41 — What Is Cooling, Declining and Reversing
Cross-industry synthesis · 25 sectors · research date 2026-09-15
0. Why this document exists, and why almost nobody else produces one
Cooling is the most systematically under-covered category in trend research, and the reason is structural rather than intellectual. Nobody is paid to report a decline. A vendor does not commission research into the contraction of its own category. A consultancy sells transformation programmes, not the finding that a transformation has stalled. An analyst covering a sector has a professional relationship with the companies in it. A conference sells tickets to a growth story. The result is an information market in which the supply of "what is rising" is effectively infinite and the supply of "what is falling" is close to zero — and in which the reader's exposure to being wrong is asymmetric, because a missed acceleration costs an opportunity while a missed decline costs capital already committed.
We therefore treat cooling as a differentiator and a deliberate output, not as a residual. This document covers 113 records, which is 22.6% of the database — a share that is itself a claim, because no commercial trend product we are aware of allocates anything like a fifth of its coverage to deceleration.
Acting on that commitment means three specific things, and each is enforced below:
- Every cooling record carries a turning indicator — the specific, dated, checkable thing that changed. Not a narrative, not a vibe: a number, a filing, a court order or an absence.
- Decline is disaggregated by cause. Structural decline, cyclical trough, policy reversal and measurement artefact look identical in a chart and require opposite responses. Conflating them is the characteristic error.
- Reversal candidates are stated with their mechanism, because a cooling trend that can turn back is a different asset than one that cannot, and the difference is identifiable in advance.
Method and counts
| Query | Records |
|---|---|
classification = cooling |
75 |
stage in (declining, reversing) |
83 |
tags.direction in (decelerating, reversing) |
113 |
| Union of all three | 113 |
Of the union: classified current |
20 |
Of the union: classified overhyped |
12 |
Of the union: classified emerging_signal |
6 |
The direction field is the widest net and captures the twenty current trends that are
large, real and decelerating — T-09-02, T-14-03, T-12-02, T-21-01, T-01-01 and others.
These belong here even though they are not classified cooling, because a decision-maker
reading only the cooling bucket would miss the biggest decelerations in the database.
Evidence is weaker on average in this set than elsewhere, and that is not an accident:
decline is under-sourced by the same mechanism that makes it under-covered. Of the 75
cooling-classified records, 22 are single-source and 8 carry low confidence. Each
table below states the evidence factor.
1. Structural decline — the cost curve, the rule or the market structure changed, and it does not revert
The test, from the definitions framework: what would have to happen for this to reverse, and is that plausible? For these records the answer requires undoing something that cannot readily be undone.
| ID | Score | E | Trend | Turning indicator — the specific thing that changed |
|---|---|---|---|---|
| T-14-18 | 74.3 | 0.9 | The cable bundle as a profit pool is reversing, not merely declining | Paramount affiliate revenue −6% year on year in Q2 2026 — outright decline, not slower growth. Cable fell to 18.7% of US TV viewing, below broadcast at 19.5%. Comcast has deconsolidated Versant and announced an NBCUniversal/Sky separation. The decisive datum is cohort, not share: UK 16–24 weekly broadcaster reach fell from 43% (2022) to 26% (2025) — a cohort that never formed the habit, which removes recovery-by-ageing |
| T-14-03 | 78.8 | 0.9 | Affiliate fees falling outright, cost cuts still outrunning | Paramount TV Media revenue −9% while adjusted EBITDA rose to $1.1bn and margin expanded 26.4% → 34%. The threshold question is when revenue falls faster than cost can be cut |
| T-02-17 | 72.8 | 0.7 | The offshore headcount-linked services pyramid | Operators now guide to it rather than deny it: HCL's CEO quantified "AI deflation" at 3–5% of revenue; TCS's CEO used the word "degrowth." TCS revenue −0.5%; Infosys narrowed FY27 guidance to 1.5–3.0% cc. Headcount has not fallen, which is precisely why margins compress |
