68m of annual cash burn. Dilution is — shares outstanding rose 18.5%, including $373m raised at $21.89, nearly double the current price.

The date is the harder problem. On the July 27 call management put gallium-nitride content inside accelerator voltage conversion at mid-to-late 2027; six weeks later, at Citi's Global TMT conference on September 8, chief executive Chris Allexandre dated native 800-volt systems to early 2028. "It would have happened without NVIDIA, it's happening much sooner due to their influence," he said of the shift. Consensus has revenue of $47.7m this year, up 4.7%, and no positive earnings before 2030, against 62.6x forward sales.

What each side earns

Monolithic Power's fall is mostly duration and an extreme starting point, not a broken business: its numerator shrank while its denominator grew 47.8%, and the three disclosures that would have to crack — inventory, book-to-bill, concentration — have not. What the business does not yet explain is why it should still be the most expensive name in the group, at 26.1x forward gross profit against roughly 17.7x for Analog Devices, when the buyer of its main socket has just bought a competitor. Navitas's de-rating is earned outright: a company selling below cost, with its defining revenue pushed two quarters further out in six weeks, was never going to hold 80x sales through a rate increase.

Neither company has published a third-quarter reporting date; Monolithic Power's own precedent is the last week of October. Until one of them prints, the only new hard fact either has given investors this month is where the wafers will come from.

The company that buys the beef is rarely the company whose name is over the door. Meritage Hospitality Group, which runs 314 Wendy's restaurants and is the chain's largest franchisee in the United States, filed for Chapter 11 on September 17 after failing to meet payment obligations under its franchise agreements. Wendy's own franchising subsidiary is listed as the largest unsecured creditor, on a $24.9m claim for deferred fees. Meritage had already closed roughly 60 restaurants earlier in the year.

A franchisor's income statement carries no line for ground beef. It carries a royalty percentage of somebody else's sales, plus rent on property the franchisor owns or master-leases at a markup. Input inflation therefore lands on the operator first and on the brand owner later — through franchisee cash flow, which decides whether units get built, whether they stay open, and how much discounting an operator can fund. Four burger businesses have been marked down for a year on that question. McDonald's, Shake Shack, Wendy's and Arcos Dorados fell an equal-weighted 19.9% in the twelve months to September 18, a grind rather than a gap: McDonald's closed lower in 13 of the 18 sessions from late August, none worse than about 2%. Restaurant Brands International, which owns Burger King, rose 15.8% over the same year, so this is not the market marking down everything that sells a fast burger.

The input is a supply story

The American beef cow herd has shrunk to 27.6 million head, the smallest since 1961 according to the Department of Agriculture's January 2026 inventory, and cattle prices are expected to average roughly $241 per hundredweight this year, about 8% above 2025. Retail ground beef set an all-time high of $6.90 a pound in May. Fewer cattle, not hungrier customers.

What McDonald's is actually paid

About 95% of McDonald's restaurants were franchised at June 30, and those revenues are rent and royalties on a percent of sales. In the June quarter franchised revenue was $4,393m, up 4%, against $2,525m of sales from restaurants the company runs itself. The franchised line produced $3,713m of restaurant margin — 84.5% — while the company-operated line produced 15.3%. And in the US, company-operated margin dollars fell 6% to $91m. The cost shock is visible exactly where the structure says it should be.

The volume meter is worse than the margin. Global comparable sales rose 1.3%, against 3.8% a year earlier, and the US gain of 0.8% came from higher average check with fewer guests. Only 60% to 65% of the US system implemented the ten-item under-$3 value menu, chief executive Chris Kempczinski said, and a loose $3 parameter let some franchisees raise prices on items like small fries — a franchisor can mandate value but cannot pay for it out of the operator's profit and loss. "We made a bad trade in Q2... And we've got to get that fixed," Kempczinski told investors on the August 4 call, attributing two-thirds of the US sales miss to the offer change. Numerator estimates the company lost $310m of sales from lower-income customers in the quarter — in the demographic that is supposed to trade down toward it.

The operator's half of the same shock

Shake Shack runs nearly all of its Shacks itself, so it buys the beef. Food and paper rose 60 basis points to 28.8% of Shack sales, and restaurant-level profit of $92.7m was 23.0% of sales, down 90 basis points on record beef costs. Average weekly sales per Shack were flat at $78,000: the 17.2% revenue growth to $417.6m came from opening more Shacks, while operating income fell 7.3%. Operating leverage running backwards. The company had already cut second-quarter and full-year guidance in June, taking restaurant-level margin guidance to 22.0-23.0% from 24.0-24.5%. "Our second quarter results reflect a business that continues to execute across sales, development, and profitability despite operating in one of the most challenging cost environments we have faced in many years," chief executive Rob Lynch said on August 5.

Arcos Dorados, McDonald's exclusive master franchisee across 20 Latin American and Caribbean markets, shows the beef shock is North American. It posted record quarterly revenue of $1.31bn with adjusted profit before interest, tax, depreciation and amortization up more than 20%, on a third consecutive quarter of falling food-and-paper cost. "Brazil was the standout performer for Arcos during this quarter. EBITDA margin expanded 180 basis points with an EBITDA margin of 14.6%," chief financial officer Mariano Tannenbaum said on the August 13 call. It is the only one of the four whose shares are higher than a year ago, up 11.6%, at 6.2x trailing earnings.

