DK Street Journal

Ormat Technologies, which owns and operates geothermal plants in Nevada, California, Kenya and Guatemala and sells their output under long-term power purchase agreements, told investors in August how long it will be before the data-center power bid reaches most of its revenue line. A block of contracted capacity sits priced in the mid-$80s per megawatt hour and cannot be renegotiated until the next decade.

"Between 2031 and 2034, we have approximately 190 megawatts under contract that are currently priced at a weighted average of approximately $86 per megawatt hour, lower than today's market pricing of over $100 per megawatt hour," chief executive Doron Blachar said on the second-quarter call dated August 12.

That is the geothermal annuity in one sentence. Ormat's plants sell substantially all their output under contracts with a weighted-average remaining term of about 14 years, most at fixed prices, and its average realized price fell to $92.6 per megawatt hour in 2025 from $94.3 in 2024 — a decline during the year firm clean power became the most discussed commodity in the electricity business. For the equity, what reprices is the value of an old contract book, on a five-to-eight-year lag.

What the fleet is actually paid

Two channels pull that forward. The first is amendment: on March 10 Ormat signed a "blend-and-extend" deal at its Casa Diablo-4 plant lifting contract pricing about 27% over a 15-year term, effective October 1. The second is expiry, and there is little of it soon: the Amatitlan contract in Guatemala runs to the end of 2027, while Heber 1 was extended 25 years to 2052. Management says over 200 MW is in negotiation above $100 per megawatt hour, against contracts signed over the prior five years at $60 to $80. The hyperscaler business booked so far is small and deferred: up to 150 MW for Google with deliveries expected between 2028 and 2030, and 13 MW for Switch.

So the growth reported in the June quarter came from elsewhere. Electricity revenue rose 5.8% to $169.3m on a full quarter of Blue Mountain, well-field optimization at Olkaria and lower curtailments at McGinnis Hills, Dixie Valley and Tungsten — capacity and availability doing the work of price.

The two books blended into the same line

Ormat's product segment, which sells geothermal turbines and engineering to third-party developers, shrank 21.6% to $46.7m at a 9.7% gross margin, with backlog of $202.8m concentrated in Asia and Oceania. The storage segment, paid merchant rates for ancillary services, grew 195.1%. Consolidated revenue rose 10.6% to $258.8m, yet operating income fell 3.2% and net income 3.4%. Growth decelerated hard inside the year, from 75.8% in the March quarter. Full-year 2025 gross margin was 27.6% against 31.0% in 2024, and consensus has 2027 revenue declining 1.7% with earnings per share down 3.5%.

Against that, Ormat trades at 19.6 times trailing earnings before interest, taxes, depreciation and amortization — roughly double the 10 to 12 times typical for owners of contracted power — 36 times forward earnings, and a trailing free-cash-flow yield of minus 4.7%. The September 8 investor day targeted 3.5 to 3.7 gigawatts by 2030 with $1.0bn to $1.1bn of EBITDA and electricity gross margin recovering to 40%, while placing enhanced geothermal at a gigawatt only between 2033 and 2035. Shares fell 8.5% the next session.

Fervo is a cost curve wearing a power company's clothes

Fervo Energy, the Houston developer that drills horizontal wells into hot rock, has the mirror-image balance sheet: no annuity, and a learning curve instead. Phase I wells at Cape Station in Utah averaged 21 days and 14,483 feet at about $7,000 per kilowatt, with Phase II targeted at $5,500 and per-foot drilling costs down roughly 70% since 2022. Its binding offtake book reached about 1,054 MW when Google signed a 396 MW contract on September 1; the 658 MW signed before it, including 320 MW with Southern California Edison, carries a $7.2bn contracted revenue backlog. None of it is collected yet. Revenue was $113,000 in the June quarter; capital spending was $226.5m, with $850m to $900m guided for the second half against $2.1bn of cash, a May initial public offering of 70m shares at $27.00 and $421m of non-recourse project debt. Consensus does not have Fervo earning a profit until 2029; at 1.43 times book, about $1.87bn of net cash sits inside a $3.90bn market value.

First power came on September 24. "This is a gamechanger for the geothermal industry," chief executive Tim Latimer said that day. It was also one 33 MW block of a three-block, 100 MW Phase I, and the shares fell 9.2% the following session.

A discount-rate month

The September decline was not geothermal's. From August 28 to September 30 Ormat fell 11.5% and Fervo 5.9%, while Brookfield Renewable fell 11.7%, XPLR Infrastructure 10.8% and Clearway 8.2%; merchant generators Vistra and Talen rose. The 10-year Treasury yield reached 5.21% on September 28, a near 20-year high, and decades of fixed PPA cash flows are discounted off exactly that. Ormat's own damage is older and larger: $145.10 on June 3 to $91.19 at the end of September, though only 4.6% below where it traded a year ago.

The verdict splits cleanly. The fall from June is work the business earns — a leveraged generator at double the multiple of its contracted-power peers, negative free cash flow, and consensus modeling a smaller 2027. The last five weeks are the bond market, shared with every owner of long-dated contracts. What nothing in either company explains is the premise that put them on data-center shelves: Ormat cannot sell today's price into today's market at scale, and Fervo cannot bill for anything yet. If Fervo's cost per kilowatt keeps falling while Ormat's realized price stalls, the two are not a hedge on each other — both need the same discount rate, and only one of them is still getting cheaper to build.

On October 1 the amended Casa Diablo-4 contract starts paying Ormat about a quarter more for one plant's output. The rest of the fleet waits until the next decade.

The federal highway program was supposed to run out of authority today. It did not. A continuing resolution signed on 2 September extended the Infrastructure Investment and Jobs Act's surface transportation authorities and pro-rated Highway Trust Fund contract authority at fiscal 2026 levels through 11 December 2026, while cutting enacted transportation funding by a quarter in total — transit down 20%, rail down 82%. Highways were held level and given ten more weeks.