| T-22-17 | 54.2 | 0.7 | The consumer homework-help subscription | Chegg revenue −51% year on year to $51.8m, guiding down sequentially to $43–44m. The cleanest fully-quantified AI substitution event in the database: when a general-purpose model does the job, the vertical subscription does not decay — it halves in a year |
| T-22-16 | 48.5 | 0.4 | Online degree programmes as a growth engine | Coursera Degrees revenue $13.4m, down from $15.7m, while the same company's Consumer grew 29% and Enterprise 118%. Fourteen years after the MOOC thesis, the degree line shrinks while everything around it grows |
| T-17-16 | 47.3 | 0.5 | The undifferentiated creator fund | Arithmetic, then absence: a fixed numerator over an unbounded denominator guarantees falling per-creator payouts, and no new major creator fund launch was identified in the research window. Platform disclosure language shifted from fixed pools to contractual revenue share |
| T-19-16 | 47.2 | 0.6 | Plant-based meat, fifth year of contraction | Beyond Meat Q2 2026 $68.8m vs $74.9m; FY2025 $275.5m vs $326.5m (−15.6%), five consecutive annual declines. Compounded by the 2025–2030 Dietary Guidelines endorsing full-fat dairy and a 1.2–1.6 g/kg protein target while instructing consumers to avoid highly processed foods |
| T-06-16 | 53.9 | 0.7 | Ex-vivo autologous cell and gene therapy as a commercial model | Patient counts against capital deployed: 64 Casgevy infusions in all of 2025 for $116m; Lyfgenia just over 100 patients ever; bluebird bio sold to private equity for under $50m after once being a $10bn company with multiple approvals |
| T-16-16 | 62.2 | 0.7 | Multi-touch attribution and the deterministic user graph | Ownership, not adoption: Publicis is acquiring LiveRamp for $2.167bn. Once the last neutral identity spine sits inside a buyer, cross-platform deterministic attribution ceases to exist as a neutral product category |
| T-08-16 | 61.5 | 0.6 | GEO satellite expansion | Cancellation, not deferral: SES cancelled IS-41 and IS-44 outright for failing its internal rate-of-return threshold, joining Eutelsat. Within SES, Networks grew >100% while Media/Video fell 11%. Starlink's $66 ARPU is the price new GEO broadband capacity must clear, and it cannot |
| T-24-17 | 59.3 | 0.9 | Ad-supported music streaming has stopped contributing growth | Spotify ad-supported revenue +1% in Q2 2026 while ad-supported MAU grew 14%; H1 revenue fell 2%. Revenue per free user is declining outright. The free tier is now a subscriber-acquisition cost, not a revenue line |
| T-24-16 | 55.0 | 0.9 | Webtoons as a growth story | Three issuers, two currencies, one direction: WEBTOON revenue −1.5%, Kakao Story −15.8%, NAVER Contents +0.5% against group growth of 16.2% |
| T-24-18 | 39.5 | 0.4 | Comics IP adaptation revenue | WEBTOON IP adaptations −22.8% to $19.8m — the steepest decline in any of its lines, and the exact mechanism the "webtoons are an IP factory" thesis rested on |
| T-15-16 | 52.9 | 0.8 | Consumer VR content | Studio closures, not hardware announcements: Polyarc shut, Vertigo Games Amsterdam closed, Schell Games cut, ~100 Apple Vision Pro staff removed, Lynx acquired out of liquidation. Meta Reality Labs lost $8,647m on $833m of H1 revenue — roughly 10:1. Valve's Steam Frame at $1,059 moves PC VR further from mass market |
| T-17-18 | 42.7 | 0.5 | Metaverse and immersive social as a consumer platform thesis | Venture funding into metaverse-labelled startups has effectively stopped since 2023; residual activity is corporate strategy decks, not consumer coverage |
| T-04-17 | 56.9 | 0.6 | US federal cyber capacity | CISA headcount ~3,200 → ~2,200; field cybersecurity advisers 164 → 97; the Cybersecurity Information Sharing Act of 2015 lapsed in the 43-day late-2025 shutdown and has run on stopgaps since; CIRCIA slipped twice |
| T-11-17 | 56.5 | 0.7 | First-time fund formation at a decade low | On pace for the lowest year since 2016; median time between fund closes 1.7 years. The mechanism is arithmetic: −$202bn of net LP cash flow since 2022 and ~$5tn of NAV in funds seven years or older leaves nothing to recycle to unproven managers |
What unites this group. In every case the mechanism of decline is identifiable and its reversal would require undoing a cost curve (T-08-16, T-22-17), a demographic non-event (T-14-18), an ownership change (T-16-16) or an exhausted balance sheet (T-11-17). None would reverse on sentiment.
2. Cyclical trough — the level is falling, the mechanism reverts
These look identical to §1 in a chart and are the opposite thing. The discipline is to ask whether the phenomenon has happened before.