What the de-rating is paying for

Wendy's decline is earned. US same-restaurant sales fell 7.0% on a 12.5% traffic decline, the dividend was cut to $0.07, the outlook withdrawn, net leverage sits at 5.0x, and 21 new US openings in the first half were swamped by 289 closures. Its shares trade at 10.2x trailing earnings but 13.3x forward, because consensus 2026 earnings per share are 41.7% below 2025 — profit falling faster than price. Shake Shack's 48.4x forward multiple sits against consensus 2026 earnings per share now expected to fall 12.2%; on enterprise value to EBITDA it has crossed below McDonald's, 12.2x against 15.3x.

McDonald's is the one the business does not fully explain. It trades at 20.1x trailing and 19.2x forward earnings, against roughly 25.8x a year ago — a 22% contraction while trailing earnings per share grew 5.5%, and while analysts trimmed price targets without changing ratings. Part of that is rate arithmetic: the 4% dividend increase to $1.93 a quarter announced on September 17, a 50th consecutive annual raise, yields 3.11% at the current price — below the 3.75%-4.00% federal funds target the Fed set on September 16 in its first hike since 2023. An annuity re-prices when the risk-free rate rises. The rest is the question Meritage just made concrete: the royalty-and-rent stream is only as durable as the operators underneath it, and for the first time in this cycle one of them stopped paying.

McDonald's brings a plan it calls "McDonald's > NEXT" to an investor day in Chicago on September 23. The targets read hardest there will not be the ones describing its own margin, but whatever it can say about the profit left in the restaurants it does not own.

Federal regulators spent September 18 refusing to end a railroad merger fight early. The Surface Transportation Board denied three motions asking it to reject Union Pacific's revised application to control Norfolk Southern for failure to make a prima facie case — motions filed on August 6 by BNSF Railway, by a shipper coalition including the American Chemistry Council and The Fertilizer Institute, and by CSX Transportation. The Board was careful about what it had done: the decision "does not reflect any determination on the merits of the proposed transaction nor is it an endorsement of Applicants' arguments, analyses, or positions." Parties may renew everything in opening comments due November 18.

That date is now the meter for Norfolk Southern's equity, because its share price has stopped being a function of what it hauls. Holders of the eastern Class I, which runs roughly 19,300 route miles across 22 states hauling chemicals, autos, coal and intermodal boxes, are owed one Union Pacific share plus $88.82 in cash. On September 18 that package was worth $368.19 against a $314.16 close — a gross discount of 17.2%, wider than any reading since April, and against 13.0% on August 19, the day the Board lifted its abeyance and started the clock.

The target trades on its acquirer

The inversion is measurable. Over the three months to September 18, Norfolk Southern's daily moves correlated 0.92 with Union Pacific's and 0.81 with CSX's. Before the deal, from January to July 2025, the readings ran the other way: 0.90 with its eastern neighbor, 0.81 with its future buyer. Decompose the last month and the point sharpens — the package Norfolk Southern is owed fell 5.0% as Union Pacific's own shares slid, while Norfolk Southern fell 7.3%. Roughly 2.3 points of its decline is widening deal risk alone.

The freight underneath is not what is moving it, though it is not flattering either. June-quarter railway operating revenue rose 11% to a record $3.5bn on 4% more volume, but six of those eleven points came from higher fuel surcharges, and higher fuel expense carried roughly a 110-basis-point operating-ratio headwind. Operating income grew 1.8%; margin fell about three points to 32.4%. "We've had basically 4 years that we've navigated through a freight recession," chief executive Mark George told the July 23 call. "It is the longest freight recession in history." Management also said the company lost some volume after the merger was announced.

CSX is an adversary with a shopping list

CSX — 19,500 route miles east of the Mississippi, unique access to Mid-Atlantic ports and PJM generation, 22,200 employees — is not objecting on principle. It is asking for trackage rights from Chambersburg to Manville, New Jersey via Harrisburg and the Lehigh Valley, plus access over Norfolk Southern lines to the port terminal at Norfolk. BNSF wants 824 miles between Chicago and Pennsylvania intermodal terminals. Union Pacific has rejected the requests. Those responsive applications land November 18 too.

On its own account CSX is the best operator in the group and the least explained share price. Revenue grew 10.1% to $3.94bn in the June quarter, the operating ratio improved to 61.7% from 64.1%, volumes rose 6% and intermodal 9%. "Our railroaders successfully managed substantial volume growth," chief executive Steve Angel said on July 22. The shares are up 40.9% over twelve months, the best of the five North American Class I roads and far ahead of the merger target's 11.1%, and down 6.3% over the last month.

The selloff has de-rated nothing: CSX's trailing earnings multiple is 27.2x, against 27.4x in early May at a lower share price. Looking forward, CSX at 23.5x and Norfolk Southern at 24.2x both sit above Union Pacific's 21.4x — despite Union Pacific carrying the best operating margin of the three at 40.3%.

