DK Street Journal

Nova, which sells process-control metrology systems — the film, materials, chemical and dimensional measurement tools chipmakers run between the steps that build a wafer — reported record revenue of $255m before the Nasdaq open on 6 August and guided the September quarter higher still. Its shares closed down 6.2% that day, and by 10 September stood 39% below their 15 June close of $605.65.

Process-control metrology is where the artificial-intelligence construction budget first becomes a measurable order: before memory stacks ship, someone buys the tools that inspect and measure them. Three months of selling have priced Nova, Camtek, Onto Innovation and KLA as a single trade. They are not one business. Nova's revenue tracks wafer-start capacity and the industry's move to gate-all-around transistors at leading-edge logic; Camtek's tracks back-end advanced packaging lines. Over the summer both de-rated hard — one into accelerating numbers, the other into decelerating ones.

Nova: records, and a lower price

Nova's growth has been speeding up, from 14.3% year over year in the December quarter to 10.3% in March and 15.9% in June. Operating margin was 29.9%, against 29.8% a year earlier, so the growth arrived without leakage. Guidance for the September quarter of $277-287m implies roughly 25.5% growth on the $224.6m booked a year ago. The company attributed the quarter to advanced logic on the gate-all-around transition and to record advanced-packaging revenue from capacity increases across logic and memory — two demand clocks inside one income statement. "This quarter marks a significant milestone for Nova, surpassing $250 million in quarterly revenue and $2.50 in quarterly non-GAAP EPS, reflecting the run-rate envisioned in our long-term strategic plan," said Gaby Waisman, president and chief executive, on 6 August.

Camtek: the order book and the P&L disagree

Camtek, a 709-person company in Migdal HaEmek that sells automated optical inspection and 2D/3D metrology systems into back-end packaging, has the opposite problem. June-quarter revenue rose 8.0% to $133.2m, operating margin fell to 20.4% from 25.9%, and operating income declined 14.8% — the second straight quarter of falling operating profit. Gross margin was flat at 50.1%, so the compression came from operating expenses rather than a slide back toward cheaper older platforms; its newer Eagle G5 and Hawk systems were half of systems revenue.

The forward-looking disclosure says something else entirely. More than $600m of orders have been received year to date, about 80% of them advanced packaging, with deliveries stretching into 2027 — more than four quarters of work against a $593m consensus revenue estimate for this year. Chief operating officer Ramy Langer told investors on the 10 August call that more than 20% of those orders came from high-bandwidth-memory manufacturers and more than half from outsourced assembly and test houses. Chief executive Rafi Amit said about 75% of second-quarter revenue came from advanced packaging, mostly AI-related. China is guided to about 45% of 2026 revenue, down from 49%, with the company's own risk disclosures naming further US export controls as a threat to that business.

What the group's numbers hide

Camtek trades at about 40x forward earnings and 29.5x the 2027 consensus, against Nova at 34x forward and 26.8x 2027, Onto at 33x and FormFactor at 37x. The slowest-growing of the four is the dearest. FormFactor, meanwhile, explains most of the group's apparent boom: its shares are up 286% over twelve months against Camtek's 79% and Nova's 33%, and it sells probe cards — consumables billed per test insertion, tracking wafer starts rather than capacity orders. Its June-quarter gross margin was 50.7%, up from 37.2%, and it guided the current quarter to $270m give or take $10m.

The selling itself was concentrated and shared. All four names peaked on 17 August and bottomed on 1 September; on 18 August, the 30-year Treasury yield topped 5.33%, a 19-year high, and Camtek fell 11.9%, Nova 9.1%, Onto 10.2% and KLA 5.3% in that one session. Since 1 September, Camtek has recovered 13.2% and Nova 9.2%.

The verdict

Camtek's de-rating is earned in part: two quarters of falling operating profit at a premium valuation is a fact, and consensus still asks for a sharp reacceleration to reach $593m this year. What nothing in Camtek's reported P&L supports is the order book, which says the slowdown is timing between shipment dates. Nova's case is cleaner and stranger — revenue accelerating, margins steady, guidance up a quarter year over year, and a share price a third lower than in June. The likelier reading is a discount-rate shock compressing a 57x multiple, not a demand signal. At 34x forward, that cushion is gone in both directions.

What sits under both order books is memory capacity that does not exist yet: Micron is adding as many as 60,000 high-bandwidth-memory wafers a month toward roughly 100,000 by year-end, capacity that ships product in 2027 and buys tools now. Micron reports on 30 September, and it will be the first outside read on whether the tools are still being bought.

NuScale Power's June-quarter revenue was $75,000. A year earlier it was $8.1m, and the entire difference is one contract ending: management attributed the drop to "completion of Fluor front-end engineering design Phase 2 work for RoPower in late 2025 with no comparable activity in current quarter." The company that owns the only small modular reactor design certified by the Nuclear Regulatory Commission was, for one quarter, billing less than a suburban dental practice.

This is the shape of the problem in listed nuclear. NuScale and Oklo are the two most-owned equity claims on data-center power demand, and neither company's income statement is a claim on electricity. The meters that matter are licensing dates, quarterly cash consumption and the share count — and over twelve months NuScale has lost 70.1% and Oklo 45.2% while the S&P 500 gained 16.2%.

NuScale sells the design and pays for the customer

NuScale supplies technology; someone else sells the power. Under a 50/50 joint venture, ENTRA1 Energy holds global exclusive rights to commercialise and deploy NuScale's 77-megawatt modules, developing, financing and owning the plants. Money flows toward the partner: NuScale paid ENTRA1 $495m as "Milestone Contribution 1" against the Tennessee Valley Authority framework for up to 6 gigawatts, with more than $3bn of further milestone payments possible. That single payment produced a $538.4m operating loss in the third quarter of 2025, against revenue of $8.2m in the same three months.