| ID | Score | E | Trend | Turning indicator | Why it is cyclical |
|---|---|---|---|---|---|
| T-09-17 | 69.4 | 0.7 | The semiconductor and electronics factory build-out has crested | Computer/electronic/electrical factory construction down ~44% from its July 2024 peak, to ~$70bn SAAR; this single component causes most of the 21.7% headline decline in T-09-02 | It is a subsidy disbursement profile, not a demand failure. CHIPS awards made 2022–24 produced put-in-place spending 2023–25 and are now tapering into equipment installation. Excluding electronics, factory construction grew ~5.6% nominally |
| T-13-18 | 65.2 | 0.7 | European industrial robot demand contracts in every major market | Europe −8% to ~85,000 units: Germany −5%, Italy −16%, France −24%, UK −35%. The 2023 high of 92,393 units was a peak, not a trend | Western Europe is at a record 267 robots per 10,000 employees. High density plus falling installations is replacement demand, which is a lower-growth service business, not disappearance |
| T-13-17 | 64.3 | 0.5 | Automotive stops being the engine of robot demand | US automotive robot installations 13,500 units, −1%, while total US installations rose 11% — all growth from food (+~30%), metals and electronics | Partly cyclical (EV retooling wave complete), partly policy (EPA GHG rescission 2026-04-20, EV credit expiry). Composition change, not contraction |
| T-09-16 | 68.9 | 0.8 | Container overcapacity: structurally intact, cyclically overwhelmed | Not the supply data — orderbook-to-fleet 31.6%, idle tonnage 0.7%, scrapping near zero, all unchanged — but the rate outcome: Drewry at $4,476/40ft on 2026-09-10 against $2,107 in January | The bear case was deferred by an event, not refuted by modelling. Hormuz closure absorbed 1.5–2m TEU of effective capacity. Every ordered ship is still on order |
| T-09-18 | 62.1 | 0.7 | Air cargo growth collapses to near zero | IATA cut 2026 growth from 2.6% (March) to 0.2% (June) after the Hormuz closure disrupted jet fuel and Gulf hubs. February 2026 CTKs had been +11.2% | Event-driven. Revenue is still forecast +7.2% on yields +6.5% — volume and value decoupled, which is a price response to a supply shock |
| T-19-17 | 54.2 | 0.5 | The egg price shock is unwinding | USDA ERS forecasts US egg prices −30.8% in 2026 (interval −35.3% to −25.3%), pulling the whole meats/poultry/fish/eggs CPI group to +1.1% | HPAI flock cycle. This research date sits immediately before the autumn migration window — the single most reversible item in the food sector |
| T-19-18 | 64.4 | 1.0 | The coffee and cocoa cost super-spike is deflating | Cocoa $5.95/kg (Aug 2026) against a $7.80 2025 average (−24%); robusta −18%; EU coffee/cocoa import value −17%; Nestlé already guiding to an H2 margin benefit | Agricultural supply cycle. West African harvest failure could re-spike it within one season |
| T-23-16 | 57.1 | 0.3 | Premium athleisure reverses in its home market | lululemon Americas comparable sales −12%, FY guidance implying −5 to −7% revenue. Reported gross margin rose 200bp only because $134.5m of tariff refunds contributed 560bp — underlying margin fell ~360bp | Single issuer, E=0.3. Brand cycle plus a leadership transition; adidas and challengers are taking the share rather than the category disappearing |
| T-25-17 | 45.3 | 0.4 | Japan inbound cools after the weak-yen surge | June 2026 arrivals ~3.15m, −6.8% y/y, after consecutive record years | A currency trade unwinding. Renewed yen weakness restarts it |
| T-18-06 | 68.8 | 0.9 | New-home spec inventory overhang | Months of supply 9.6 against a 607,000 SAAR pace; 117,000 completed and unsold; 35% of builders cutting prices by an average 6%; single-family starts −15.7% | Housing cycle, though the land-basis problem runs through 2027 pipelines. Note the genuine internal inconsistency: permits rose 5.0% m/m and 3.1% y/y in the same month |
The most instructive record in this section is T-09-16, because it is the case where a rigorously constructed consensus was comprehensively wrong for a reason that had nothing to do with the analysis. The correct reading — the bear case is deferred, not falsified — is available only if the analyst separates the structural supply series from the cyclical price series, which the sell-side did not.
3. Policy-caused decline and reversal — a rule changed, and it can change back
This is the largest and fastest-moving group, and the one where persistence scores are lowest by construction. Every record here is reversible by the same instrument that caused it.