What the decline is, and is not

All five Class I railroads fell between 6.0% and 7.3% from August 18 to September 18, with Canadian National and Canadian Pacific Kansas City squarely inside that range, which is the evidence that the month is macro. Diesel set a record $6.285 a gallon on September 14 while rail fuel surcharges reset on a lagged formula — grain surcharges held at 48 cents a mile through September — so the cost lands before the recovery does.

The verdict splits cleanly. CSX's year is earned by the accounts: accelerating revenue, expanding margin, an operating ratio in the low sixties. Its month is a sector de-rating on fuel that its own multiple has not yet absorbed. Norfolk Southern's month is mostly a regulatory reprice, and its multiple is not a judgment on its earnings at all — it is an exchange ratio wearing a price-to-earnings ratio's clothes. Anyone valuing it on carloads is reading the wrong instrument.

A railroad that booked record revenue in June now trades on a filing deadline two months out — and on whether the rival asking for its track gets it.

Intercontinental Exchange, which owns the New York Stock Exchange, a fixed-income data franchise and the Encompass mortgage-origination platform, earned more in the June quarter than in any quarter of its history and still trades below where it did a year ago. Revenue grew 10.7% year on year and operating income grew 53%, lifting the operating margin to 55.0% from 39.8%. The shares are down about 11% over twelve months.

That gap is not an argument about the fee stream. It is an argument about what ICE is doing with the cash the fee stream produces, and it has a date: on July 30 the company reported adjusted earnings of $1.90 a share against $1.88 expected and, in the same breath, agreed to buy MarketAxess for $167 a share in cash — a 33% premium, about $5.7bn of enterprise value. ICE fell 1.3% that session, with elevated spending plans and the size of the deal cited as the objections.

What ICE actually sells

Of $2.7bn in second-quarter net revenue, a record roughly $1.35bn was recurring and grew 8% in constant currency: exchange data and connectivity of $416m, up 10%, and Fixed Income and Data Services recurring revenue of $531m, also up 10%. Guidance is for 7% to 8% recurring growth in the data segment.

The per-loan meter is the part that should be broken and isn't quite. Mortgage Technology revenue reached $557m, up 5%, with transaction revenue up 11% as Encompass closed-loan revenue outran industry origination volumes and customers exceeded contractual minimums — share taken in a frozen market rather than a refinancing wave. There is no wave coming soon: the Federal Reserve raised rates in September for the first time since 2023, and the 30-year fixed mortgage averaged 6.95%.

MarketAxess was acquirable because it was losing. In June 2026 it fell behind Tradeweb in US credit electronic-trading share for the first time. "We were late to getting into the execution of fixed income securities," founder and chief executive Jeffrey Sprecher said on the July 30 call, per Markets Media. Tradeweb, the rival that gains a better-capitalized competitor, grew revenue 9.0% to $558.9m last quarter at a 43.9% operating margin and is capitalized at 24.8 times trailing earnings — richer than the acquirer.

CME did not de-rate

CME Group, which runs the futures markets and clearing houses for interest rates, equity indexes, energy and agriculture, is up 5.2% over the same twelve months. Second-quarter volume averaged 29.8 million contracts a day, the second-highest second quarter on record, and market data — billed per screen and per license — was the fastest-growing line at a record $238m, up 20%. The average fee earned per contract was $0.678, down from $0.690 a year earlier but up from $0.652 in the March quarter, which CME attributes to volume tiering and member mix. Investment income on the cash collateral members must post fell 6%, on a lower average rate of return against higher balances — a headwind that a hiking Fed reverses.

"The first half of 2026 was the strongest in CME Group's history," chairman and chief executive Terry Duffy told investors on the July 22 call; chief financial officer Lynne Fitzpatrick, who becomes CEO in March 2027, put first-half volume 10% ahead and revenue up 8%. CME trades at 23.3 times trailing and 22.5 times forward earnings, against 24.0 times its 2024 earnings at that year's close — its price has roughly tracked its earnings, which grew 14.1% on a trailing basis.

Where the de-rating really sits

It is not in everything that sells a data feed. Over twelve months S&P Global is down 25.1%, Morningstar 20.7% and FactSet 17.8%, while Nasdaq is up 1.1%. The catalyst is identifiable: S&P Global fell as much as 13% on 2026 guidance of $19.40 to $19.65 a share against $19.96 expected, dragging peers with it. Morningstar, whose PitchBook and research subscriptions grew revenue 9.6% to $663.2m with operating margin widening to 24.2%, is capitalized at 16.6 times forward earnings. Subscription analytics got marked down on client spending; venues that clear contracts did not.

So ICE's operating record earns the 26.5% recovery from its June 29 low of $122.91. What nothing in the accounts explains is the remaining distance to its own history — 19.2 times this year's expected earnings against 28.1 times at the end of 2025. The honest reading is that the market is charging ICE today for leverage of about 3.4 times gross that arrives only when MarketAxess closes, and discounting a trailing earnings base flattered by one unusually high-margin March quarter. Both objections are dated, and both expire.