The TVA power purchase agreement is still unsigned; chief executive John Hopkins has said he expects it finalised by the end of 2026. Until then, the offsetting assets are real but static: two standard design approvals, conventional low-enriched uranium fuel rather than the high-assay fuel rivals need, forgings ordered from Doosan Enerbility, and $1.9bn of cash against operating expenses near $50m a quarter. Financing continued anyway. Class A shares rose from 318.5m at year-end to 410.4m by 30 June, and on 11 August the company launched a $750m at-the-market programme.

Oklo builds, owns and has not yet signed a price

Oklo's model is the mirror. It plans to own its Aurora powerhouses and sell electrons under long-dated contracts, which makes its forward case a licensing calendar and a fuel supply rather than an order book. The calendar has held: the NRC approved the Aurora principal design criteria in May on an accelerated schedule, the combined licence application is in review, and the Groves isotope test reactor in Lockhart, Texas reached first criticality on 5 August, which management called the fastest greenfield-to-criticality transition "for a full-scale, privately funded and privately sited reactor in history" on the 7 August results call.

What that reactor makes is radioisotopes. The roughly 14 gigawatts of announced demand — a 12-gigawatt Switch agreement plus letters of intent with Equinix, Diamondback Energy and Prometheus Hyperscale — remains, by Oklo's own filings, preliminary arrangements it is still negotiating into power purchase agreements. Fuel is similarly provisional: a Centrus letter of intent for up to five powerhouses with deliveries expected in 2029, plus a pending Department of Energy plutonium allocation. Meanwhile the bill grew — 2026 operating cash guidance was raised to $120–150m from $80–100m and capital spending to $400–500m — funded by $1.9bn of stock sales at a realised price that fell about 35% between the first and second quarters.

What the de-rating is actually pricing

The two have separated recently. NuScale is up 7.2% over thirty days, including a 15% session on 8 September when it and partner MillenniTEK produced first-of-a-kind boron-oxide pellets, a manufacturing demonstration rather than an order. Oklo is down 12.3%. On 10 September Piper Sandler initiated with a Buy on Oklo and a Sell on X-Energy, and the whole complex fell about 5%, selling the Sell rather than buying the Buy.

The year-long de-rating is not a milestone failure — Oklo's schedule is intact and NuScale's design approvals stand. It is a repricing of duration. With the 30-year Treasury yield near 5.27%, cash flows dated 2028 and beyond are worth less, and neither price is anchorable to earnings in any case. Oklo trades at about 2,337 times forward sales and 2.15 times book, with roughly $3.0bn of cash inside a $7.5bn market value. NuScale screens at 160.7 times forward sales only because the reported diluted share count is stale; on the fully exchanged Class A and Class B counts, the market value is nearer $4.4bn and forward sales roughly 233 times, with $1.9bn of that cash.

So the honest split: Oklo's fall is the cost of capital rising against an unchanged plan, and its equity is now partly a cash pile with a construction site attached. NuScale's is the plan itself going quiet — one signature away from a combined licence application, engineering contracts and OEM negotiations, and with no revenue worth the name until it arrives.

The fastest reactor to criticality in modern American history went critical last month to make medical isotopes. The one certified to make power for the grid is waiting on a piece of paper from a federal utility.

Supernus Pharmaceuticals, a Rockville, Maryland developer of central-nervous-system medicines, told investors on 3 August that 2026 revenue would land between $860m and $890m, raising its own forecast after a June quarter in which sales grew 32.4% to $219.1m. The same morning it announced something that made the guidance largely irrelevant to its shareholders: an all-stock merger of equals with Indivior, in which every Supernus share converts into 1.5401 Indivior shares and Indivior holders take about 56.5% of the combined company.

A fixed exchange ratio is a wire, and it has been carrying current in one direction. From 5 August to 10 September Supernus fell 12.0% and Indivior fell 12.2%. Rebuilding Supernus's price each day as 1.5401 times Indivior's close, less the roughly $7.75 a share of the $1bn special dividend Indivior holders collect immediately before the exchange, reproduces Supernus within about two percent through mid-August and within five percent since. Qelbree prescriptions no longer price this equity. Buprenorphine does.

The business the price left behind

Supernus's own quarter was the best in its recent history. Qelbree, its non-stimulant attention-deficit treatment, grew net sales 15% to $89.2m on 264,545 prescriptions, up 17%. The four products management counts as growth assets — Qelbree, GOCOVRI, ZURZUVAE and ONAPGO — together reached $175.7m, up 52%, enough to bury the decline in the legacy epilepsy drugs Trokendi XR and Oxtellar XR.

The reported loss of $58.4m is purchase accounting and a write-down, not trading: a non-cash $54.9m impairment of the APOKYN intangible plus $25.3m of acquired-intangible amortization, most of it on ZURZUVAE from last year's Sage transaction. Gross margin, at 84.6%, slipped from 89.8%.

What Qelbree does not have is a protected socket. It competes against generic atomoxetine in a market where stimulants take about 76% of drug-class revenue, so growth is bought with sales representatives and payer coverage. Collegium Pharmaceutical, the Massachusetts pain specialist that cut its own 2026 guidance three days after Supernus raised its, is buying the same prescribers: Jornay PM grew 41% to $46.1m in the June quarter, took 29.2% of the branded long-acting methylphenidate market, and Collegium now fields 190 attention-deficit representatives. Collegium's shares are down 39% over twelve months and trade at 2.5x trailing enterprise value to earnings before interest, taxes, depreciation and amortization.

What sets the price now

Indivior, based in North Chesterfield, Virginia, sells buprenorphine medicines for opioid dependence and had its strongest quarter on record: revenue of $343m, SUBLOCADE — its monthly injectable — at $253m, up 21%, adjusted EBITDA up 111% to $186m, and full-year SUBLOCADE guidance lifted to $1.01–1.05bn. "Every fundamental metric in support of SUBLOCADE is trending in a strong direction," chief executive Joe Ciaffoni said on the 4 August merger call. "In the quarter, we had record new patient starts. We've seen stable market share at 76%."

