17–28x forward earnings
Mandated, price-free intermediaries are currently trading at an AI-fear discount of 17–22x. As the market recognises their cost-base deflation against a defended price, the cleanest franchises should re-rate toward 24–28x.
The Portfolio at a Glance
The following overview summarizes the strategic building blocks, position directions, and allocated companies for this research theme.
Long
Mandated, Price-Free Intermediaries (The Core)
Index and benchmark franchises, clearing and settlement rails, statutory credit data monopolies and compliance-content publishers. Their cost base is 40–70% cognitive labour, which AI deflates, while statute or inclusion effects hold their price. No capital is required against a tail.
Long
Concentrated Adopters
Premium-denominated adopters with externally set prices and labour that can decline through attrition. The saving lands here without the market paying a software multiple for it.
Long
Risk Capital and Conformity Assessment
Reinsurers, specialty MGAs and notified testing bodies. Capturing a share of AI-created value requires either capital against a tail or proximity to the ledger. Own them for earnings compounding, not for a multiple re-rate.
Hold
Ledger Plus Machine Identity
Systems of record are migrating from per-seat pricing to agent-action metering. This defensive re-denomination defends ARPU rather than multiplying it. The return is the de-rating that does not happen.
Long
Physical Bottlenecks
Power, grid, thermal, HBM and advanced packaging are a tax on everyone, but cyclical, capital-hungry and politically exposed. Trade the shortage; never capitalise the peak.
Short
Labour-Denominated Revenue
IT services, BPO, staffing and hourly professional work. Anything denominated in hours, heads or seats is structurally impaired. Size for convexity and event risk rather than carry, and expect gaps rather than grinds.
The analysis
The Permission-and-Proof Economy (Early-to-Mid 2030s)
Capability-adjusted inference is effectively abundant and near-free at the margin, held there permanently by multiple frontier suppliers and sovereign open-weight models from the EU, Gulf, India and Korea acting as a price ceiling on commodity cognition.
Through the early 2030s, the global economy will run a queue. Agents will remove the bulk of routine cognitive labour in high-volume digitisable workflows—code maintenance, claims, medical coding, KYC and tier-1 support—but they will never remove the accountable human or the accountable balance sheet from high-liability decisions.
Regulation will land on deployment tempo and energy through state ratepayer levies, works-council consultation and algorithmic-dismissal statutes rather than as a direct compute tax. It functions as a governance toll that slows rollout and heavily favours balance sheets that can self-insure. Both sides of every intermediation contract are AI-armed, meaning opacity rents are competed away. Only legally or structurally mandated intermediation retains a take-rate.
Where the Analysis Looks Beyond Consensus
The framing error
Investors are treating AI as a traditional software cycle where the vendor captures the value. In reality, AI is a deflator of cognitive labour. The tool layer currently monetises a low single-digit percentage of the labour cost it displaces (roughly $200–$600 per engineer per month against $150k–$300k fully loaded). The first-round surplus sits with the buyer, not the vendor.
The consensus trap
The fashionable answer—buying programming and agentic software vendors—fails on economics. Code authoring is only 11–30% of developer time, and developer compensation is about 15% of a software vendor’s revenue. Because the savings are machine-verifiable and legible to buyers, they will be competed into price.
The real bottleneck
The binding constraint is not cost per task; it is permission and proof. Scarcity has migrated from GPUs (2023–26) to power and substrates (2026–29), and is now settling on verification capacity, liability capital, proprietary process data and agent authorisation rights (2029–35).
The central test
Friction does not block the surplus; friction is the only thing that stops it leaking to customers. To keep the surplus, a business requires a non-deflatable complement. There are only five: a legal mandate, settlement finality, capital standing behind a fat tail, a physical input such as megawatts, or the authorisation to write into a record of legal effect.
The “Write-Right” Rent (The Strongest Counter-View)
If agent authorisation consolidates into two or three stacks—hyperscaler identity, core ledger and attestation—before an open, portable standard emerges, a massive rent exists that needs no statute and no capital. Agent-action metering would then be a genuine take-rate on the adopter’s labour savings, and the systems-of-record gatekeepers would capture the multiple expansion.