The deal is not expected to complete until the first half of 2027. Until then ICE is being priced on a balance sheet it does not yet have.

Target Hospitality, which owns roughly 15,528 beds across 27 communities and rents them by the person to crews working in remote places, told investors on September 18 that a top-five hyperscaler will pay it about $250m through August 2030 to house roughly 1,100 people near a data-center development in the Pecos region of West Texas. The company says it can serve the contract by modifying under-utilized assets for less than $15m of capital.

That announcement, and the guidance raise that came with it — to $435-445m of 2026 revenue and $105-115m of adjusted EBITDA, from $410-420m and $85-95m guided in August — settles an argument about what this company is. It is no longer paid mainly out of appropriated federal funds. Its customer is data-center construction capital.

The mix flipped in about eighteen months

In the June quarter the Workforce Hospitality Solutions segment, which houses industrial and data-center construction crews, grew 142% year on year to $36m, with average utilized beds in that segment passing 4,000. Management guides it above half of consolidated 2026 revenue. Government revenue was $13.5m, up from $7.5m a year earlier on the reactivation of the Dilley, Texas community, whose contract runs to 2030 — growing in dollars, shrinking in mix.

Behind that sit three signings: more than $550m in North Texas on April 1, structured with a guaranteed minimum and extension options; a 48-month agreement announced May 11 expected to generate more than $750m housing about 3,370 people for AI infrastructure work; and the September contract. "Since January 2026, we have secured over 9,000 contracted beds, representing more than $1.4 billion of multiyear contracts," chief executive Brad Archer said on the August 10 call; with September's award the disclosed total exceeds $1.7bn.

The shares have followed the paper, not the quarters. Over twelve months Target Hospitality rose 146.7%, and the three largest up-sessions of that year — 36.4% on April 1, 17.9% on May 11, 8.0% on September 18 — each landed on a contract announcement.

What the contracts have not yet produced

Profit. June-quarter revenue rose 38.7% to $85.5m, but gross margin was 18.5% against 46.1% for full-year 2024, operating margin was negative, and the net loss was $9.0m. Full-year 2025 was the trough being measured against: revenue down 17.0% to $320.6m and a $37.1m loss. Capital spending guided at $490-510m for 2026 exceeds the entire guided revenue, funded by a $660m credit facility and by customers paying up front. "Year-to-date cash flows from operating activities exceeded $110 million and included more than $100 million of advance payments from customers," chief financial officer Jason Vlacich said on August 10. Net leverage was 0.6x at the June quarter; management expects to exit 2027 under 3x.

At 6.1 times trailing sales and 66.8 times trailing EV/EBITDA, the $2.11bn equity is not priced off what has been earned. It is priced off consensus 2027 revenue of $742m and $219m of EBITDA — about 9.6 times that 2027 figure — a ramp the company says its existing contracted portfolio alone supports, before any of a pipeline it puts above 20,000 beds.

The contractor that actually invoices Washington

Maximus, which runs Medicaid and marketplace eligibility, disability assessments and federal case management for governments, fell 37.3% over the same twelve months. Revenue has declined three straight quarters, down 5.1% in the latest, and the company cut fiscal 2026 adjusted earnings guidance to $7.90-8.20 a share after the Department of Veterans Affairs paused performance incentives on its medical disability exam program; it trades near 8.2 times trailing earnings against roughly 16.4 times a year ago. Operating margin still expanded, and U.S. Federal Services margin reached 18.6%.

So the year's move in workforce housing is not a re-rating of companies that bill the federal government — that trade went the other way. It is hyperscaler capital expenditure reaching a supplier small enough for it to matter, and the market has paid for the signature rather than the income statement. The gap between the two is the whole position.

The beds in Pecos are supposed to be occupied this quarter. That is the first of these contracts where revenue, not a press release, becomes checkable.

CF Industries, North America's largest nitrogen fertilizer maker, closed at a twelve-month high of $139.27 on September 2. By then the granular urea it sells had lost about half its value since April, easing to $443 a tonne from $935 as China's exports resumed and Strait of Hormuz fears faded. The rally that carried the shares to that high was not a fertilizer-price rally. It was a gas rally — someone else's gas.

That distinction is the whole stake. CF is paid the gap between a global ammonia and urea price set at the marginal European producer's cost and the Henry Hub gas it burns. Dutch TTF gas reached €78.71 a megawatt-hour on September 9, up 136% on the year after Qatar's force majeure, and European ammonia capacity is cost-bound: when TTF rises, those plants idle and the region buys from the Gulf. CF's own cost, meanwhile, is forecast to move the wrong way. The Energy Information Administration expects Henry Hub to rise from just under $3.50 an MMBtu in 2026 to just under $4.60 in 2027 as Plaquemines, Corpus Christi Stage 3 and Golden Pass add 1.7 billion cubic feet a day of liquefied-natural-gas feed demand. CF's 10-K sensitivity puts a $1.00 move at roughly $33 on the cost of a ton of ammonia and $22 on a ton of urea.