Ten weeks is not a program. No successor multi-year bill has been introduced, and county and state bodies warn that if none arrives, formula funding for highways, bridges and transit reverts to pre-IIJA levels — a $36.8bn annual shortfall — because trust fund revenues cannot carry current spending. The question the quarry operators' shares have been marking down since July was deferred, not answered.

The tons went up

The premise that demand cracked is wrong. Vulcan Materials, the largest US producer of crushed stone, sand and gravel, shipped 59.9m tons in the June quarter, up 1%, with freight-adjusted prices up 3.9% and mix-adjusted prices up 4.7%. Martin Marietta Materials, the number-two aggregates producer, which also sells magnesia chemicals and dolomitic lime, shipped a record 61.6m tons, up 17% on acquisitions and 2.3% organically, at mix-adjusted pricing up 3.7%. Knife River, which consumes its own aggregates in ready-mix and asphalt and bids heavy-civil and paving work for public agencies, lifted mix-adjusted pricing 8% on volumes up 14%.

What failed was the conversion. Vulcan's cash gross profit per ton went from $11.88 a year earlier to $12.02 — fourteen cents, against a price increase four times that size — because diesel cost it nearly $40m in the quarter. "Price is our biggest lever when it comes to overcoming headwinds like this and inflationary pressures," chief executive Ronnie Pruitt told analysts on Vulcan's 29 July call. Martin Marietta's aggregates gross profit per ton fell 17% to $6.78, of which 84 cents was the accounting charge for selling acquired inventory written up to fair value. Operating income fell at all three on revenue that grew.

The control is holding

Knife River is the check on whether project starts are deflating, and they are not: backlog rose roughly $50m sequentially to a record $1.2bn, and the company raised 2026 revenue guidance to $3.4bn-$3.6bn while trimming expected full-year aggregates margin expansion from two percentage points to about one. "With the majority of the construction season still ahead of us, we have opportunities to execute on our $1.2 billion contracting services backlog," chief executive Brian Gray said on the 4 August call. Starboard Value wrote to the board on 24 September demanding a plan for at least 22% adjusted EBITDA margins by fiscal 2029, or a look at strategic alternatives.

The de-rating has been broad and gradual. Vulcan closed at $242.75 on 29 September, down about a fifth over twelve months; Martin Marietta at $479.09, down about a quarter; Knife River at $52.68, the deepest faller. Neither pure quarry operator has a session worse than 3.8% in the last thirty. CRH, Eagle Materials and Cemex all fell roughly a tenth over the same month — this is the heavy-materials complex, not three companies.

Vulcan's price to trailing gross profit has compressed to 14.1x from 17.7x in early May, and it trades at 26.7x forward earnings — 23x the 2027 consensus. Martin Marietta is at 26.4x forward, with a trailing EV/EBITDA of 16.6x that describes a company superseded by the $13.5bn Lhoist North America combination closed on 21 August, which adds $786m of EBITDA. Knife River is the cheapest on the measure that fits a leveraged, low-margin producer, at 10.6x trailing EV/EBITDA.

What the evidence settles

Pricing power is intact and volumes are rising; neither earns the de-rating. What does earn it sits between the two: an input cost that took the entire price increase, and a funding calendar that now runs in ten-week increments. On the first there is finally movement — US on-highway diesel averaged $6.382 a gallon in the week to 28 September, down 14.7 cents and the first weekly decline in twelve weeks. One week does not undo $2.63 a gallon of year-on-year inflation, and consensus already has 2026 EBITDA below 2025 at all three names before a 2027 recovery nobody has authorized highway money for.

None of the three has announced a third-quarter date. Knife River earns its year in two quarters — it lost $84.2m at the operating line in March — and whatever that quarter shows will be public before Congress returns to the December deadline it wrote for itself.

Planet Fitness does not bill a single gym-goer. It licenses its brand to franchisees, who bill the members, and collects a royalty of roughly 7% on the dues they take in — which is why a piece of consumer software turned into the most important thing to happen to the stock this month. Meta released Muse, a personal artificial-intelligence agent, on 8 September; among the chores it will run in the background, with its owner's permission, is cancelling subscriptions that have stopped being used. Two weeks later Planet Fitness fell 9.3% in the single session of 22 September, on a day the S&P 500 tracking fund closed essentially flat and Dick's Sporting Goods rose 8.6%. Bloomberg reported that afternoon that Muse was dragging down stocks that depend on "consumer inertia"; Axios followed two days later, noting that "customers often keep paying for subscriptions long after they stop valuing or using the service."

That one day was 40% of Planet Fitness's decline over the past month, and the four sessions to 24 September accounted for 81% of it. The fall had been read as fallout from the chain's own marketing misfire and from $9.99 rivals — an explanation that fits the year and not the week. What is being repriced is the share of a fitness dollar collected from people who do not use what they pay for, and the three listed ways of collecting that dollar carry very different exposure. Demand itself is not the issue: US gym membership stands at 81 million, an all-time high.

Three meters

Measured from the launch to 29 September, Planet Fitness is down 18.9%, Life Time down 5.9% and Peloton up 0.2%. Planet Fitness's royalty rides on auto-pay dues billed to approximately 21.5m members across 2,930 clubs, and June-quarter same-club sales grew 1.7% entirely on rate, with membership level against March. The company does not disclose visit frequency; outside estimates put the share of members who do not visit in a given month near 60%. Attrition ran 3.5%, mid-range for the chain. The problem is joins.

Life Time, which operates resort-style athletic centers with pools, courts, a spa and cafés, charges average dues of $245 a month and collected $993 of revenue per membership in the June quarter — roughly a quarter of that from training, spa and food bought inside the building, which is hard to pay for and never use. Memberships rose 1.2% to 860,041, comparable-center revenue grew 9.1%, and full-year comparable guidance went up a second time, to 7.9%–8.3%. Revenue rose 13.7% to $866m and operating margin reached 17.3% from 14.2%. "We continue to see strong performance across all aspects of our business," founder and chief executive Bahram Akradi said on the 30 July call. The shares fell 5.6% on 22 September regardless, then recovered to $40.08 by 29 September.