Cognitive Deflation vs. Bifurcation
Will frontier models de-commoditise in regulated verticals and claw back the surplus via per-claim outcome pricing? If scaling returns flatten and the binding input becomes expert-generated environments and audited deployment, providers might charge for outcomes. However, charging for outcomes means signing for outcomes, which imposes insurance economics. This only holds if liability is genuinely transferred rather than allocated by tight sublimits.
The Value Chain: Where the Economics Concentrate
| Rank | Layer | Rationale for margin capture |
|---|---|---|
| 1 | Legally protected intermediaries | The cleanest cell. The cost base is heavily cognitive and therefore deflationary, but prices are set by statute or network inclusion. It captures the spread without posting capital against tail risks. |
| 2 | Registers and machine identity | The gatekeepers of the write-right. They tax agentic action, successfully re-denominate seats into actions to defend ARPU and annex the attestation slice of assurance. |
| 3 | Risk capital | Reinsurers and specialty MGAs price autonomous error. The liability ceiling is converted into recurring premium, earned at insurance ROIC and valued at 9–12x earnings. |
| 4 | Metered Cognition Utilities | The three to five frontier providers and sovereign enclaves. They generate enormous absolute earnings, but their moats increasingly rest on residency and distribution rather than raw capability. |
| 5 | Physical bottlenecks | Power, HBM and substrates are a cyclical tax on the ecosystem. Returns are politically capped and capital-hungry. Expect a panic and washout before capital-goods pricing settles. |
| 6 | Conformity assessment | Notified bodies, testing and audit evidence. Mandated volume compounds at mid-teens-to-twenties multiples with high certainty and low glamour. |
Economics leak away from narrow per-seat point software, hours-denominated services, opacity-rent intermediation and rate-regulated adopters, such as health payers and utilities, where formulas ratchet savings back to consumers.
Portfolio Construction
The portfolio expresses the strategic building blocks of the permission-and-proof economy. The ratings attached to each company come from Leeway’s general equity analysis and are independent of this research topic.
The Business-Rating scores the quality of the business model, independent of the share price. The Market-Fit-Rating tests eighteen fundamental figures for how well the company currently fits the market; a negative reading implies expected negative performance. The Cycle-Rating places the valuation in the stock’s own history: a higher figure means the shares are historically cheaper. The Leeway-Score combines the three in equal parts. How the ratings are calculated
The recommendations since publication
From 19 August 2026 to 4 September 2026: Portfolio+2.4%ACWI+0.7%
Mandated, Price-Free Intermediaries (The Core)
LongIndex and benchmark franchises, clearing and settlement rails, statutory credit data monopolies and compliance-content publishers. Their cost base is 40–70% cognitive labour, which AI deflates, while statute or inclusion effects hold their price. No capital is required against a tail.
Role in Thesis: This is the core position. The bet is straightforward: index-licence revenue is denominated in client AUM, pricing is set by private contract rather than a regulated fee schedule, and the cost base is research, data operations and index administration—precisely the labour AI deflates. The strategy is to own outright and add on AI-commoditisation drawdowns.
Valuation Context: MSCI is the cleanest expression of the trade. The market is marking the company down on the fear that artificial intelligence cheapens data. The mechanism actually runs the other way: the cost base falls, but the index licence holds the price. Caveat: It is not cheap in absolute terms. Benchmark-provider fee scrutiny is a live, long-run political risk, and its AUM beta means it will fail the correlation-shock test alongside the rest of the market.
Invalidation Risk: Watch cost-to-revenue and headcount versus run-rate revenue each quarter. The thesis is invalidated if asset owners internalise custom and direct indexing at scale, or if the AUM cycle turns down hard enough to swamp the margin gain.
Role in Thesis: LSEG provides settlement finality, alongside index and regulatory data. We are long the equity to capture a genuine cost-to-income improvement in data operations. The bet is that headcount remains flat-to-down while revenue grows and pricing remains unchanged across index, clearing, post-trade and analytics.
Valuation Context: This is a large-cap expression of friction protecting the surplus, trading at a multiple that currently embeds disruption fear. The trade is the closing of this AI-fear discount. Caveat: It is a complex, acquisitive story with a long history of margin promises. Market-data fee regulation is the specific statutory-rent risk here.