The year was two rallies, and the first one is gone

CF's twelve-month gain of 51.8% was no grind. The shares were down 6.4% by mid-December. Then two sessions after US and Israeli strikes on Iran — March 11 and 12, up 9.2% and 13.2% — delivered $25.95 of the year's $43.60 point gain. That product-price rally has fully round-tripped: over six months CF is up 0.8%, Nutrien down 1.0%, LSB Industries down 26.4%. What is standing is a second, cost-side move dating from mid-June, worth 24.1% at CF and 14.6% at Methanex.

The businesses earned it. CF grew June-quarter revenue 17.6% to $2.22bn and lifted net income 88.3% to $727m; diluted shares fell 5.6%, so buybacks account for roughly a ninth of the near-doubling in earnings per share and the spread did the rest. "Higher global capital costs have structurally raised the incentive price required for new global nitrogen capacity," chief executive Christopher Bohn told investors on the August 6 call. "This is before we factor in any geopolitical premium."

Methanex, the world's largest methanol supplier, swung harder: revenue up 75.2% to $1.40bn, operating income up 356%, on a realized $529 a tonne against $351 in the first quarter and a record 1,027,000 tonnes from Geismar. Only Geismar buys American gas; its Trinidad, Egypt, New Zealand and some Chilean contracts are indexed to the methanol price itself, which cushions the downside and caps the upside — and there the binding constraint is availability rather than price. Methanex idled the 860,000-tonne Titan plant indefinitely on June 29 after failing to agree a gas contract, taking a $115m impairment.

The same gas, two different outcomes

LSB Industries, which makes ammonia and ammonium nitrate in Oklahoma off the identical Henry Hub feedstock, grew revenue 11.1% and watched gross margin collapse to 6.8% from 15.3%, with operating income negative. It carries the group's dearest forward multiple at 18.5x. Cheap gas is necessary and nowhere near sufficient. Nutrien's 37% year, for its part, came from record potash volumes and retail while its first-half nitrogen volumes fell 7%.

CF at 8.5x forward earnings looks cheap only because "forward" means 2026. Consensus has 2027 earnings per share falling 27.7% to $10.84, which is 11.8x today's price, and revenue down 10.6%. The nineteen-analyst average target of $126.11 now sits below the close; Freedom Broker cut the stock to Hold, noting nitrogen prices had returned to pre-conflict levels by June. Methanex trades at 6.7x its 2026 consensus and 11.0x 2027, with a 15.7% trailing free-cash-flow yield and leverage near three times EBITDA that has deferred buybacks — the cheaper peak year, and the one whose multiple has compressed while the shares rose.

The verdict the numbers support: the June quarters are real and the cost advantage is real, but the part of this year still standing rests on European gas staying expensive through a winter, while the consensus underneath both stocks already assumes it does not last. Methanex has told the market the reversion has started, guiding July and August realized methanol to $460–485. CF has not yet had to.

On September 18 the whole complex fell — CF 4.6%, Mosaic 2.6%, Intrepid Potash 3.0% — while the S&P 500 rose, on volumes two to three times normal in a quadruple-witching and index-rebalance session. No company news was discoverable, so mechanical flow is the likelier reading. The date that matters is the next quarterly report, when the gas line inside cost of sales speaks for itself: $3.37 an MMBtu bought CF a 51.5% margin in June, and $4.57 bought it 37.6% three months earlier.

Nucor told investors on September 17 that third-quarter earnings would land between $5.55 and $5.65 a diluted share. Analysts were carrying roughly $6.17. The next session the Charlotte electric-arc-furnace producer — it melts recycled scrap into sheet, plate, bar and structural steel and fabricates joists, deck, racking and tube downstream — fell 7.1% on roughly three times normal volume, its sharpest single-day drop of the year and the first gap in a twelve-month advance that had otherwise been a grind. Its shorter moving average has stayed above its longer one every session for a year.

The guide missed expectations, not the business: the midpoint is about 11% above the $5.04 Nucor earned in the June quarter and more than double the $2.63 it earned a year earlier. What matters is which meter moved, because Nucor is not paid for "the steel price." It is paid the metal spread — realized price per ton minus the cost of the scrap it melts — times tons shipped, against a largely fixed conversion cost of power, electrodes and labor.

The spread, measured

In the June quarter the steel mills sold to outside customers at $1,145 a net ton, up 7% sequentially, while scrap and scrap substitutes cost $422 a gross ton, up 4-5%. Converted to the same 2,000-pound ton, the melt cost about $377, leaving a spread near $768. Output rose into it: a record 7.1m tons shipped and 91% mill utilisation, a second consecutive quarterly record. Group revenue grew 23% year on year while operating income grew 77%, and gross margin has climbed four quarters running, from 11.2% to 19.6%. The mills segment alone earned $1.6bn, nearly double the year-ago quarter. "The demand drivers across the spectrum are incredible," chief executive Leon Topalian said on the July 28 call.