The shares got cheaper against all of it — 8.1x forward earnings against 12.1x trailing. Supernus, by contrast, is at 16.2x forward earnings and 3.29x trailing gross profit; the comparison runs on gross profit because gross margins under this label differ too widely for sales multiples to mean anything.

The verdict

Nothing company-specific was discoverable to explain a steady slide that began the week both companies raised guidance. Because Supernus trades at a small premium to dividend-adjusted parity rather than a discount, the likelier reading is not deal-break risk but a re-rating of the currency itself: Indivior's holders are diluting a business earning a 46% adjusted EBITDA margin into one earning 18%, then paying themselves $1bn of it — partly with a $650m Citibank term loan — on the way out. Indivior's own merger filing says the dilution is real and the benefits are not guaranteed. Management targets $125m of annual cost savings within a year of closing and has given no 2027 forecast.

Until the ratio is broken, Supernus's earnings reports are informational. The merger is due to close in the fourth quarter, subject to two shareholder votes whose dates have not been set — and the last legacy opioid settlement instalment, $50m, falls due in December 2027, inside the combined company's first full year.

The docket changed subject

Virginia's State Corporation Commission this week ordered three in-person public hearings across Dominion Energy's service territory in the case that would hand the state's largest electric utility to NextEra Energy, an unusual addition to three days of telephonic testimony already scheduled. For the company whose wires serve the densest cluster of data centers on earth, the live regulatory question is no longer what it will be allowed to earn on what it builds. It is who will own it.

The distinction has already emptied Dominion's fundamentals of any effect on its share price. On 18 May NextEra agreed to buy Dominion in an all-stock deal at a fixed ratio of 0.8138 NextEra shares per Dominion share, with a one-time $360m cash payment at closing and completion targeted for the second half of 2027. Both shareholder bases approved the $66.8bn combination on 3 September. Rate base, allowed return, the new large-load tariff, an offshore wind farm four-fifths built — all of it now informs an arbitrage spread rather than a valuation.

Nine-tenths of a decline that belongs to someone else

Over the thirty days to 10 September, Dominion fell 3.9% and NextEra fell 3.6%. The ratio between the two closes moved from 0.7919 to 0.7894 — three-tenths of one percent. On 15 May, before the announcement, that ratio was 0.6612; three sessions later it was 0.7588. At 0.7894 it sits roughly 3% under the 0.8138 the deal promises, which is the market's combined price for deal risk and the wait. Dominion has not been in a downtrend at all this summer; its 50-day average has stayed above its 200-day throughout.

The clock is regulatory. The companies filed their joint petition with the Virginia commission on 15 July under the Utility Transfers Act, and the 60-day statutory deadline falls on 13 September, extendable by 120 days to 11 January 2027.

What the deal froze in place

Dominion's June quarter showed exactly the squeeze the merger now obscures. Revenue rose 19.6% to $4.558bn while operating income fell 2.4% and net income fell 55.3% to $340m. The company reported operating earnings of $0.79 a share against $0.37 on a reported basis and reaffirmed full-year operating guidance of $3.45 to $3.69. Virginia's commission last year raised the authorized return on equity only to 9.8% from 9.7%, against 10.4% requested, and cut the 2026 base-rate increase to $565.7m from the $822m sought. The demand itself is less contracted than the headlines suggest: Dominion discloses more than 53 gigawatts of data-center capacity in various contracting stages but about 12 gigawatts under executed service agreements. The path from load to revenue runs through a new rate class for customers above 25 megawatts, effective January 2027, requiring minimum payment on 85% of contracted transmission and distribution demand and 60% of generation demand. Coastal Virginia Offshore Wind is 81% complete, and its levelized cost has risen to about $62 a megawatt-hour from $56, with owners bearing half of any costs above $10.3bn under the 2022 settlement. At 22.6x trailing and 18.2x forward earnings on consensus growth of 5% this year, none of that clears the price. NextEra does.

The one still trading on itself

Entergy, the New Orleans utility serving about three million customers across Arkansas, Louisiana, Mississippi and Texas from roughly 26,000 megawatts of capacity, about 6,000 of it nuclear, is the remaining unencumbered version of the same demand story — and it has been rolling over since late August, down 3.8% across three months. Its business went the other way. Second-quarter revenue rose 5.9% to $3.524bn and net income 3.4% to $487.8m; adjusted earnings slipped to $1.03 a share from $1.05 as weather normalized, and 2026 guidance of $4.25 to $4.45 was affirmed. Industrial sales grew 10% excluding weather. "We continue to have seven to 12 gigawatts of hyperscale data center potential in our pipeline, as well as three to five gigawatts of interest from traditional industrial segments," chair and chief executive Drew Marsh told investors on 29 July.

The credit and regulatory meters the bear case needs are not cooperating. Entergy Louisiana reported an as-reported return on equity of 11.0% for the twelve months to March in its formula-rate filing — earning above what Dominion is allowed. Funds from operations to debt is running at or above 15% and is projected to stay above S&P's 13% threshold through 2030. Under the Meta agreements, the customer finances more than 5.2 gigawatts of new gas plants and roughly 240 miles of 500-kilovolt transmission while Entergy keeps the regulated assets; Louisiana rate base is expected to roughly double to near $50bn by 2030 inside a $67bn five-year plan.

What that plan costs is the real story in the multiple. Diluted shares went from 431.6m in 2024 to 466.3m by the June quarter, about 8% dilution in eighteen months, including a $2.175bn equity forward in May. The 30-year Treasury yield touched about 5.32% in mid-August, its highest since 2002, and Duke Energy sold $1.75bn of equity units that month at a 7.75% all-in distribution rate against a 9.8% allowed return — roughly two points of spread between what new capital costs and what regulators permit.

The verdict

Entergy's decline is a de-rating, and an honest one, but nothing in the operating record earns it: revenue, industrial volumes, guidance and credit metrics all improved or held. The multiple is falling from 29.9x trailing in May to 26.9x now, 24.2x forward, against a long-run median near 16.8x for vertically integrated utilities — which means the compression is arithmetic catching up to an expensive starting point and a higher discount rate, not a verdict on Meta's megawatts. Consensus still has earnings accelerating 12.3% this year and 16.1% next.