| ID | Score | E | Trend | Turning indicator | Reversibility |
|---|---|---|---|---|---|
| T-05-16 | 77.9 | 0.7 | Federal clean-energy tax credit support withdrawing | Not installations, which are at records — the queue: solar and wind interconnection requests each −19% in the 2026 data. Treasury eliminated the 5%-of-cost safe harbour in Aug 2025; OBBBA set a 2026-07-04 begin-construction deadline; FEOC rules from Jan 2026 | Developers safe-harboured 216–240 GW of solar and ~16 GW of wind before the deadline — enough to cover forecast installations to decade-end. The cliff is at the four-year placed-in-service limit, ~2029–2030, not in 2026. A federal court has already struck down one IRS attempt to narrow eligibility |
| T-10-16 / T-10-03 | 73.0 / 74.5 | 0.7 | US battery-electric adoption in its first year of decline | EIA/Omdia: BEV share 6% of Q2 2026 light-duty sales against 7% a year earlier, after a pull-forward spike to 12% in September 2025 immediately before the credit expired 2025-09-30. Luxury BEV share 22% → 14%. Hybrids at a record 16% | Isolates the variable cleanly: US EV adoption was incentive-elastic, not preference-driven. Reverses on price, not policy — the indicator is a mainstream sub-$30,000 BEV without incentive dependence |
| T-10-18 | 72.5 | 0.6 | Unwinding of BEV-only dedicated platforms at legacy OEMs | Stellantis €25.4bn of one-time charges against a €22.3bn FY2025 net loss, dropping PHEV technology in parts of the lineup, with the CEO citing over-estimation of the energy transition's pace. VW's four ID plants under review to 2031; Audi's US plant on hold | Regulatory whiplash is the reversal mechanism in both directions. EPA rescinded US vehicle GHG standards effective 2026-04-20; the EU's proposed 90% 2035 target preserves PHEV/EREV/ICE and has not been adopted |
| T-07-17 | 69.0 | 0.7 | US open banking as a federal mandate is reversing | A Kentucky federal court enjoined the CFPB's Section 1033 rule as likely exceeding statutory authority; the Bureau is rewriting it to permit data-provider fees after a set number of free requests, and it sits on the 2026 rescission list | New York Assembly Bill 10640 / Senate Bill 9483 show the state substitution path. The UK went the other way — FCA open-finance roadmap, April 2026. Fifty-state fragmentation is worse for aggregators than the federal rule was |
| T-05-17 | 67.2 | 0.9 | Coal retirement schedule stalling and partially reversing | Operators planned 12.3 GW of 2025 retirements and completed 4.6 GW — the least since 2008. Coal-specific retirements 2.6 GW, least since 2010, against 13.7 GW in 2022. 4.8 GW postponed, 1.1 GW cancelled outright | The reversal is itself now reversing: the DC Circuit vacated DOE's Section 202(c) order on 2026-09-11. The retirement schedule has become a political variable and is no longer a reliable model input in either direction |
| T-20-16 | 63.1 | 0.6 | The assurance escalation has been cancelled | Directive (EU) 2026/470 deleted Article 26a(3) and its October 2028 reasonable-assurance deadline, and deleted the Article 29b(1) sector-standards empowerment. Limited assurance is now permanent | Assurance requirements essentially never go backwards once legislated. This one did. The Big Four staffed for a revenue event that is not coming |
| T-20-18 | 63.8 | 0.7 | Target-setting rigour softens at SBTi | Corporate Net-Zero Standard v2.0 (2026-06-11) offers "a suite of options to reflect business reality"; SBTi Services launched a paid Readiness Assessment Service five weeks later. Participation is not cooling — 11,948 validated targets. Rigour is cooling | Version drift makes "SBTi-validated" meaningless without version metadata. The private-sector echo of the EU's public retreat |
| T-20-17 | 50.2 | 0.5 | Carbon-market equity funding has collapsed | Sightline Climate records carbon equity funding collapsing in H1 2026 while total climate-tech funding rose 55% to $26.1bn. Corroborated by CCUS at 0.23% of clean-energy investment and durable CDR delivery at 3.4% of contracted tonnes | Reverses via an obliged buyer: an EU CRCF delegated act creating a compliance-grade removal unit, or CORSIA unit scarcity. Honest caveat: "collapsed" is the tracker's word; no numeric figure was retrievable |
| T-01-16 | 51.5 | 0.6 | The EU AI Act high-risk compliance wave stalls | The Digital Omnibus entered into force 2026-07-27; high-risk employment obligations deferred 2026-08-02 → 2027-12-02. Transparency obligations did take effect on schedule, and the two are routinely conflated | The deferral could itself be reversed. Meanwhile every vendor claim built on the August 2026 date is stale |
| T-25-16 | 49.3 | 0.5 | Venice access fee terminated | The city's own site records that from 2026-07-27 the access fee is no longer in effect; the 2024–26 scheme ran as a trial and was not made permanent | The template dozens of destinations cited has lapsed. Mechanism of failure: administrative cost and political friction exceed revenue, so schemes lapse at renewal rather than being repealed |
| T-16-17 | 44.2 | 0.6 | Structural break-up as the expected antitrust outcome | 2026-09-02: Judge Brinkema declined to order divestiture of AdX or DFP and refused to require open-sourcing of the auction logic | A live DOJ appeal is the reversal mechanism, and the sealed opinion could contain more than reported. This record could itself reverse within twelve months and says so |
| T-03-17 / T-03-05 | 49.2 / 59.3 | 0.7 / 0.8 | China as a merchant-GPU revenue line is written to zero | NVIDIA's guidance language: it assumes no China data-centre compute revenue, having written down $0.4bn of H200 inventory after Beijing restricted purchases despite US licences being granted from February 2026 | The least persistent trend in the sector. US policy loosened (case-by-case review from 2026-01-15) and the trade still did not happen — the binding constraint moved from Washington to Beijing, and no US policy change can relieve it |
| T-04-16 | 44.3 | 0.5 | SEC material-incident disclosure deflating into voluntary filings | 29 mandatory Item 1.05 filings against 50 voluntary Item 8.01 filings over two years; petition SEC File 4-856 seeks to rescind Item 1.05 | A worked example of a regulation whose behavioural effect was to move disclosure into a safer channel rather than to increase it |
| T-05-20 | 52.2 | 0.9 | Federal emergency orders as a reliability instrument | Four of eleven ordered units were not operating as of mid-2026; net compliance cost $259m for J.H. Campbell alone; DC Circuit vacated the order on 2026-09-11 | Reserve-margin models that assume a federal backstop are now wrong. Whether DOE appeals or reissues narrower orders is the open question |
4. Measurement artefact — the number moved, the world may not have
The most dangerous category, because it produces confident wrong conclusions in both directions.