Peloton's $49.99 membership is tethered to a machine the household already bought, so cancelling strands the asset — and Peloton rose 2.8% that day. Its annuity is nonetheless the one visibly shrinking: connected-fitness subscribers fell 8.8% to 2.553m in the year to 30 June, with monthly churn at 2.2% against 1.8%. Price covered the gap; subscription revenue grew 7% to $436.6m, fiscal 2026 was the first profitable year in company history, and fiscal 2027 revenue is guided to $2.3bn–$2.4bn, a decline.

What the businesses earn

Planet Fitness trades at 12.5x forward earnings and 9.6x trailing enterprise value to EBITDA, with an 8.1% trailing free-cash-flow yield, against a 52-week high of $114.26 — about 35x forward. Life Time is on 13.9x trailing EV/EBITDA against consensus 2026 EBITDA growth of 41%, with the catch that its free-cash-flow yield is minus 1.7%: roughly $400m of sale-leasebacks, not operating cash, funds the new clubs. Peloton is on 15.6x forward earnings and an 18.8% free-cash-flow yield on $378m of fiscal 2026 free cash flow.

Planet Fitness's 61% twelve-month fall is earned by its own disclosures — May's cut to roughly 1% comparable growth from 4–5%, withdrawn three-year targets, the $10 promotion restored, and consensus 2026 EBITDA of $611m, down 2.1%. "We are concentrating our efforts this year on two priorities that are central to reigniting net member growth, driving acquisition and reinforcing affordability," chief executive Colleen Keating said on the 6 August call. The September leg is different: operating margin expanded to 33.9% from 30.0%, nothing has been disclosed since August, and JPMorgan reiterated Overweight at $62 on 25 September calling the week an overreaction. The same trade knocked Booking Holdings and SiriusXM down since the launch, which is the point: this is a bet on cancellation behavior across subscriptions, not a verdict on gyms. Life Time's one-day drop is the least explained of the three, and Peloton moving the other way is the cleanest evidence that what was sold was inertia rather than fitness.

Whether idle members actually leave is a fact only franchisees can report, one attrition line at a time, and the first reading arrives with the third quarter.

American Electric Power signed 6 gigawatts of new large-customer load agreements in the June quarter and raised its full-year guidance to $6.25–$6.55 a share. Per-share operating earnings for that same quarter went the other way, to $1.36 from $1.43, while revenue rose 7% to $5.4bn.

Nothing broke. The gap is the mechanism by which a regulated utility actually gets paid for a gigawatt of new load, and it deserves spelling out, because the revenue line makes almost none of it legible. AEP is not paid for "data-center demand." It earns a return that a state commission sets on plant the commission approves. Those allowed returns cluster just under 10%: the Virginia State Corporation Commission authorized 9.80% for Dominion's Virginia Power, Entergy Louisiana asked for 9.70% in its forward-looking formula rate plan, Kentucky allows PPL 9.775%, and NextEra's Florida return is locked at 10.95% through 2029. Each is fixed for years. The equity that funds the plant reprices daily.

The arithmetic of a fixed return

At roughly twice book value — AEP trades at 2.02x, Entergy at 2.48x — a 9.7% return on book equity is something closer to 4% to 5% on what a buyer pays today, against a 30-year Treasury that reached 5.57% in late September, its highest since June 2007. PPL supplied the cleanest measure of the squeeze when it priced $1bn of equity units in February at 7.00% a year against the 9.775% Kentucky lets it earn: under three points of spread, before regulatory lag.

Entergy, which sells electricity to 3 million customers in Arkansas, Louisiana, Mississippi and Texas, is where this shows up in reported results. June-quarter revenue rose 5.9% to $3.5bn and operating income was flat, but the diluted share count climbed 4.6% to 466m and earnings per share fell to $1.03. It is funding a $67bn five-year plan with roughly $7bn of equity, about $4.1bn of it settled or contracted by 30 June. Consensus still has its earnings rising 12.3% this year.

AEP's version is gentler — shares up 2.65%, and a $3bn marketed equity transaction settling under forward contracts by May 2028 that management says covers all anticipated marketed equity needs for the current plan, which targets nearly 11% annual rate-base growth.

Who eats the stranded asset

The obvious fear — that ratepayers or shareholders absorb a half-built plant if a hyperscaler walks — is the part the industry has largely closed. Ohio regulators approved AEP's data-center tariff requiring loads above 25 MW to pay for at least 85% of subscribed capacity for at least 12 years, with an exit fee of three years' minimum charges and collateral requirements; AEP has comparable terms in Indiana, West Virginia and Kentucky. Virginia's new GS-5 class, effective January 2027, imposes a 14-year contract at 85% of contracted transmission and distribution demand and 60% of generation demand whether the building runs or not. In Louisiana, Meta is financing seven new gas plants totalling more than 5.2 GW directly, under an agreement Entergy says pays full cost of service.

"That is why we have led efforts to implement large load tariffs and structure contracts to ensure growth helps pay for growth," AEP chief executive William Fehrman said in the company's 30 July second-quarter release.

The live risk is conversion, not abandonment. Dominion disclosed more than 53 GW in stages of contracting in its June quarter, of which only 12.0 GW sits under firm electric service agreements — about a fifth — against roughly 70 GW of interconnection requests it has described to Virginia regulators. PPL expects about 2 GW of a 31.8 GW Pennsylvania pipeline to be drawing power by 2031. The Ohio Manufacturers' Association is still contesting AEP's demand forecast after the utility halved it.