One more correction is worth making about the group these two are filed under. Its 4.3% average thirty-day decline is one broken name: PG&E fell 16.5% after California's legislature passed a wildfire bill stripped of the liability protections utilities expected. Excluding it, the other seven averaged a 2.6% loss.

Virginia spent years litigating what Dominion may charge the data centers. It now has until January, at the outside, to decide who collects.

Talen Energy books much of its power years before it delivers it, which is why its most recent quarter shows a 64.5% jump in revenue and a $92m net loss on the same page. The Houston-based independent power producer sells electricity, capacity and ancillary services from roughly 10.7 GW of nuclear, gas, coal and solar generation, and on August 5 it raised 2026 guidance to $2.025bn–$2.225bn of adjusted earnings before interest, taxes, depreciation and amortization. The shares are down 22% over twelve months and slipped into a downtrend on September 3, when the 50-day average crossed below the 200-day.

What is at stake is the roughly $4bn of cumulative adjusted free cash flow Talen expects across 2026 through 2028, and whether a power market that keeps setting records can reach it. On the two legs that carry most of its margin, for now, it cannot.

Sold forward, and capped

At June 30, including the nuclear production tax credit, Talen had hedged about 85% of expected 2026 generation, about 70% of 2027 and only about 30% of 2028. Management cited West Hub spark spreads up nearly 50% year on year in its second-quarter materials — a move that reaches reported income as mark-to-market noise rather than as revenue. The accounts show it as violence: June-quarter gross profit rose 291% to $368m while operating income was minus $72m; the December 2025 quarter paired 58% revenue growth with a $363m loss.

The second leg is set by administrators. PJM's 2028/29 base residual auction cleared at $325 per megawatt-day across the footprint — the preset cap, for the third auction running, and 2.5% below the prior year's $333.44 ceiling. PJM's own simulation put the uncapped price at $554.72, with the auction still leaving a 6.8 GW shortfall against the target reserve margin. Talen keeps 98% of its generation in PJM, per Mizuho's August 24 initiation at Outperform. Scarcity worsened; its capacity revenue per megawatt-day went down.

Even the celebrated data-center contract is a volume schedule rather than a price bet: the Amazon agreement runs to 2042 at up to 1,920 MW from Susquehanna, ramping to 840–1,200 MW in 2029, with roughly $18bn of revenue expected at full quantity. "We believe blending new capacity with existing energy on a front-of-the-meter grid-connected site is more reliable and durable, and in fact, less expensive than any behind-the-meter solution," chief executive Mac McFarland told investors on the second-quarter call.

The per-share number is a financing decision

Talen closed the ~$3.5bn Cornerstone purchase of about 2.6 GW of PJM gas plants, funded with roughly $2.6bn of cash and 2,399,998 shares issued to Energy Capital Partners, $983.5m of which was registered for resale in June with lock-ups beginning to expire in mid-September. Against that, some 15 million shares have been repurchased for about $2.3bn since 2024, taking the diluted count from 53.17m to 45.90m. Management targets net leverage below 3.5x adjusted EBITDA by year-end.

The decisive session was August 18, when Talen fell 11.4% on 2.17m shares — attributed to a post-earnings reset, the share registration and rotation out of power names — while Constellation fell about 4%. Over three months Constellation is up 16.1% and Vistra 3.2%, so the complex is not de-rating together.

The other name in the group

TransAlta, the Calgary generator sorted into the same industry classification, is a different trade entirely. Alberta spot power averaged C$29 per megawatt-hour last quarter against C$40 a year earlier, its own hedges realized C$63, and it reaffirmed C$950m–C$1.05bn of adjusted EBITDA. Its shares are down 3.6% over twelve months and its trend has been flat since late July. At 11.3x trailing enterprise value to EBITDA it sits between Vistra and Constellation; at 38x forward earnings on profits guided down 40%, it is dear.

The verdict

Nothing in Talen's reported business explains a 22% decline. Revenue accelerated, gross margin more than doubled, cash guidance went up, and the shares now fetch 15.2x 2026 consensus earnings against Vistra's 17.6x and Constellation's 24.8x, and 9.0x trailing gross profit against the 16.5x recorded in early May while that gross profit grew. What the business does explain is why record power headlines never showed up in the numbers investors were repricing: the hedge book fenced 2026 and 2027, and the capacity cap took the rest. This was a valuation reset rather than an earnings one — but a company whose reported loss is a hedge artifact has few defenses when sentiment turns, because its trailing earnings multiple offers no floor to argue from.

The part of Talen a rising power curve can still reach is 2028, about 70% of it unsold. Shareholders wait two years to find out what it fetches.

New York has become the first state to freeze new data-center construction for a year, and around the country roughly $64bn of projects are now blocked or delayed by local, bipartisan opposition — some 50 active local moratoriums, with Loudoun and Spotsylvania counties, the heart of Data Center Alley, weighing bans of their own. None of this existed in the July earnings reports that set expectations for the two big listed landlords of computing space.

It matters because it changes which side of these businesses is scarce. Digital Realty, which rents wholesale halls by the megawatt to cloud and enterprise tenants across 309 facilities, and Equinix, which sells cabinets and cross-connects inside its interconnection-heavy centers, both earn money two ways: rent on the buildings they already own, and returns on the buildings they are still constructing. A permitting freeze raises the value of the first and threatens the second.

The buildings that already exist

The supply data says the first effect is real. CBRE put North American primary-market inventory at a record 10,903 megawatts in the first half, up 33.7% year on year — and vacancy still fell, to 1.4%. In Northern Virginia it reached about 0.24%, with 10.8 megawatts available and contiguous blocks of 5 to 10 megawatts hard to find at all. Power, land and permits, not tenant appetite, are the gating item, and leases there are shifting toward triple-net structures and "pay-or-lose" power floors.