| ID | Score | E | What appears to be happening | What is actually being measured |
|---|---|---|---|---|
| T-15-17 | 76.2 | 0.8 | Roblox revenue +36% | Bookings +8% and guided to a 14–18% decline — the first in company history. Recognised revenue is deferred-revenue release from prior bookings. DAU 123m against a 152m peak; hours 29bn against 40bn. The 36% is a lagging accounting artefact. Part of the engagement decline is self-inflicted by age-assurance measures at 57% penetration, which is a policy cost, not a demand collapse |
| T-14-16 | 44.2 | 0.7 | The streaming subscriber series has flattened | It has been withdrawn. Disney's Q3 FY2026 release did not headline Disney+/Hulu counts; Netflix stopped reporting quarterly membership. Third-party estimators (Antenna, Ampere) are filling the gap with modelled figures of unknown accuracy. A broken series is not a trend |
| T-02-16 | 64.8 | 0.5 | SaaS net revenue retention is deteriorating | NRR has disappeared from disclosure. Among the vendors reviewed only Snowflake — consumption-priced — published one (126%). The disappearance is the indicator; the underlying direction is inferred from Indeed's software postings index at 76.12 against a 100 baseline, and from cRPO, which shows deceleration rather than collapse |
| T-21-17 | 55.2 | 0.7 | DeFi is contracting | TVL is a price multiplied by a quantity, reported as a quantity. With ether −37.2% over twelve months, much of the $88.166bn series is price. The audited cross-check is the one that counts: Coinbase blockchain rewards revenue −42% to $83.3m |
| T-18-18 / T-09-02 | 73.5 / 80.8 | 0.9 | Nonresidential construction is booming (+21.3% "office") | US Census reports data centres inside the Office category. Decomposed: data centres drove all nonresidential growth in July 2026 while non-data-centre spending fell to its lowest since September 2023, and private manufacturing construction fell 21.7% |
| T-17-08 | 66.8 | 0.9 | Snap is growing (DAU 493m, revenue +19%) | North America DAU −7% to 92m and Europe −2%; all growth is in the geography where a user is worth ~2.8% of a US/Canada user. Blended user metrics are now actively misleading for the challenger platforms |
| T-01-18 | 59.2 | 0.5 | Pre-training scale is failing | Low confidence, single source. Independent indices show convergence (GPT-6 Astra and Claude Fable 5.1 both at 53 on Artificial Analysis v4.3), and effort-tiered scoring now treats inference budget as a capability variable. But labs do not disclose training compute, so this is inferred, not observed. Recorded honestly as low-confidence |
5. The complete cooling register — remaining records with turning indicators
Records not covered above, for completeness. All are cooling-classified.