What the shares earn and what they do not

Fifteen regulated and renewable names fell over the thirty sessions to 29 September, AEP by 3.0% and Entergy by 5.7%, and between roughly half and three-quarters of each loss arrived in three sessions from 21 to 24 September, as the 10-year Treasury yield went from 4.70% on 24 August to 5.18% a month later and the Federal Reserve raised its target range on 16 September to 3.75–4.00%, its first increase since 2023. Over three months AEP is down 13.2% and Entergy 13.1%; over twelve, both are still higher.

So the split. Entergy's de-rating, from 29.94x trailing earnings in May to 25.19x now and 22.68x forward, is earned: it is the most expensive of the three while delivered per-share earnings move backwards on dilution. AEP's is harder to pin on its own results — 18.69x forward against 20.50x trailing, price to gross profit down from 7.17x in May to 5.89x, market value off 12.8% in a stretch when guidance went up. The likelier reading is that the long bond repriced every fixed allowed return at once. Dominion, at 20.99x trailing and 16.91x forward, is not a verdict on Dominion at all: since May it has traded as 0.8138 of a NextEra share, pending approvals in three states.

Turbines are the one input where scarcity still favors the seller: AEP has secured roughly 13 GW of gas-fired capacity in a market where the three dominant makers are sold out into the end of the decade and slot prices are tracking toward $600 per kilowatt. Both AEP and Entergy report third-quarter results on 28 October, and Louisiana's commission is expected to vote on the Meta build in December. The gigawatts are not the open question. Who buys the shares that build them is.

Ceva, which builds no chips at all and instead licenses digital-signal-processor cores, neural accelerators and Bluetooth, Wi-Fi and cellular radio designs to semiconductor makers and device OEMs, collected $10.8m of royalties in the June quarter on 567 million devices shipped by its licensees. The unit count was 16% higher than a year earlier. The royalty was 1% higher.

So the money collected per shipped device fell, from roughly 2.19 cents to about 1.91 cents. That arithmetic is the central question for anyone owning edge inference through intellectual property rather than through silicon: adoption in the camera, the earbud and the car can be entirely real and still not arrive as revenue, because the units shipping today are cheap connectivity sockets — Bluetooth up 16% to 295m units, Wi-Fi up 28% to 80m, cellular internet-of-things at a record 68m — while the expensive vision and AI cores being licensed now do not pay a royalty until the chips carrying them reach production years later.

The license is the revenue

What carried the quarter was the up-front fee. Licensing and related revenue rose 21% year over year to $18.2m, 63% of the total, on ten agreements including two with first-time customers and two struck directly with OEMs — among them a neural-processing-unit license to what Ceva called a leading global AI and computing platform company for a custom silicon program. "We delivered another strong quarter, with revenue increasing 13% year-over-year to $29 million, fueled by licensing and related revenue growing 21% to hit highest level in three years," chief executive Amir Panush said on the August 10 call. Customers, he added, "are adopting broader platforms and deeper collaborations that strengthens both our near-term licensing business and our long-term royalty opportunity."

The operating record underneath is genuinely improving: gross margin of 87.4%, an operating loss narrowed by more than half to $2.08m, revenue growth accelerating for four consecutive quarters to 13.1%, and full-year growth guidance raised to 13-15% from 12%. The shares fell about 10% on the day anyway.

They made it back in a week that had nothing to do with Ceva. Between September 17 and September 25 the stock rose 23.4% — the entire month's gain — alongside Arm up 19.5%, Lattice Semiconductor 15.6%, onsemi 14.0%, Synaptics 12.6% and Microchip 10.7%. indie Semiconductor was the lone edge-adjacent decliner, off 4.0%. No Ceva announcement was discoverable in that window; the nearest event was a StoneX upgrade to Buy on September 14. The likelier reading is sector rotation expressed through the highest-beta name in it. Even after the run, Ceva sits 18.9% below its June level, and its 50-day average has been below its 200-day since September 1.

The per-chip version of the same bet

Ambarella, which sells computer-vision system-on-chips per unit into dashcams, driver-monitoring systems, security cameras and robotics, shows what the other collection method is doing. June-quarter revenue of $108.1m grew 13.2%, the fourth straight deceleration from 31.2%, with GAAP gross margin at 57.7% against 58.9% a year earlier, inventory days up to 157 from 145, and a guide of $115.0m-$124.0m whose midpoint implies about 10% growth. Automotive revenue set a record; the constraint sits elsewhere. Chief executive Fermi Wang said the company is working with customers to make sure they can secure enough memory for fourth-quarter business, calling it "an uncertainty that we are dealing with" — the consequence of AI data centers absorbing roughly 70% of high-end DRAM output this year, with J.P. Morgan Global Research putting DRAM prices up more than 400% from the start of 2024 to the end of 2026. Ambarella buys no memory itself. Its customers do.

Both lose money on a GAAP basis, and gross margins of 57.7% and 87.4% are too far apart for sales multiples to compare, so price against gross profit does the work. Ceva is the cheaper of the two at 8.80x forward gross profit against Ambarella's 11.75x — and Ambarella's reading is above Microchip Technology's 11.10x, a profitable embedded-controller maker consensus has growing revenue 37% this year against Ambarella's guided 10%.

On earnings, the ranking reverses. Ceva trades at 62.9x consensus 2026 earnings of $0.55 a share and 31.8x the 2028 figure — a price already marked to profit two years out, demanding that earnings roughly double, on a royalty line that grew 1%.

What each side earns

Ambarella's twelve-month decline of 16.7% is earned: decelerating growth, a margin that missed its own guided range, and inventory building for a second quarter. Ceva's advance is half earned. Rising licensing revenue at 87% gross margin against a shrinking loss is a real business improving, and the unit base is expanding. What is not earned is the rate: the volume arriving is worth less per chip each year, and the week that produced the move belonged to the sector.

The number that settles it is the one Ceva reports every quarter and nobody quotes — royalty dollars divided by units shipped. Until that figure turns up, the edge is shipping and somebody else is keeping the margin.

The price of a residential broadband connection in the United States is no longer set by the company that owns the wire into the house. Comcast's chief financial officer said so in public this month, and named the offer he will not match.