That pricing power shows up in the quarters. Digital Realty renewed $262m of annualized cash rent in the June quarter at rates 25.4% above expiring levels, grew revenue 28.9% to $1.9bn, carried a record $1.9bn backlog of signed-but-not-commenced rent at full share — $1.4bn at its own share, the rest accruing to joint-venture partners — and raised 2026 core funds-from-operations guidance to $8.15–$8.20 a share excluding promote income. "Digital Realty delivered record Core FFO per share in the quarter, reflecting robust customer demand and strong execution," chief executive Andy Power said on July 23.

Equinix accelerated too: revenue up 16.4%, operating margin widened to 25.3% from 21.9%, adjusted funds from operations per share up 18%, a record 9,700 net interconnections added and churn of 1.8% against a 2–2.5% target range. "Customer demand is broad-based and growing," chief executive Adaire Fox-Martin said on July 29, announcing raised 2026 guidance of $10.205–$10.285bn of revenue and $42.69–$43.29 of adjusted funds from operations per share.

The buildings that do not exist yet

Both are also enormous builders. Digital Realty's construction pipeline doubled in the first half to 1.4 gigawatts costing $20bn, 63% pre-leased at an expected stabilized yield near 11.5%. Equinix guided 2026 capital spending to $5.0–$6.0bn, told investors its blended cost of capital would rise by about 150 basis points, and set a through-2029 outlook of 9–12% annual growth in adjusted funds from operations per share — a step down from the 11–13% guided for this year.

That is the pipeline the moratorium wave taxes. "There is a lot more pushback in the communities across the U.S.," Berkshire Hathaway chief executive Greg Abel told CNBC on September 2. Mizuho's Vikram Malhotra reads it the other way for incumbents: while new projects may be delayed, it "could be a positive for existing projects/DC REITs which have pricing power".

What the shares have been doing instead

Neither argument moved the price in August. Both stocks topped out on the same session, August 14 — Digital Realty at $200.15, Equinix at $1,102.10 — and have given back about 6% since, the day the 30-year Treasury printed 5.34%, its highest since 2007. Both bottomed in the first days of September and have recovered as yields eased. Very long rent streams get discounted at the long bond; a synchronized fade in two businesses whose operating numbers both accelerated is a discount-rate event.

So the honest split: the operating results earn what they show, and the political overhang is not yet visible in any reported meter. What the moratoriums do is turn a growth story into a scarcity story — good for the 3.0 gigawatts Digital Realty already runs and the 82%-utilized estate Equinix already owns, worse for the compounding both managements have guided investors to expect from capital not yet in the ground. Digital Realty's key risks were always power availability and rate sensitivity; September added the county board.

The test arrives with the third-quarter prints, and one number will settle more than the others: whether a market with 10.8 free megawatts in its biggest metro shows up as rent, or as projects that never get built.

Universal Insurance Holdings, which writes homeowners, condominium and dwelling cover through Florida's independent agents and its own Clovered.com digital agency, reported second-quarter net income available to common stockholders of $59.2m and a net combined ratio of 91.6%. Both numbers were produced by an empty sky: through early September the Atlantic had generated five named storms, no hurricanes and the lowest accumulated cyclone energy on modern record, the least active start in at least 60 years.

A Florida homeowners carrier's year is decided by three variables its income statement never displays: what it pays for reinsurance each June 1, the litigation rules the legislature rewrote in 2022, and the storm count. All three broke the same way this year, and the market has already conceded they will not stay that way — at Universal and at HCI Group, the Tampa insurer that also controls the underwriting-software business Exzeo, the forward earnings multiple sits above the trailing one.

The revenue line is a residual

Universal wrote $621.3m of direct premiums in the June quarter and ceded 30.8% of premium away before it could earn any of it, leaving $377.3m of net premiums earned. That is why the reported top line behaves like an accounting remainder rather than a demand signal: revenue rose 6.2% in the second quarter after falling 0.9% in the first. Sitting ahead of any retained risk are fee streams that get paid on volume written — policy fees collected by Universal's wholly owned managing general agent, brokerage commissions on the reinsurance it places through its in-house intermediary Blue Atlantic Reinsurance Corporation, and installment fees from policyholders.

They bought more cover, not more risk

The June 1 renewal was the cheapest in years. Guy Carpenter put Florida property-catastrophe pricing down 15% to 20% risk-adjusted across many layers, with individual accounts down as much as 25%; the state-run Citizens Property Insurance renewed its $2.82bn program at a net rate on line of 8.46%, against 11.95% a year earlier.

Neither protagonist used the discount to keep more risk. Universal lifted the top of its combined tower by about $50m to $2.623bn while leaving its first-event statutory retention unchanged at $45m, and locked in $352m of capacity extending into the 2027-2028 treaty year, most of it below the Florida Hurricane Catastrophe Fund layer. The renewal is "now fully supported and secured," chief executive Stephen Donaghy said. HCI renewed about $4.1bn of catastrophe excess-of-loss limit, a 16% increase, for $381m of net reinsurance premium, down 10%, with its maximum first-event retention up 4% to $163m.

Cheaper input costs in a regulated line do not stay with the carrier. On HCI's August 6 call, chairman and chief executive Paresh Patel said the saving would be incorporated into a future rate filing, likely late this year, after actuarial review. The giveback is already visible statewide: Florida's insurance regulator reports rates falling in 51 of the state's 67 counties, with an approved 8.7% average decrease for Citizens, and seventeen new carriers have entered since the 2022 and 2023 reforms.

Two businesses, moving apart

Universal's quarter was a margin story rather than a volume one: gross margin of 42.5% against 19.5% a year earlier, direct premiums up 4.1%, book value per share of $22.89, up 39.7%. "In the quarter, we delivered a very strong 38.8% annualized return on common equity, driven by solid underwriting and revenue performance," Donaghy said on July 24.