The same rising mill price is a cost inside Nucor. Steel Products earnings fell year on year as higher steel input costs more than offset volume and price; those plants ran at about 65% utilisation against 88% in the mills. The backlog still grew 10% sequentially, with joist and deck visibility into 2027 above realized pricing on warehouse, data-center and advanced-manufacturing work. And the scrap arm, which earned $146m in the quarter, is the one guided lower — Steel Dynamics split its own guide the identical way, steel operations significantly higher on metal margin, recycling lower, and fell 3.6% alongside.

The other side of the heading

ArcelorMittal, the Luxembourg group that mines much of its own iron ore and runs blast furnaces across Europe and the Americas, is on the opposite meter. Its June-quarter EBITDA was $2.1bn, or $155 a tonne, with European EBITDA per tonne at a three-year high of $98 against $70 in March — real, and still far below Nucor's per-ton economics. Reported operating income fell 45% and net income 62%; trailing free cash flow is negative. "The implementation of the new tariff rate quota alongside CBAM is creating a more balanced competitive environment," chief executive Aditya Mittal said with those results. That quota, live since July 1, is expected to cut EU imports from 22.0m to 12.2m tonnes.

That is the whole difference. Nucor's 85% year came with trailing earnings per share up 126%, to $12.54, and its price-to-earnings ratio down to 19.8x from about 24x — below the 23.05x it carried in May. ArcelorMittal's 109% year came with trailing earnings per share down 27%, a ratio at 30.3x, and book value at 1.01x: payment in advance for a 2027 recovery consensus has not yet seen, 16.3x this year's estimate collapsing to 9.7x next year's. Nucor's 13.0x forward carries its own warning — the same consensus has earnings plateauing at $19.64 in 2027 and drifting to $17.44 by 2030.

So the guidance day did not break the spread. Scrap settled flat at $420 in September, hot-rolled coil futures sit above $1,200 and maintenance outages will take about 1m tons out of the market this fall. What broke was the assumption that it keeps widening at June's pace, and the domestic mills had already given ground before the release: Nucor is down 6% over thirty days, Steel Dynamics 5.8%, Commercial Metals 9.1%. ArcelorMittal, whose earnings case is a regulation rather than a spread, barely moved.

Nucor repurchased about 2.03m of its own shares during the quarter at an average of $247.04. The close that followed its own guidance was below it.

Celestica sold $4.70bn of hardware in the June quarter, and the half of its cloud business it designs itself grew more slowly than the half it builds to someone else's print.

That distinction is the entire equity case. Celestica, the Toronto contract manufacturer that builds switches, servers, storage and interconnects for hyperscale cloud operators alongside aerospace, industrial and healthtech electronics, books most of its reported revenue as customer-specified components bought and passed straight through at a few points of margin. What it owns is Hardware Platform Solutions — its own switch and storage designs, sold per unit at materially better margins. In the June quarter that line reached roughly $1.9bn, up 58%, while the Connectivity & Cloud Solutions segment around it grew 84% to $3.81bn. Owned design therefore fell from about 58% of that segment to about half. Gross margin followed it down, to 12.29% from 12.82% a year earlier.

The quarter the shares ignored

Everything else in the print went the other way. Operating margin reached a record 9.75%, net income rose 75%, and full-year guidance was raised to $20.5bn of revenue and $11.30 of adjusted earnings per share — growth of 65% and 87%. "Celestica delivered very strong performance in the second quarter, achieving revenue of $4.70 billion and adjusted EPS (non-GAAP) of $2.54, each exceeding the high end of our guidance ranges," chief executive Rob Mionis said in the July 27 results release. Revenue growth has now accelerated four consecutive quarters, from 26.4% to 62.4%.

The balance sheet is carrying that ramp better than the headline suggests. Inventory reached $3.4bn, up $1.5bn on the year for coming cloud program ramps, yet cash cycle days improved to 47, a nineteen-day gain year on year. The concentration did not improve: three customers accounted for 32%, 17% and 14% of the quarter's revenue, roughly 63% between them.

What a year of gains actually bought

Sales multiples do not compare these companies — a dollar of Celestica revenue leaves 12.29 cents of gross profit, a dollar of Jabil's 9.46 cents. On price to trailing gross profit, Celestica trades at 21.14x, against roughly 22.4x twelve months ago and 26.26x in mid-May. The stock's 32% twelve-month gain was earned entirely by gross-profit growth; none of it came from a higher multiple. On fiscal 2027 consensus earnings the two names are struck at nearly the same price, about 17x each.

Jabil, the St Petersburg manufacturing-services group with 135,000 employees spanning healthcare, packaging, automotive, capital equipment and cloud infrastructure, is the one with a de-rating it earned. Revenue growth halved to 11.8% from 23.1%, and operating income grew 10.4% — slower than revenue, with leverage running backwards. Its diversification is also thinner than advertised: AI-related revenue of about $13.6bn against a roughly $35bn outlook is nearly two fifths of the company. "Our diversified model continues to work, allowing us to support strong growth while also driving higher margins and strong free cash flow," chief executive Mike Dastoor said on June 17. Jabil bought back $291m of stock in the May quarter and has retired 8.7% of its shares since August 2024; Celestica went the other way, selling $3bn of new stock at $310 on August 6, some 15% below the prior close.