| ID | Score | E | Trend | Turning indicator |
|---|---|---|---|---|
| T-06-17 | 76.4 | 0.7 | Medicare Advantage as a growth engine | Enrolment +3% to 35.5m against a historical rate of up to 10%; UnitedHealthcare −9% (10.3m→9.4m), Elevance −14%, Centene −4% — deliberate margin-over-membership contraction, with Humana and Devoted taking the share |
| T-08-18 | 74.3 | 0.6 | Heavy-lift launch diversification reverses | A New Glenn stack exploded during static fire at LC-36 on 2026-05-28; repairs exceed a year; BE-4 main oxygen valve confirmed as cause 2026-08-05. SpaceX flew 78 times in H1 2026. Amazon Leo's 396 satellites launched on Atlas V, Falcon 9 and Ariane 64 — none on New Glenn |
| T-18-18 | 73.5 | 0.9 | Manufacturing/reshoring construction collapsing as data centres take the crews | Data-centre starts $84.1bn YTD, nearly triple, with cost/sq ft +57% to $818.20, while manufacturing construction fell 21.7% and contractor backlog stayed flat at 8.8 months |
| T-10-17 | 70.9 | 0.6 | Korean and Japanese cell makers losing share | SK On installations −9.8% Jan–Jul 2026 while the market grew 20.4%; LGES +4.5% and Panasonic +7.6%, both below market. Seven Chinese firms hold 72.8% of the top ten, +3.1pp |
| T-07-16 | 72.4 | 0.8 | Property-cat reinsurance pricing softens while the loss trend rises | Five consecutive quarters with no single insured loss above $10bn; only 11 billion-dollar events in H1 2026 against a ten-year average of 16. Swiss Re nonetheless maintains 5–7% annual structural growth in insured losses. The price is cooling; the risk is not |
| T-12-16 | 68.8 | 0.8 | China-direct low-price platforms decelerate | PDD Q2 2026 revenue +8% and a consensus miss; net income −12%; online marketing services +3% while sales and marketing expense rose 9% — growth is being bought |
| T-11-18 | 68.5 | 1.0 | Non-AI venture and US startup formation contracting | Every count-based series down while every dollar series is up: 8,440 deals in Q2 2026, lowest since Q3 2017; pre-seed instruments 14,825 → 11,500+ on flat dollars; Census business applications −7.8% m/m in August 2026 |
| T-18-17 | 68.2 | 0.7 | Multifamily supply wave passes while its 2021–22 vintage debt fails | Multifamily CMBS delinquency +46bp to 7.69% in July 2026 — the largest monthly rise of any property type — while advertised rents rose for a sixth consecutive month. Recovering rent rolls do not fix a loan struck at a 2021 cap rate |
| T-06-18 | 66.7 | 0.8 | Systemic AAV gene therapy for neuromuscular disease | FDA imposed a Boxed Warning on Elevidys 2025-11-14 and excluded non-ambulatory patients after two fatal acute liver failures. Sales $898.7m (2025) with 2026 forecast below $500m; the CEO announced retirement |
| T-14-17 | 66.0 | 0.8 | The content-spend arms race | Explicit efficiency targets replace volume: Paramount's run-rate goal raised to $2.7bn, plus $3bn+ from the Skydance combination and $6bn+ stated in the WBD transaction |
| T-25-18 | 65.5 | 0.8 | US domestic air travel declines in absolute passenger terms | Seasonally adjusted June 2026 passengers 80.1m, 3.7% below the June 2024 peak of 83.2m; three consecutive months of US domestic RPK decline; 2,783 airline jobs lost in July |
| T-23-17 | 64.4 | 0.6 | Mass-fashion store estates shrink | H&M at 4,038 stores against 4,166, ~170 net closures planned, SEK 679m of restructuring — while gross margin rose to 56.6%, so profit recovery is coming from sourcing, not customers |
| T-11-16 | 64.0 | 0.4 | The semi-liquid liquidity promise breaks | Blue Owl Capital Corp. II closed quarterly redemptions on 2026-02-18; the $33bn Cliffwater fund received requests for 14% against a 7% cap; non-listed BDC redemptions hit 4.71% of NAV, nearly tripling q/q |
| T-05-18 | 62.9 | 1.0 | Unbounded data-centre load forecasts deflate | PJM cut summer 2027 by ~4 GW and 2028 by 4.4 GW after stricter vetting; AEP Ohio halved its forecast; Exelon disclosed 22% expected materialisation; ERCOT suspended Batch Zero against 474 GW of requests. Honest caveat: PJM simultaneously raised its long-term growth rate 3.1% → 3.6%. Deflation at the margin, not reversal |
| T-21-18 | 62.0 | 0.7 | Standalone bitcoin mining abandoned by miners | Cipher Mining renamed itself Cipher Digital and raised $810m of secured notes against data-centre assets; TeraWulf signed a 20-year, 410 MW AI lease. A company removing its core activity from its own name is the indicator |