HCI's is decelerating. Revenue grew 11.1% and net income attributable to shareholders 11.5%, against triple-digit growth in the two quarters prior, and operating margin has fallen from 58.5% in the fourth quarter of 2025 to 45.0%. The pivot away from hurricane risk is real but small: other revenue was $5m in the quarter, roughly 2% of the total, even though Exzeo's November 2025 initial public offering valued that arm near $1.75bn against HCI's own $2.39bn market capitalization. Growth has migrated to new vehicles — TypTap's premiums were flat at $123m while the Tailrow reciprocal exchange's gross premiums earned rose 236% to $29.7m — and the policy pipeline that fed the whole group is draining, with Citizens down to roughly 391,768 policies from a peak near 1.4 million.

What the shares already know

Universal's stock is up 74.1% over twelve months, and the path was not a grind: it fell 9.7% in one session on July 15 after Piper Sandler cut it to Neutral on valuation, then gapped 15.0% higher across the second-quarter release. HCI is up 5.8% over the same year; American Coastal Insurance, the commercial-residential specialist in the same corner, is down 16.1%.

The verdict splits cleanly. Universal's improvement is earned — a six-point better combined ratio, a compounding book value, a cheaper reinsurance tower with unchanged attachment. What the business does not explain is the multiple: at 1.91x book and 5.5x trailing earnings against 9.0x forward, the price outran book value, and consensus already models 2026 earnings per share of $4.85 against $6.32 delivered in 2025. HCI trades richer on every measure — 7.2x trailing, 10.0x forward, 1.97x book — on decelerating earnings and a software business still too small to carry the story. A single-digit earnings multiple here is a storm count, not a valuation.

Both retentions were set on June 1 and neither has been tested by an event. Forecasters put roughly 60% of a typical Atlantic season's activity after September 10, and this one runs to the end of October.

Akamai told investors this summer that the business which gave the company its name is shrinking because customers pay less each time a contract comes up for renewal. Its Delivery and other cloud applications segment brought in $396m in the June quarter, down 6% year on year, and the quarterly filing attributes the decline to "downward pricing of contract renewals," with media and gaming customers optimizing costs in ways that cut traffic across the network — a moderation management expects to continue through the rest of 2026. The decline is deepening: down 2% in the December quarter, then 7%, then 6%.

What replaces it is arriving, and it costs money first. Akamai — a content-delivery network that has added web-application security and an edge compute cloud sold to enterprises and carriers — has signed more than $2.8bn of multi-year cloud-infrastructure commitments this year, including a seven-year, $1.8bn contract with Anthropic disclosed on May 7 and a $600m, four-year robotics compute deal announced with second-quarter results. Capital spending hit $347m in the June quarter, 32% of revenue against 21% a year earlier, and is guided to roughly 40% of revenue for the full year. The buyback is suspended after $616m of repurchases, and the balance sheet now carries about $7.6bn of convertible debt against $4.6bn of cash and securities.

The arithmetic of a front-loaded contract

Management has been explicit about the shape of these deals: the large compute contracts carry 60-70% cash gross margins and 20-30% operating margins after depreciation, with costs incurred before revenue and target profitability reached roughly a quarter after deployment. Run that through the income statement and the result is what the June quarter showed. Revenue rose 5.4% to $1.0997bn while gross profit fell 0.5% to $613.8m, gross margin slipped to 55.8% from 59.1%, and the GAAP operating margin came in at 7.3% — a fourth consecutive quarter of compression from 15.7% last September. Revenue growth, meanwhile, has sat in a 5-7% band for four straight quarters. Security, at $604m and up 10%, is now more than half the company; cloud infrastructure grew 39% to $99m.

"Akamai continues to build momentum and gain wide industry recognition as a key infrastructure provider for the AI-driven economy," chief executive Tom Leighton told investors on August 6. The finance side supplied the cost: all existing graphics-processor capacity is sold out, requiring another $500m of investment, about $60m this year and the rest in early 2027. The revenue guide has barely moved since February — $4.4bn-$4.55bn then, $4.445bn-$4.530bn now — while the non-GAAP operating margin guide has been cut by roughly a point and a half, to 25-26%. Consensus models 2026 revenue up 7.2% and earnings per share down 4.6%, then a 13.0% revenue re-acceleration in 2027 on the contracted book.

The mirror that declined the buildout

Cloudflare, which sells firewall, bot mitigation, content delivery, zero-trust access and a serverless developer platform on the same kind of global edge, monetizes it as expansion inside existing accounts. June-quarter revenue rose 35.9% to $696.1m, a fourth consecutive acceleration; dollar-based net retention reached 120% and remaining performance obligations $2.732bn, up 38%. Non-GAAP gross margin ticked up 30 basis points sequentially, its first sequential gain in eight quarters. Network capital spending ran at 7% of revenue through June against a 14-15% full-year guide, so the step-up is all in the second half. "We're in a very different business than the hyperscalers," chief executive Matthew Prince said on the August 6 call, arguing the company sells completed work rather than rented servers. That posture has its own bill: $99m of severance in the quarter and full-year restructuring charges raised to $165m after May's decision to cut about 1,100 employees.

What the shares have earned

Akamai closed at $109.49 on September 10, 25.9% below its $147.71 close on the day the Anthropic deal landed and 15.8% below mid-June, with its 50-day average crossing below its 200-day in early September. Cloudflare is up 48.9% over six months. The gap tracks gross profit, not orders. HSBC cut Akamai to Hold on August 10, taking its target to $123 from $171 and its valuation multiple to 17x near-term earnings from 25x on cloud margin concerns; JPMorgan upgraded to Neutral in mid-August on inference-led growth. At 16.3x forward earnings against 38.6x trailing, Akamai looks cheap only on earnings that are depressed by design — on enterprise value to trailing EBITDA it is 18.5x, and free cash flow is being consumed. Cloudflare, at 38.3x forward sales and 52.8x forward gross profit, has re-rated well past the 25-28x of enterprise value to 2026 revenue it carried in May.