"So when we see fiber pricing, standalone fiber pricing, in the $30-$40 range for a gig, when we say irrational, that's what we mean by irrational," Jason Armstrong told the Goldman Sachs Communacopia + Technology Conference on September 9. He had already said competition "popped up a little bit in the second quarter" and "continued into the third quarter." The stake is not one quarter of subscriber numbers. It is whether an incumbent charging roughly three times the marginal fiber price can hold either its customers or its rate card, and three companies are answering that question with incompatible business models.

Charter: both meters moved

Charter, which sells Spectrum internet, video and resold mobile service to about 32 million customers across 41 states, lost 172,000 internet customers in the June quarter, against 116,000 a year earlier. The premise that price offsets volume failed in the same release: monthly residential revenue per residential customer was $117.52, down 1.8%, and residential internet revenue fell 3.2% to $5.8bn. Revenue has now fallen year over year for four straight quarters; adjusted EBITDA fell 4.3% and operating margin compressed to 22.6% from 24.4%.

Mobile is growing and not paying for it. Charter added 406,000 Spectrum Mobile lines and grew mobile service revenue 18.9% to $1.1bn on somebody else's network, while consolidated earnings still shrank. Chief executive Chris Winfrey defended the discounting on the July 24 call: "The last thing I want to do … is hamstring the ability of the company to go do some things to accelerate our growth."

The 2.71x forward earnings multiple is a distraction. At 5.53x enterprise value to EBITDA against a $14.9bn equity value, the listed stock is a stub on roughly 13% of the enterprise — implying about $100bn of net debt, before the roughly $12bn of Cox obligations that came with the $34.5bn Cox combination closed August 19-20. Consensus still models 2026 earnings up 14.5%; management guided standalone EBITDA to fall about 1%. Buybacks are paused, and chief financial officer Jessica Fischer — who told investors "Broadband ARPU will improve sequentially in Q3" — leaves on October 15.

Comcast: the conglomerate is not a hedge

Comcast's Connectivity & Platforms revenue fell 3.2% to $19.8bn with segment earnings down 5.8%, broadband revenue per user down 3.8%, and 167,000 domestic broadband customers lost — an improvement of 34,000, still a fourth consecutive quarterly loss. The theory that parks and content absorb the shock failed too: theme-park earnings fell 5.1% to $609m, and Peacock's first profitable quarter contributed $189m.

The shares now fetch 6.23x forward earnings, 4.67x EV/EBITDA and 0.86x book, on a trailing free-cash-flow yield of 26.7%. The mechanism that normally arbitrages that gap is switched off: the $15bn repurchase programme was suspended from July 1 until the NBCUniversal and Sky separation targeted for mid-2027. KeyBanc cut the stock to Underweight on September 25, modelling 2027 broadband losses of 665,000 and framing the trap plainly — match the cheap gigabit and lose revenue per user, refuse and lose customers, "Neither situation is a positive outcome".

Lumen sold the weapon

Lumen left the household altogether, selling its Quantum Fiber consumer base — over 1 million customers and more than 4 million enabled locations — to AT&T for $5.75bn, handing the buyer exactly the network type now undercutting cable. It is now paid by a different meter: hyperscalers prepay for conduit and capacity. Its Private Connectivity Fabric book reached about $13bn of contract value, and in the June quarter it took $476m of cash against $91m of recognised revenue. Reported revenue still fell 9.3%, legacy revenue 15%. Leverage is below 4x and 2026 free cash flow is guided to $1.9-2.1bn, yet Lumen carries the highest EV/EBITDA of the three at 7.75x on the fastest decline, and its forward price-to-sales multiple of 0.51x sits above the trailing 0.48x — the market pricing more shrinkage, not less.

What the shares did, and what it means

Over thirty sessions Charter fell 26.9% and Comcast 20.0%. This was a grind: strip Charter's two worst days and the remaining twenty still fall 17.7%. Verizon, on the winning side of the same price war, is up 7.6% over twelve months, so the money moved within the sector. Canada's regulated payers — BCE, Rogers, Telus — fell roughly half as much.

So the verdict splits. Charter's and Comcast's declines are earned: the price of a connection is being set by a competitor with a cost structure Comcast calls irrational, and New Street Research does not expect cable subscriber additions to turn positive this decade. Lumen's fall is a bet against a growth book no income statement makes legible. The unexplained part is Comcast's price against its own cash generation with the repurchase suspended — the cheapest of the three is also the one that cannot buy itself.

Charter reports on October 30, the first quarter consolidating Cox, which resets the denominator of every per-customer figure it publishes. The number that matters will still be set by somebody else's price list.

Ituran Location and Control, which bills a monthly fee for every vehicle it tracks in Israel and Brazil, has found a second way to get paid. On September 9 it signed the first commercial contract for its traffic-data platform: a software deal with entities of Israel's Ministry of Transport, running up to four years and worth more than 21m shekels, or about $7m, built on anonymised data from more than a million vehicles — roughly a quarter of those in the country. Against 2025 revenue of $359m, spread over four years, it is small. It is also the only new way of getting paid that any of these three companies has produced since early August.

That matters because Zebra, Digi International and Ituran are the standard way a portfolio owns scanned parcels, tracked vehicles and connected machines — and all three reported accelerating revenue for four straight quarters. What separates them is the meter. One sells a box once and waits for the next refresh; one is converting boxes into subscriptions; one has only ever been paid monthly. Which meter is running decides whether this is an enterprise capital-spending cycle with a date on it or an annuity.

Zebra: one sale, then the wait

Zebra sells the barcode scanners, radio-frequency identification readers, thermal printers and rugged mobile computers used in warehouses and stores, through a partner network it puts at more than 10,000 firms. June-quarter revenue rose 20.4% to $1.557bn, but 9.2% of that was organic, the Elo Touch acquisition supplying the rest; the full-year 7% organic guide contains roughly two points of price, putting unit volume growth nearer 5%. Gross margin reached 53.0% against 47.6% a year earlier, helped by $73m of tariffs refunded after the Supreme Court struck the levies down in February. Transportation and logistics — parcel scanning itself — was flat, with the large deployments starting in 2027.