| T-21-16 | 55.9 | 0.8 | Digital asset treasury companies break | Strategy's BTC Monetization Program sold $218.4m of bitcoin YTD, primarily to fund preferred dividends, while repurchasing STRC at a 13% discount to par and paying a 12.00% coupon. The flywheel running backwards, disclosed by the company itself |
| T-12-18 | 61.0 | 0.7 | The retail apocalypse narrative cools | Six major US retail Chapter 11 filings through mid-July 2026 against nine in all of 2025, and the filings that occurred were leverage stories (Saks $3.4bn, Sleep Number $1.3bn), not format failures |
| T-04-18 | 60.9 | 0.7 | Cyber insurance loses its grip as a de facto regulator | Twelve consecutive quarters of cyber rate decline, with carriers now competing on broader coverage and lower deductibles — the exact inverse of the 2021–22 hard market when insurers effectively mandated MFA and EDR |
| T-16-18 | 60.0 | 0.7 | Third-party cookie deprecation as a planning deadline | The deadline moved 2022 → 2023 → 2024 → 2025 and was abandoned as a forced migration. The sector's best-documented false positive: six years of roadmaps built on a date that never bound |
| T-07-18 | 59.9 | 0.4 | The sponsor-bank BaaS model | Over 25% of FDIC and over 20% of OCC formal enforcement actions now target sponsor banks in embedded-finance partnerships; 80% of sponsor banks report multi-partner compliance strain. Demand is not cooling; cheap charter access is |
| T-15-18 | 59.1 | 0.6 | Day-one-on-subscription abandoned by its inventor | Microsoft removed day-one Call of Duty on 2026-04-21 and cut the price it had raised ~30% six months earlier; Xbox content and services revenue −10% |
| T-03-18 | 51.9 | 0.7 | The AI-PC upgrade supercycle stalls on memory economics | Samsung's own Q2 2026 commentary acknowledging "partial demand moderation" in mobile and PC — from the company selling the memory, in the same document as record memory profits. DRAM reported at ~60% of a budget handset BOM |
| T-02-18 | 50.3 | 0.5 | Standalone per-user AI copilot SKUs | Adobe net new ARR −36–37% y/y with management explicitly declining pricing actions in favour of freemium reach; AI migrating onto credit pools at Salesforce and HubSpot. Counter-indicator keeping this at "cooling": Microsoft's 30m+ paid Copilot seats |
| T-17-17 | 49.9 | 0.5 | Standalone AI companion apps as a venture category | Fixed compliance costs landing on a ~$120m annual app-store revenue base: FTC 6(b) orders to seven companies, statutes in four-plus states with a private right of action, and the category leader voluntarily removing its under-18 open-ended product |
| T-03-16 | 46.9 | 0.7 | Near-term High NA EUV adoption | TSMC stated on 2026-09-08 it will deploy High NA in HVM only from 2030, with a 12-inch mask pilot in 2031. ASML's 2027 expansion is low-NA EUV and DUV immersion, not High NA |
| T-22-18 | 47.4 | 0.5 | Coding bootcamps and the learn-to-code pipeline | Capital rotating away from coding academies as the junior-developer rung is compressed from both directions — hiring contraction and AI assistants absorbing junior tasks |
| T-13-16 | 48.0 | 0.9 | Humanoid valuations deflate | Unitree fell from an RMB 1,100 first-day high to RMB 513.93 on 2026-09-09 — 53% off the high, 39% off the first-day close, erasing ~$35bn — while nothing changed operationally |
| T-08-17 | 44.4 | 0.4 | eVTOL certification slips again | EHang — the only holder of a type certificate — withdrew its 2026 revenue guidance. When the certified player cannot forecast revenue, the uncertified players' timelines carry no information |
| T-18-16 | 44.2 | 0.4 | Proptech venture funding | $4.53bn across 231 rounds in H1 2026 against $16.7bn tracked for full-year 2025 — an annualised pace ~46% below, during the largest half-year of global venture funding on record. Decoupling, not correction |
| T-01-17 | 46.0 | 0.4 | Open-weight models lose ground in Western enterprise workloads | Open-source share of enterprise workloads 19% → 13%, with 87% of workloads on closed models. Low confidence: this directly contradicts the developer-token-volume picture, and the divergence is recorded as unresolved |
6. Reversal candidates — what could turn back, and by what mechanism
A cooling trend with a live reversal mechanism should be held differently from one without. Ordered by how identifiable the trigger is.