One session resists explanation. On September 9, Cloudflare rose 10.6% after launching a vulnerability discovery and remediation service built on OpenAI models inside its managed defense product. Akamai rose 9.2% the same day with no contract or earnings of its own — the discoverable event was Leighton at Citi's technology conference, repeating the low-teens 2027 outlook and warning that near-term margins could stay under pressure. The S&P 500 exchange-traded fund fell that day; Fastly and DigitalOcean rose with the complex. The likelier reading is sympathy, not news.

Akamai's 2027 promise now rests on capacity it has pre-sold and must still build, with seven years of economics tied to one private artificial-intelligence company through 2033. Its processors are sold out and the next $500m of them is already spoken for. The growth is contracted; the margin trough is what shareholders own until it arrives.

Alignment Healthcare spent less of every premium dollar on medical care in the June quarter than in any quarter since it listed, raised its full-year outlook, and then told investors it intends to spend the difference. The company, which runs Medicare Advantage plans in California, North Carolina and Nevada and also delivers care to members of unaffiliated health maintenance organizations, reported an adjusted medical benefits ratio of 86.3%, about 40 basis points better than a year earlier and the lowest of its public life. The same measure computed from reported gross margin ran between 86.9% and 88.3% across the four preceding quarters.

That matters because Alignment has no other lever. It has no employer book and no commercial business; close to all of its revenue is a per-member-per-month price set by the Centers for Medicare and Medicaid Services through county benchmarks and risk scores — roughly $1,514 a member a month on about 294,100 members last quarter. The plan cannot raise its price. It bids a benefit package each June and then manages what the members cost. A record-low cost ratio should have been the whole story. Instead the shares closed at $13.03 on September 9, down 32.1% in three months.

Two sessions, both dated

The decline was not a drift. On July 8 the stock fell 16.7%, from $24.05 to $20.03 on 15.4m shares against 4.2m the session before, after a former executive, ex-chief data and transformation officer Hakan Kardes, sued in California alleging the company misclassified roughly $8m–$10m of operating expenses as capital expenditures in 2024. Correcting that would have turned 2024's first positive adjusted EBITDA of $1.3m into a loss of $7m–$9m — and adjusted EBITDA carried a 35% weight in the incentive plan that funded a $1.3m bonus for chief executive John Kao. Alignment says its audit committee retained outside counsel and a national accounting firm and concluded the concerns were unfounded. Kao sold 298,000 shares for about $5.92m on July 10 under a trading plan adopted the previous November.

On July 31 the shares fell another 21.3% — the day after a quarter that beat the high end of guidance on every metric. Revenue was $1,335.6m, up 31.6%, with operating income of $42.1m and net income of $36.6m. Management lifted the low end of full-year adjusted EBITDA to $145m and guided membership to 298,000–301,000. What investors sold was the phasing: with the company reinvesting the outperformance in clinical infrastructure and 2027–2028 market entry, only about 30% of full-year adjusted EBITDA now falls in the second half, against 40% last year. Third-quarter adjusted EBITDA was guided to $20m–$30m.

Nobody else moved on either day. Humana went from $394.62 to $396.29 on July 8. Over the three months in which Alignment fell 32.1%, Humana gained 10.1% and Oscar Health 15.0%.

What the company says it is buying

Alignment's case rests on member tenure. A first-year member runs a medical benefits ratio near 93%, falling to roughly 82.1% by year five, and about half of current members are in their first or second year with the plan — which is why the company puts the embedded gross profit inside its existing book at $880m, against $600m disclosed in early 2025. "We are making these investments now in a year in which we are meeting high end expectations," Kao told investors on the July 30 call.

The pressures are real but named. About half of 2026's new members are chronic-condition, dual-eligible or special-needs enrollees, who cost more early. The Inflation Reduction Act's Part D redesign raised plan liability above the $2,000 out-of-pocket cap from 20% to 60%, the flatter drug-benefit earnings slope management flagged for the second half. The full-weight risk model CMS phased in for 2026 cut payable diagnosis codes from 9,797 to 7,770, compressing industry risk scores. And Alignment kept 100% of members in plans rated four stars or better for a second year, so the quality bonus that funds 2027 revenue is intact.

The other pure-play went the other way

Clover Health, a Tennessee insurer whose book is roughly 98% preferred-provider-organization plans built around its Clover Assistant physician software, reported the same direction of travel: an insurance benefits expense ratio of 87.6%, 80 basis points better year over year, revenue up 55.6% and net income of $28.0m against a loss a year earlier. It guided to its first full year of profit under generally accepted accounting principles. Its shares are up 117% over six months.

Clover also has something Alignment does not: a judge. Judge Lisa Godbey Wood's May 27 opinion found CMS had improperly included 20 star-ratings measures and ordered recalculation, lifting the contract covering more than 97% of Clover's members from 3.5 to 4.5 stars and roughly $120m of 2027 payments with it. CMS appealed on July 21 to the Eleventh Circuit. "Four and a half stars matters. It gives us more room to reinvest in members, maintain a highly competitive product, support growth, and expand profitability. It does not create the economics of our model," chief executive Andrew Toy said on the August 5 call.

The verdict

The operating case against Alignment — utilization running above what the June bid assumed — is not supported by anything the company has reported. The cost ratio improved, the guidance went up. What the de-rating fairly prices is an unresolved accounting allegation and a management team that has now shown it will convert a beat into spending rather than into second-half earnings.

That repricing has been severe on revenue-based measures: Alignment trades at 0.59x trailing and 0.52x forward sales, against roughly 1.09x trailing at the July 7 close, a compression of about 46% in nine weeks. On earnings it is not cheap at all — 72x forward, on a consensus 2026 margin before interest, tax and depreciation of 2.0%. Clover, after doubling, sits at 79x forward earnings.

And on forward gross profit, which is the only measure that compares two insurers carrying very different gross margins, Clover at 4.10x and Alignment at 4.13x are priced almost identically. Buyers are paying the same for the falling chart and the rising one.