Invalidation Risk: The position is invalidated if EU/UK consolidated tape and market-data fee reform compress the defended prices, or if the Microsoft partnership proves to be a cost line without a corresponding revenue line.
Role in Thesis: Wolters Kluwer delivers compliance content plus audit evidence. The bet is that PLD, CRA and AI Act conformity will drive content and workflow-software demand, while the company deflates its own editorial and support cost base. Accumulate on weakness.
Valuation Context: This is the purest listed compliance-content and audit-evidence franchise, yet it is incorrectly sold as an AI victim. There is statutory demand for proof combined with internal cost deflation, all with no capital posted against a tail event. Caveat: The disruption fear is not baseless in generic legal research, and ledger/ERP incumbents may annex compliance workflows.
Invalidation Risk: Watch organic growth in the tax/audit and legal/regulatory segments, plus EBITA margin against headcount. The thesis breaks if generative AI genuinely substitutes for paid regulatory content in professional workflows.
Role in Thesis: Experian holds a statutory data monopoly on the credit file. The bet relies on a statute-created data monopoly with a large operations and analytics cost base that AI deflates, supplemented by emerging-market volume growth. We are long, but with explicit political sizing discipline.
Valuation Context: No AI-armed counterparty can arbitrage a legal duty to consult the credit file. However, the opportunity is smaller than index and settlement because politics hits here first. Caveat: It carries consumer-credit cycle beta and is the most politically exposed of the mandated longs. Take-rate reform is a real terminal risk.
Invalidation Risk: Watch North America organic growth, headcount and any US or UK access/pricing reform. The thesis is invalidated if regulators attack file access, pricing or data scope.
Role in Thesis: B3 operates as a monopoly exchange, clearing house and central depository with legal finality and a heavily manual cognitive cost base. The bet is that cost-to-income falls while volumes and statutory functions hold.
Valuation Context: This is the cheapest globally available version of the mandated-intermediary trade. The exchange offers the highest cost-out leverage per unit of multiple in the mandated cell, trades at a single-digit-to-low-teens earnings multiple and remains unowned as an AI theme. Caveat: Emerging-market macro and currency dominate short-run returns. There is competition-policy risk to the monopoly, alongside governance and volume cyclicality.
Invalidation Risk: The thesis is invalidated if Brazilian rates or foreign exchange, or a domestic exchange-competition mandate, hits before the cost-out becomes visible in the margins.
Role in Thesis: European mandated-intermediary exposure through settlement finality, clearing and exchange infrastructure, with a cognitive cost base that can decline through attrition rather than forced displacement.
Valuation Context: This is not a multiple-expansion story. It is a slow cost-to-income improvement in a protected market, testing whether rigid European labour delays but does not erase AI-driven operating leverage. The valuation reflects a mature exchange franchise rather than a clean AI re-rating.
Invalidation Risk: European post-trade and market-data fee regulation, works-council consultation, rate-sensitive net-interest income and competition or volume pressure can obscure the operating signal. The thesis is invalidated if headcount and cost-to-income do not improve while pricing and core volumes hold.
Concentrated Adopters
LongPremium-denominated adopters with externally set prices and labour that can decline through attrition. The saving lands here without the market paying a software multiple for it.
Role in Thesis: Concentrated, premium-denominated adopter exposure through an insurer whose claims and underwriting workflows are cognitive, while Japan's shrinking workforce reduces displacement friction.
Valuation Context: This is the adopter leg, not a software re-rating. If AI lowers claims and administration costs while premium pricing holds, earnings compound through quiet attrition. Japan provides the cleanest comparison with Germany and France because labour scarcity makes substitution politically easier.
Invalidation Risk: The payoff is earnings, not multiple expansion. Catastrophe, reserve and autonomous-error losses can overwhelm expense savings, while yen and investment-book volatility obscure the signal. The thesis is invalidated if expense ratios do not improve against headcount and premium growth, or if pricing discipline breaks.