| Cooling trend | Reversal mechanism | The indicator to watch |
|---|---|---|
| T-23-18 — Swiss watch downturn (already reversing) | Operating leverage off a 3.4% divisional operating margin. Richemont Specialist Watchmakers returned to +8% at constant rates in the June 2026 quarter after a 24-month contraction | Whether Chinese demand — still declining while the Americas and Japan lead — joins. A recovery led by a different demand base than the last cycle |
| T-19-17 — Egg deflation | A single autumn or spring HPAI wave. This research date falls immediately before the autumn migration window | APHIS commercial-flock detections. Note the data gap: the APHIS dashboard could not be retrieved |
| T-09-16 — Container overcapacity | Suez normalisation releases the 1.5–2m TEU absorbed by Cape routing into an orderbook delivering 2.8m TEU in 2027 and 3.5m in 2028. Maersk and Hapag-Lloyd are returning six Gemini services from late September 2026 | Weekly Suez transit counts (IMF PortWatch); blank-sailing discipline |
| T-25-16 — Visitor access charges | Reinstatement by a future Venice administration, or adoption elsewhere. The underlying pressure has not abated: EU platform guest nights +11.4% a year | Whether any city adopts a day-visitor charge citing Venice after its lapse |
| T-03-17 — China GPU revenue | A political decision in either capital: a US–China settlement reopening licences, or Beijing concluding domestic supply is inadequate. The revenue line is written to zero, not structurally destroyed | NVIDIA restoring a China line to guidance |
| T-16-17 — Antitrust structural remedies | A successful DOJ appeal reinstating divestiture; EU DMA enforcement as a separate structural lever | A notice of appeal within the statutory window after final judgment |
| T-10-16 — US BEV decline | Price, not policy. Chinese-cost LFP economics reaching US assembly via Korean or Japanese licensees, or via Mexico | A mainstream US-market BEV launching below $30,000 MSRP without incentive dependence |
| T-05-17 — Coal retirements | Already double-reversing. The DC Circuit vacatur removed the federal instrument; whether DOE appeals or reissues narrower orders decides the direction | DOE's response to the 2026-09-11 ruling; RTO retirement filings |
| T-13-16 — Humanoid valuation deflation | One credible, audited, multi-hundred-unit paying industrial deployment. The category is not disproven; it is undisclosed | Harmonic Drive consolidated bookings and Schaeffler's 2027 strain-wave line — component orders lead demo videos |
| T-15-18 — Day-one subscription | A post-GTA-VI competitive environment in which Microsoft reinstates flagship day-one access to defend share | Whether the next Call of Duty appears on Game Pass at launch |
| T-02-16 / T-02-03 — SaaS de-rating | Two or three consecutive quarters of incumbent cRPO reacceleration alongside visible AI ARR. Salesforce's Q2 FY27 cRPO already accelerated to +14% cc | cRPO growth, since NRR is no longer disclosed |
| T-05-16 — Clean-energy credits | A federal court has already struck down one IRS attempt to narrow eligibility; an administration change reopens it. But the binding cliff is the four-year placed-in-service limit around 2029–30, colliding with five-year interconnection timelines | The share of the 216–240 GW safe-harboured pipeline that energises before its limit |
| T-20-17 — Carbon funding | An obliged buyer appearing: an EU CRCF delegated act creating a compliance-grade removal unit, or CORSIA scarcity bidding prices up | A second large-scale airline CORSIA retirement; any CBAM scope extension |
| T-01-16 — EU AI Act high-risk | Re-tightening after the deferral window, or member-state enforcement of the transparency obligations that did take effect on 2026-08-02 | Whether harmonised standards publish before 2027-12-02 |
| T-24-17 — Ad-supported music | Video and podcast inventory inside audio platforms; programmatic fill-rate improvement; emerging-market free tiers starting from near zero | Two consecutive halves of ad-supported revenue growth exceeding ad-supported MAU growth |
| T-19-16 — Plant-based meat | A livestock disease shock, a carbon price on ruminant protein, or fermentation ingredients reaching cost parity as a B2B input rather than a consumer analogue | A major CPG adopting fermentation-derived protein in a mainstream SKU on cost grounds |
Reversals that are unlikely and should be said to be unlikely
- T-14-18 cable bundle. Reversal requires a cohort that never formed the habit to form it. UK 16–24 broadcaster reach fell 43% → 26% in three years. There is no mechanism.
- T-22-17 Chegg. Reversal requires a general-purpose model to become worse at homework, or paid.
- T-08-16 GEO expansion. Reversal requires new GEO broadband capacity to clear a $66 Starlink ARPU. The arithmetic does not close.
- T-11-17 first-time funds. Reversal requires LP cash, and DPI is 0.05x on the most recent large vintage. This is a four-to-five year lag by construction.
7. Three cross-cutting observations about decline
Decline is better evidenced than growth in exactly the places where it is least convenient. The highest evidence factors in this entire document sit on the records that no vendor wants published: T-11-18 (E=1.0, non-AI venture contraction), T-05-18 (E=1.0, load-forecast deflation), T-19-18 (E=1.0, commodity spike deflating), T-13-16 (E=0.9, humanoid valuation deflation). The data exists; the incentive to publish it does not.
The most reliable turning indicator across all 113 records is a disclosure change, not a data point. Microsoft stopping Game Pass subscriber reporting; Cipher Mining removing "Mining" from its name; Roblox moving to quarterly-only guidance; SaaS vendors dropping NRR; Baidu dropping robotaxi ride counts; Disney dropping subscriber counts. A company that stops reporting a metric is telling you about that metric.
Policy-caused declines outnumber structural ones roughly two to one in 2026, and they have a half-life measured in months. Of the cooling records, seventeen are driven by a rule change made within the last eighteen months, and in three cases (T-05-17, T-05-20, T-16-17) the reversal has already begun to reverse. The correct persistence score for a policy-caused trend is low by construction, and a research product that scores it otherwise is mis-selling durability.
Research provenance
- Source artifact
- 04-analysis/41-what-is-cooling.md
- Corpus date
- 15 September 2026
- Prepared for this site
- 16 September 2026
- Site publication
- 18 September 2026
- Verification
- Inherited; not fully rechecked