The binding constraint is memory. "In second half, we're continuing to see a challenging and dynamic environment around memory," chief executive Bill Burns told investors on the August 4 call, saying demand signals pointed to the high end of the range while the guide sat at the midpoint. Chief financial officer Nathan Winters described qualifying ten new suppliers, targeting five to seven sources for each memory type. Server memory contract prices are still rising 13-18% quarter on quarter. Third-quarter guidance is $4.70 to $4.90 in non-GAAP earnings a share against $6.35 in the refund-aided second quarter, on about 22% adjusted profitability. At 17.8x forward earnings the stock sits below its 20-25x five-year range, against roughly 14x in mid-July; consensus has revenue growth falling from 15.5% this year to 6.1% next. The 35.2x trailing figure is mostly a GAAP-versus-adjusted artefact rather than a growth expectation.

Digi: the only mix shift

Digi, a $2.9bn maker of cellular routers, embedded radio modules and console servers, reported revenue up 29.0% to $138.7m and annualised recurring revenue of $191m, up 52% — about a third of annualised sales, against roughly 29% of fiscal 2025 revenue. "Growth in ARR reflects achieving ROI for our customers through remote presence and control over their mission-critical and business-critical assets," chief executive Ron Konezny said. Gross margin has climbed to 64.8%. Part of the ramp was bought, through the Particle and Jolt Software deals, and consensus models revenue growth of 8.7% next fiscal year off four analysts. Investors now pay 8.93x trailing gross profit, up from 7.18x in early May, and 29.9x trailing operating cash profit.

Ituran: paid regardless

Ituran's subscription fees rose 25% to $79.8m, 76% of a record $104.8m quarter, with 41,000 net additions taking the base to 2,711,000. That is 6.4% more subscribers than a year earlier, so some 17 to 18 points of the revenue gain is dollar revenue per subscriber: price, mix, and translation the company does not report on a constant-currency basis, having put the effect on operating profit at roughly $1m. New subscribers are coming from motorcycles in Brazil under deals with Yamaha and BMW, not from anyone's capital budget. Operating margin has held between 21% and 23% throughout. It is the cheapest of the three at 14.2x forward earnings and 8.68x operating cash profit, with an 8.2% free-cash-flow yield, $103.7m of net cash and no debt — against Zebra's roughly $2.66bn of borrowings. One analyst supplies its estimates.

What the prices earned

Zebra's 44.6% three-month gain is one session: it rose 26.5% on August 4 and has added 0.3% in the eight weeks since. Digi gapped 14.4% higher on August 6 and trades 7.4% below that close. Ituran is down 18.7% over three months into record results. Over twelve months the dispersion is wider still, Digi up 107.2% against Zebra's 24.6%.

So what binds these three is a mechanism, and their share prices have almost nothing to do with one another. Zebra's re-rating is earned by operating leverage — June-quarter operating income up 75.4% — but the part that depends on a logistics refresh remains a 2027 claim, and the next quarter guides down; the average sell-side target of $383.08 across 13 brokerages sits roughly at the price, even after Needham went to $435 on September 15. Digi is the only one whose revenue quality is measurably improving, and the one paying most for that fact. Ituran is the only business here immune to a deferred refresh, and the cheapest — with the caveat that its per-subscriber gain, the bulk of its growth, is the line no outsider can decompose.

Zebra's second half rests on memory it is still qualifying suppliers for and on parcel carriers spending next year. Ituran's shareholders get 50 cents a share on October 14 either way.

Fair Isaac's mortgage score fee is no longer merely being undercut. It is being written into the pricing machinery of the companies that buy most American mortgages. On the night of Monday, September 28, Federal Housing Finance Agency Director Bill Pulte said Fannie Mae and Freddie Mac are moving to a single mortgage pricing grid that incorporates VantageScore alongside Classic FICO — reaching into the framework used to set upfront fees on the loans they acquire, with no effective date and no replacement matrix published. The next morning TransUnion extended 99-cent standalone pricing for VantageScore 4.0 through the end of 2028. Fair Isaac closed Tuesday at $599.51, down 29% from Monday, on roughly 36 times its normal share turnover.

What is at stake is not software demand. Fair Isaac, which sells credit scores and decision software from Bozeman, Montana, is paid a royalty every time a score is pulled, at a wholesale price it sets by itself and collects through the three credit bureaus and the resellers who assemble tri-merge reports. That mortgage-origination royalty was roughly $284m of the $674.19m the company booked in the June quarter — about 42% of the whole company, against roughly 27% a year earlier.

The price step was collected, essentially in full

Fair Isaac more than doubled the wholesale mortgage price for 2026, from $4.95 to $10 per score, after taking it from $0.60 to $4.95 over the preceding five years. The arithmetic of what followed is unusually clean: mortgage-origination revenues rose 97% year over year in the June quarter, against a list-price increase of about 102%. Units and mix contributed close to nothing. Scores revenue reached $458.9m, up 41%, with the business-to-business line up 49% and the company attributing that primarily to the higher mortgage score unit price.

Chief executive William Lansing said as much about volumes on the July 29 call, and about the policy: "we're not crazy about lenders choice as a policy. We think it's bad policy because it encourages gaming," he told analysts, adding that Fair Isaac was not seeing volume loss "which suggests that they're pulling both scores." He also described rates and affordability as keeping originations below historical norms. The Mortgage Bankers Association's forecast has 30-year rates stuck in a 6% to 6.5% range for several years, which caps the unit leg regardless of who wins the pricing fight.