Both are now bidding into the same fall. Humana plans to close plans covering roughly 600,000 members in 2027, and UnitedHealth is protecting margin through benefit cuts and market exits in its own 2027 bids. Those members learn by letter this autumn that their plan is going away, and they get seven weeks to choose a new one. Alignment and Clover filed their bids in June; neither can change a benefit now. Whatever each wins in December it carries, at the price the government has already set, for the whole of 2027.

Six Flags' remaining parks are working better than the share price implies, and that is the problem: the improvement is spoken for before it reaches shareholders. Same-park attendance rose 4% in the June quarter, adjusted earnings before interest, tax, depreciation and amortization rose 7%, and the active pass base grew 6% — while the equity fell 37.9% over twelve months. With about $4.9bn of net debt against a $1.46bn market value, shareholders hold roughly a quarter of the enterprise.

United Parks is the mirror. Revenue slipped 1.4% and adjusted EBITDA fell, yet per-share earnings dropped only 11% because 12.4% of the share count was retired. One operator has been de-rated by its balance sheet; the other by a softening Orlando market and the arithmetic holding up its earnings per share.

An operator that got busier and cheaper

Six Flags Entertainment, which runs amusement and water parks across 17 US states, Canada and Mexico under licences including Looney Tunes and DC Comics, closed the sale of six American parks to the landlord EPR Properties in April, part of a seven-property deal worth about $331m, and directed the cash to debt. The parks it kept then had a better summer than the year before. Same-park attendance rose 4% to 13.1 million visits in the June quarter, the company reported on August 6, and adjusted earnings before interest, tax, depreciation and amortization rose 7% to $243m.

The shares fell 37.9% over the past twelve months regardless, to $14.29. The reason is not at the turnstile. Six Flags ended the quarter with roughly $4.9bn of net debt against a market value near $1.46bn — shareholders own about 23% of the enterprise — and in January it owes an estimated $332.6m to buy the remaining 68.5% limited-partnership stake in Six Flags Over Georgia and two nearby water parks, a price indexed to consumer prices since 1997 and not yet finally fixed.

The meter these companies publish

Neither operator is paid at the gate for most of what it sells. Season passes and memberships are sold at a discount months ahead, booked as deferred revenue, and recognized across the operating season — so the deferred balance, not reported sales, is the forward reading. Six Flags carried $431m of it at June 28, lower on a reported basis after the disposals but up $8m, or 2%, on a same-park basis, which the company attributed to membership and advance sales.

The cost of that pass base shows up in per-capita spending, which slipped 1% to $62.88 same-park. Six Flags said the decline "primarily reflected lower admissions per capita spending associated with expanded season pass benefits and increased cross-park visitation, partially offset by continued strength in guest spending on food, extra-charge attractions and other in-park offerings." Sell the pass cheaper, recover it inside the gate: season-to-date pass sales rose 7% and the pass base 6%.

Where the operating result goes

Below adjusted EBITDA the reported line moved the other way — a $203m net loss against $100m a year earlier, carrying a $38.6m impairment on the Six Flags and Schlitterbahn trade names and a $37.8m loss on the disposal group across the first half. After a $1.6bn loss in fiscal 2025 and leverage above 7x, management now targets about 4x net leverage while spending $400–425m of capital expenditure. Consensus has revenue falling 4.7% this year to $2.91bn with EBITDA of $414m, recovering to $669m in 2027 — enterprise value near 15x the current-year figure and 9.5x next year's, with earnings per share still negative in both. There has been no dividend since July 2024.

United Parks runs the same machine, differently

United Parks & Resorts, the Orlando operator of twelve SeaWorld, Busch Gardens, Aquatica and Sesame Place properties, sells the same passes to a smaller, richer park set: about $32 of adjusted EBITDA per guest last quarter against Six Flags' $18.6. June-quarter revenue fell 1.4% to $483.3m and attendance 2.9% to 6.1 million, with adjusted EBITDA down $10.8m to $195.5m. In-park spending per guest hit a record $39.51, up 5.1%; deferred revenue rose 2% to $211.9m; the paid pass base was down 1%.

The per-share numbers are arithmetic. Net income fell 21% to $63.3m while diluted earnings per share fell only 11%, to $1.29, because 12.4% of the shares were retired — $217.7m of buybacks in the first half, leaving $19m of cash and $658m of liquidity entering peak season, and shareholders' equity negative. Management blames the market and itself: Orlando softened broadly, with Comcast reporting attendance across the market weakening from June and international visitation falling for every operator. "We've had less than stellar execution in our marketing activities this year," chief executive Marc Swanson told investors on the August 4 call. He also argued that "multiple highly credible third parties assign significant value to our real estate that we do not believe is currently reflected in the public market price of our common equity." The shares trade at 12.98x forward earnings against 15.4x trailing, 8.55x trailing enterprise value to EBITDA, on a 14.5% trailing free-cash-flow yield.

What the businesses earn and what they do not

Both stocks rallied into late July and have given the whole move back — Six Flags down 18.8% from its July 27 close, United Parks 13.8%. Neither drop on September 9, when Six Flags fell 9.9% and United Parks 5.6%, maps to any company announcement; the likelier reading is that heavily indebted, drive-to leisure names amplified a day when Brent rose 3.4% to $101.21 and the ten-year Treasury yield reached 4.857%.

Six Flags' de-rating is earned, but by the balance sheet rather than the parks: attendance, passes and deferred revenue all improved same-park, and every dollar of that recovery is committed to lenders and to a January cash call. United Parks is the harder case, because its operating slippage is modest and largely explained by a weak Orlando, while the durability of its earnings per share depends on retiring stock with liquidity that a bad weather season would need.

The next honest reading for each is a date, not a quarter. For Six Flags it is the Georgia payment in January. For United Parks it is Black Friday, when the relaunched 2027 pass goes on sale and the deferred balance either builds or does not.