One session, six companies

On September 14 all six US contract manufacturers fell together in a 6.09% to 9.24% range with no release from any of them — Sanmina worst, then Celestica, Benchmark, Flex, Plexus and Jabil. It was the day the ten-year Treasury yield touched 5.01%, its first reading above 5% since 2023, and the session on which, as this paper noted, an AI-safety essay knocked a semiconductor gauge down 5.9% while the S&P 500 lost under half a percent. Long yields discount 2027 order books and price data-center project finance. The order book itself has not deteriorated: the largest five hyperscalers are on course for more than $600bn of infrastructure spending in 2026. Celestica's shares are up about 7% over thirty days and roughly 20% above their September 2 low of $277.77; the bearish moving-average reading still attached to them dates from the August equity-offering gap.

So the sell-off is a cost-of-capital event, and the business earns none of it. But that verdict cuts both ways. If nothing in the multiple was expansion, nothing in the next year can come from it either, and the meter that matters is mix, not demand — gross margin has now slipped while revenue accelerated, because rack and assembly volume is outgrowing the designs. Jabil's problem is the reverse and simpler: its multiple rose roughly 10% over a year in which its growth rate halved, and the unwind is arithmetic.

Celestica's new 1.6-terabit switches run on Broadcom's Tomahawk 6 — merchant silicon any competitor can buy. What Celestica owns is the system design wrapped around it, and that is the part growing slower.

Star Bulk Carriers, the Greek owner whose 138 ships carry iron ore, bauxite, grain and coal, is paid a daily rate that has nearly doubled in a year against a daily cost that has hardly moved at all. In the June quarter the fleet earned a time-charter-equivalent rate of $24,486 per vessel per day and spent $5,180 per vessel on crew, stores, insurance and maintenance. After cash overhead, roughly $17,944 a day per ship was left before interest and capital spending.

That subtraction, multiplied by available vessel-days, is close to the whole of a dry bulk owner's earnings. It is why Star Bulk's revenue rose 44.5% to $357.4m in the quarter while operating income rose more than tenfold, lifting the operating margin from 5.3% to 43.4%, and why the segment's shares have risen all year without a single dramatic session. Over the twelve months to September 18, Himalaya Shipping gained 119.5% and Safe Bulkers 98.7%; Star Bulk rose 63.7%, Genco and Pangaea 48.6% each. For four of the five, the largest single day of the year was under 10%.

The leverage, spelled out by the people who own it

Genco Shipping & Trading, a New York owner chartering Capesize and geared vessels to traders and steelmakers, publishes the sensitivity directly. "Every $1,000 fleet-wide TCE increase equates to $16 million of incremental annualized EBITDA or $0.36 per share," chairman and chief executive John Wobensmith told investors on August 6. Genco's cash breakeven is $10,000 a day against a realized $24,273 in the quarter; its dividend rose to $0.80 a share, and Wobensmith projected "another dividend north of $1 per share" for the fourth quarter based on the forward freight curve.

Himalaya Shipping, a Bermuda owner of twelve dual-fuel Newcastlemaxes run by three employees, is the undiluted version: ten of its twelve ships sit on spot exposure. It earned $50,600 a day in the quarter against operating costs steady at $6,500. "The all-in cash breakeven equivalent to the Baltic Capesize Index is about $17,500 per day," chief executive Lars-Christian Svensen said on August 11. Safe Bulkers, a Monaco owner of 40 Panamax and Post-Panamax ships, made the same trade quietly — revenue up 33% to $87.5m, operating margin of 41.1% against 16.0% a year earlier.

The name paid by a different meter

Pangaea Logistics Solutions, a Rhode Island operator that moves grain, bauxite and cement clinker under long-term contracts and runs port terminals in Tampa and Lake Charles, sells cargo first and finds ships after. Its chartered-in day cost rose 24% year over year to roughly $16,816 while its own vessel operating cost was flat. It has locked September-quarter revenue at $20,258 a day across 4,873 booked days against chartered-in days at $17,537 — a spread of about $2,721, capped, at the moment the Capesize spot rate printed $54,791 on September 4. Its terminals added about $4m of revenue, under 3% of the quarter's top line. Pangaea raised its dividend to $0.10; the owners pay out whatever the board does not reserve.

Class mix compounds it. The Baltic Capesize Index averaged $36,000 a day in the quarter, its best since 2021; Supramax averaged $17,000.

What is actually short

The supply case is not an empty orderbook. Bulker orders rose to 11.0% of the active fleet from 9.5% a year earlier, with 285 contracts placed in the first half against 172 a year before, and the Newcastlemax ratio nearly doubling to 34.2%. What is short is delivery slots — Star Bulk says yard availability is limited until late 2029 — and young ships: roughly half the fleet passes fifteen years by the end of 2027. Demand is tonne-miles rather than Chinese steel, which is running about 4% below last year even as Chinese iron ore imports rose 6.3% to 628.9m tonnes in the first half, with Guinea's Simandou adding a haul roughly twice the length of Australia's.