The company has its own counter-lever. Through its Mortgage Direct License Program it licenses resellers to generate and deliver scores directly, bypassing the bureau markup; Informative Research joined on August 13, and management said on the July call it had signed partners representing about 60% of mortgage volume, with negotiations under way that would take it near 90%. Meanwhile Senator Josh Hawley asked the Federal Trade Commission on March 23 to investigate the increases.

A 14x multiple on an accelerating business

The operating record through June is the opposite of distress: revenue up 25.7%, operating margin 53.8% against 48.9% a year earlier, diluted earnings of $10.45 a share. Fiscal 2025 revenue was $1.991bn with margins that have widened every year since 2021. The stock nonetheless trades at 17.2x trailing and 14.0x forward earnings, against roughly 62x trailing in early January and around 94x at the October 2024 peak. Enterprise value is 14.4x trailing EBITDA and the trailing free-cash-flow yield is 7.7%. Book value is meaningless here — equity is negative, the product of $5.58bn of debt and $7.54bn of treasury stock, including a June term loan of $1.50bn that funded an accelerated repurchase.

The forward figure is where the dispute actually lives. Fiscal 2026 is three quarters reported and close to locked at consensus revenue of $2.541bn. The fiscal 2027 consensus of $52.62 a share, which is what makes the stock look like 11x, still embeds the $10 price.

The shelf fell for a different reason

Nine financial-services software names dropped an equal-weighted 38% over twelve months, and almost none of it is about scores. Q2 Holdings, which sells digital banking platforms to community banks, grew June-quarter revenue 12.6% and widened operating margin by 8.3 points to 13.4%, and is down 25.5%. Tyler Technologies, the public-sector software supplier, grew 8.2% with no pricing controversy at all and is down 38.8% at 25.4x forward earnings. Vertex, which sells sales-tax determination software to retailers and manufacturers and has no lending exposure, grew annual recurring revenue 10.5% to $703.4m with adjusted EBITDA up 33%, yet fell 56.3%; its stumble was the pace of on-premise-to-cloud migration, which forced cloud growth guidance down from 25% to 18%. It now trades at 2.18x forward sales and 3.51x forward gross profit, roughly half Q2 Holdings' 3.92x and 6.88x on similar growth. The common cause is a sector reset: software price-to-sales has compressed from 9x to 6x on fears that AI agents cut licence counts.

So the verdict splits. The shelf's decline is a multiple reset applied to businesses that are mostly still compounding, and Vertex's is closer to a migration-timing problem than a demand one. Fair Isaac's is the only one with a named mechanism pointed at a named line, and the business has not yet felt it — the toll was collected in full in the year just ending. What the market is repricing is fiscal 2027 and beyond: not the count of scores pulled, but whether the seller still sets the price.

That right has never been tested in public. The grid has no effective date, no published fee matrix, and a rival score sitting at 99 cents with a contract that runs to 2028. Fair Isaac's September quarter is not yet reported, and it will describe a year in which none of this had happened.

Alcoa's order book for the rest of 2026 is "almost completely sold out," chief financial officer Molly Beerman told the Jefferies Global Industrials Conference on 10 September, and customers in North America and Europe are "actively looking for our supply." On the same day, the surcharge American buyers pay over the London Metal Exchange price for duty-paid aluminum was assessed back at $1.09 a pound, or $2,403 a tonne. In the fortnight that followed, Alcoa shares fell a further 10.2% and Century Aluminum 7.7%.

That gap is the segment's open question going into earnings. Century, a Chicago smelter of primary aluminum in the United States and Iceland, guided third-quarter adjusted earnings before interest, taxes, depreciation and amortization to $325-345m on an assumed Midwest premium of exactly $1.09. The premium is there. Alcoa — the Pittsburgh producer of bauxite, alumina and aluminum, and the segment's first print — reports after the New York close on Thursday 15 October.

The premium round-tripped

The Midwest premium is largely a policy number. Section 232 duties on imported aluminum stand at 50%, widened in April to apply to the full customs value of derivative goods rather than only their metal content. The premium set a record near $1.19 a pound in June. Then reports on 20 August that Washington would halve the duty on Canadian metal took 8.2% out of the September assessment in one session, to 95 cents, with October and November contracts down more than 12%. The talks collapsed two days later, the duty survived, and the premium worked its way back. Beerman's arithmetic for why it holds even under a deal: the United States must import roughly 4 million tonnes of aluminum a year and Canada can supply only 3 million, leaving a million tonnes that still has to be paid to show up.

Three different meters, one tariff

Century is paid the metal price plus that premium on tonnes it smelts domestically. Its June-quarter revenue rose 19.7% to $752.1m and gross margin went from 5.8% a year earlier to 30.3%, a fourth straight expansion. "Today, both plants are producing at full capacity into a market that needs every unit we can produce," chief executive Jesse Gary said on the 6 August call, after the Mt. Holly restart finished in June. Its largest cost is contracted rather than spot: power at Mt. Holly runs through 2031 under Santee Cooper. Management guided $10-15m of summer energy cost and $20-25m of hedge settlements against the September quarter.

For Alcoa the same policy is a bill. It makes about 900,000 tonnes a year in Canada and is paying more than $1bn in Section 232 duties to bring most of it south; its second-quarter 10-Q states that at recent premium pricing, the premium earned on US production covers the tariff cost on the imports. Roughly 35% of 2026 primary shipments price off the duty-paid premium. Revenue grew 31.4% to $3.97bn with operating income of $718m.

The converters are the check. Kaiser Aluminum, which rolls and extrudes mill products for aerospace, automotive and beverage packaging, is paid a conversion fee and passes metal cost through: record conversion revenue of $437m on shipments up 6%, packaging conversion up 34%. Constellium, the Paris-headquartered and Europe-weighted roller, raised full-year guidance to $900-940m of adjusted EBITDA excluding metal-price lag, hitting its 2028 target two years early. Neither reads like weak end demand.

What the discount is actually for

Every name held an uptrend on 2 June, its 50-day average above the 200-day. Alcoa lost that by late August; Kaiser, down only 4.4% over thirty days, broke on 25 September — and much of its August drawdown came the day it named Fred Stephan chief executive from 1 November, succeeding Keith Harvey.

One caution on the longer window. LME aluminum peaked at $3,854 a tonne on 2 June — the same day Alcoa and Century topped — on Gulf smelter outages, then fell after a tentative US-Iran deal and sits near $3,250. The three-month give-back is metal and policy together. Only the last month is policy alone, with cash metal near a two-month high.

Century now trades at 6.1 times trailing and 3.7 times forward earnings, and 5.3 times trailing EV/EBITDA, against a trailing 140 times in early May; Alcoa at 8.6 and 6.5 times, against 16 times then. Kaiser's trailing and forward multiples are both about 10.7, meaning nothing is being paid for consensus 2026 earnings per share of $14.07 against $6.77 reported for 2025.

The verdict the numbers support is narrow: the market is not pricing a demand recession, and it is not pricing a premium collapse either, because the premium came back. It is pricing duration. Consensus itself has 2027 earnings per share below 2026 at Alcoa, Kaiser and Constellium — 17.7% lower at Kaiser — and only Century higher. B. Riley cut Century's target from $86 to $74 in late September. What nothing in the fundamentals explains is the further leg down since the premium recovered; the 18 September session, when all four traded three to four times normal volume with no discoverable company news, looks more like a quarterly rebalance than information.

Century's guide is the cleanest test anyone has printed, and it will not be tested until early November. Alcoa gets there three weeks earlier, with a sold-out book and a billion-dollar duty bill it says the premium already pays.

Flex gave its planned data-center spin-off a name on 15 September and filed the paperwork to list it. The business will be called Axiom Solutions International, is expected to trade on Nasdaq under the ticker AXM, and the tax-free separation is targeted for the first quarter of calendar 2027. Flex — a Singapore-founded manufacturing-services group of 149,686 people whose plants turn out switchgear, busway, power-distribution units and liquid cooling alongside cars, medical devices and consumer goods — will be left running the assembly work.

What gets separated is the answer to the question hanging over every contract manufacturer in the artificial-intelligence build-out: whether the work is being repriced, or merely multiplied. A contract manufacturer books the customer's component cost as its own revenue and earns a few points on top. Conventional memory contract prices rose 90-95% quarter on quarter in the first quarter of 2026 and a further 58-63% in the second, and memory's share of a reference server build has climbed from about 18% to roughly 53% of the bill of materials in eighteen months. Revenue growth built out of that inflation arrives at near-zero incremental profit. Gross profit dollars are the honest meter.

The meter reads three different things

Celestica, the Toronto group that designs and builds switches, interconnects and full racks for hyperscale cloud operators, grew June-quarter revenue 62.4% to $4.70bn — the fourth straight quarter of acceleration — while gross profit rose only 55.7%, pulling gross margin to 12.29% from 12.82%. "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 27 July results release. The dilution is mix: its owned-design Hardware Platform Solutions line reached roughly $1.9bn, up 58%, inside a cloud segment that grew 84% to $3.81bn, so owned design fell from about 58% of that segment to about half. Segment margins still widened, and operating margin hit a record 9.75%. Celestica raised 2026 guidance to $20.5bn of revenue and $11.30 of adjusted earnings, funding the ramp with a $3bn equity sale priced at $310 in August.

Sanmina did the opposite. The San Jose builder of backplanes, enclosures and cable assemblies grew revenue 69.7% to $3.46bn and gross profit 100.7%, widening gross margin to 10.49% from 8.87%. The acquired factories are the reason: inside the segment where the ZT Systems manufacturing business sits — bought from Advanced Micro Devices for $2.55bn in cash plus contingent consideration, with AMD keeping the design engineers — gross margin widened 270 basis points to 10.2%. Cloud and AI infrastructure is now 62% of Sanmina's revenue. The cost shows in cash: free cash flow of $24m on $3.46bn of sales, with inventory up 87.2%. Management told the June-quarter call that an engineering benefit helping margin ramps down after the fourth quarter and that working capital will build further.

Flex's gross profit grew 30.6% on 20.6% revenue growth, and its Cloud and Power Infrastructure unit grew 35% to $2.2bn at a 9.7% operating margin — roughly double the parent's 4.94%. Chief financial officer Kevin Krumm told the fiscal first-quarter call that segment revenues are expected to grow 65% to 75% this fiscal year.

What the rebound paid for

The group has recovered about a tenth over thirty days, but from a hole: on 14 September all six US contract manufacturers fell together between 6% and 9%, with no company news, the day the ten-year Treasury yield breached 5% for the first time since 2023. Celestica gained 19.4% over the month, Sanmina 13.4%, Flex 2.3%. Every one remains below its June high, Flex by 30.3%.

Judged against the gross profit each business actually earns, Celestica costs 21.99x trailing, against 26.26x in mid-May and roughly 22.4x a year ago — its twelve-month gain was earned by growth rather than by a higher multiple. Sanmina costs 10.26x, close to Jabil's 10.37x ahead of Jabil's fiscal fourth quarter on 30 September. Flex costs 15.04x, down from 21.42x in mid-May, while its gross profit grew. Its 24.04x forward earnings sit against 43.64x trailing, with consensus revenue accelerating.

So two of the three are being paid for margin and one for throughput, and the market has inverted that. Sanmina's re-rating is earned at the gross line but financed on the balance sheet. Celestica's operating record is real, while the designed content its premium assumes is shrinking as a share of the work. Flex is the only one where the de-rating has no equivalent in the numbers — the likelier explanation is the shrinking automotive and consumer book dragging on a business whose fastest, richest unit is about to leave it.

That is what the separation ends. Once Axiom lists, power and cooling will carry its own price and assembly will carry its own, and neither can be averaged into the other. Flex has promised the strategy detail at an innovation day in Austin on 10 November.