The verdict

The businesses earn the move. Revenue grew between 19% and 80% across the five, margins expanded, and the dividends were declared and paid rather than promised. What the shares also embed is duration, and that is where the evidence thins: consensus already has 2027 earnings below 2026 for every name — Genco 29% lower, Pangaea 33% — which turns headline forward multiples of 7x into something closer to 8x to 14x on 2027 numbers. Genco is the dearest at 10.2x forward earnings with a negative trailing free cash flow yield of -12.8% as it renews its fleet, and at $27.96 it trades above the $27.34 Diana Shipping bid its board rejected in July, so part of its year is a lapsed tender rather than freight. Star Bulk is at 7.1x forward against 12.7x trailing and 1.44x book; Pangaea, whose upside is contracted away, is cheapest at 6.9x; Safe Bulkers carries the lowest price-to-book at 1.07x.

Seventy percent of this year's scheduled Capesize dry-dockings were still pending in August, holding ships off the water into the fourth quarter. Tightness that comes from a maintenance calendar expires when the calendar does.

Modine Manufacturing stops being half a truck-parts company on October 1. The Racine, Wisconsin heat-exchanger maker is handing its Performance Technologies arm — powertrain and battery cooling sold to truck, bus, agricultural and construction manufacturers — to Gentherm in a tax-free spin-and-merge, with a record date of September 28 and Gentherm shareholder approval already secured on September 10. Modine holders keep about 40% of the combined vehicle company, and the remainder will ask to be renamed Modexus Solutions.

What survives is the business everyone said they wanted — data-center cooling and commercial heating and air conditioning — and it has been marked down by more than a third since June. The reason sits inside the June quarter, and it is not an order problem.

Volume doubled, profit did not follow

Group sales rose 28% to $874.1m in the quarter ended June, and group operating income fell 1.2% to $74.8m. The Data Centers segment — Airedale chillers, precision air conditioning, coolant distribution units — sold $348.6m against $183.7m a year earlier, a 90% increase on North American hyperscale demand, close to 40% of the whole company. Segment adjusted earnings before interest, taxes, depreciation and amortization rose 27%. That gap is a margin: 14.8%, against 22.1% a year earlier.

Chief executive Neil Brinker blamed supply-chain constraints flagged a quarter earlier, which he said limited production and temporarily reduced data-center margins. On the remedy — locking in volume with existing partners and qualifying additional suppliers — he added in the July 29 release: "These initiatives are yielding positive results, driving sequential volume and margin improvements as the quarter progressed."

The half being sold is not the drag, and neither is the other half staying. Commercial HVAC grew 22% to $261.6m. Performance Technologies, at roughly $264m of the quarter, leaves at about $1bn, near 6.8x trailing EBITDA, plus $210m of cash to Modine.

The number the price is arguing with

Modine left its fiscal 2027 guidance untouched: 20% to 35% revenue growth and adjusted EBITDA growth in excess of 40%. The first quarter delivered 5%, to $106.5m. Against roughly $456m last year, the guide implies more than $638m — leaving over $532m to be earned in three quarters that produced about $349m a year ago, roughly 52% growth. That arithmetic, rather than any wobble in AI spending, is what a 36.5% three-month decline is disputing.

Demand is contracted and partly prepaid. In May, Modine reserved capacity for more than $4bn of Airedale product through 2029 for one data-center customer, who paid $165m upfront to fund the factories. Two hyperscalers already make up the majority of data-center revenue.

The shares now fetch 13.41x trailing gross profit, against 17.96x in early May, with market value down to $10.02bn from $12.31bn. Forward earnings of 25.1x rest on consensus assembled while Performance Technologies was still inside the company, which ends in under two weeks.

The same trade, one state over

AAON, the Tulsa maker of semi-custom rooftop units whose BasX division builds custom air handlers and data-center cooling plant, is running the identical experiment. June-quarter revenue rose 101% to $627m, BasX sales tripled and total backlog nearly doubled to about $2.0bn. It has raised 2026 sales guidance twice, to 55–60% growth, and cut gross-margin guidance both times, to 25–26%. Overhead at its new Memphis plant ran $18.1m in the quarter against $3m a year earlier. AAON trades at 12.73x trailing gross profit, from 18.35x in May, and its shares are down 43.9% in three months.

What the fall earns

Part of this is paid for. Both companies are absorbing the cost of capacity before the revenue it carries, and Modine's segment margin fell more than seven percentage points while it did — a disclosed deterioration, whatever the order book says. The rest is borrowed from elsewhere: a July credit scare around AI tenants, a 30-year Treasury yield at a 19-year high in August, and the September 14 session when AI-slowdown warnings hit hardware names and liquid-cooling shares fell about 8%. Those repriced the payback period on factories being built today, which is precisely what a prepaid capacity agreement is meant to insulate.

After October 1 there is nowhere for a bad cooling quarter to hide. The vehicular business that smoothed Modine's results for a century leaves with the spin, and the next report is a pure reading on whether the chillers get built at the margin the guidance assumes.