Risk Capital and Conformity Assessment
LongReinsurers, specialty MGAs and notified testing bodies. Capturing a share of AI-created value requires either capital against a tail or proximity to the ledger. Own them for earnings compounding, not for a multiple re-rate.
Role in Thesis: Munich Re operates in the risk-capital layer as the only party that signs for the tail risk. The bet is that affirmative AI performance and E&O cover becomes a durable specialty growth line while mainstream carriers exclude AI outright. Simultaneously, the reinsurer deflates its own actuarial and claims cost base.
Valuation Context: The liability ceiling binds AI adoption. Turning that liability into premium is a real, investable trend, but it will be valued at insurance multiples, not technology or theme multiples. Long the equity for earnings, not for multiple expansion. Caveat: Insurance ROIC means this is never a 20x-plus re-rate. Correlated-risk accumulation is hard to underwrite, and visible AI revenue is not separately disclosed.
Invalidation Risk: Watch specialty and cyber premium growth, disclosed AI-product commentary and loss experience on the first affirmative covers. The thesis is invalidated if a correlated systemic agent failure arrives before pricing and sublimits mature.
Role in Thesis: Bureau Veritas assesses conformity. We are long as a slow-burn volume compounder on mandated conformity assessment. Drivers include full CRA application, EU high-risk AI obligations in December 2027, machinery and product regulation, and ISO 42001-linked insurance discounts creating private demand for third-party assessment.
Valuation Context: This is the listed, mandated and under-modelled slice of assurance. The proof is required by statute, the deadlines are law rather than speculation, and the valuation already exists. Caveat: AI and cyber schemes will remain small versus total revenue for years. This is not a re-rating story, and competition from SGS, Intertek, TÜV and DNV is heavy.
Invalidation Risk: Watch new-service disclosure, organic-growth mix and any notified-body designation for AI and cyber schemes. The thesis is invalidated if conformity assessment is absorbed by self-declaration regimes or annexed by ledger and identity incumbents or cloud compliance services.
Ledger Plus Machine Identity
HoldSystems of record are migrating from per-seat pricing to agent-action metering. This defensive re-denomination defends ARPU rather than multiplying it. The return is the de-rating that does not happen.
Role in Thesis: SAP holds the write-right in customers’ systems of record. The bet is on re-denomination: as human seats erode, the record-keeper migrates to agent-action and consumption metering to defend ARPU. Keep the position small, or simply keep it off the short list.
Valuation Context: If permission is the scarce asset, the toll sits here. SAP can hold its high multiple instead of losing it as seat licences become billed agent actions. Caveat: The valuation already embeds re-denomination success. Its own developer cost-out is legible to customers, and hyperscaler identity may ultimately annex agent authorisation rather than the ERP.
Invalidation Risk: Watch for revenue per customer rising materially rather than merely re-mixing, and for consumption-revenue disclosure. The thesis is invalidated if an open, portable agent-authorisation standard strips the write-right gate, or if migration costs collapse and customers actually leave.
Role in Thesis: Palo Alto Networks supplies machine and agent identity, credential management and authorisation. We are long the equity as the machine- and agent-identity expression: securing secrets and non-human identity for agents acting inside systems of record.
Valuation Context: This represents the profitable slice of assurance that incumbents will annex rather than leave to standalone verification vendors. Caveat: It is expensive and carries large acquisition-digestion risk. Platform giants are the more likely ultimate annexers, and this is a crowded quality-technology long that will correlate to a beta of one in a capex shock.
Invalidation Risk: Watch identity and secrets ARR growth and non-human identity attach rates. The thesis is invalidated on integration failure, or if hyperscaler identity stacks commoditise agent authorisation into a free platform feature.
Physical Bottlenecks
LongPower, grid, thermal, HBM and advanced packaging are a tax on everyone, but cyclical, capital-hungry and politically exposed. Trade the shortage; never capitalise the peak.
Role in Thesis: Physical-bottleneck exposure through ABF build-up film for advanced packaging, held as a time-limited trade rather than a compounder.
Valuation Context: The value comes from a real packaging constraint and qualified supply, not from AI software sentiment. The position captures scarcity while the supply gap persists and should be harvested before new capacity changes the pricing regime.
Invalidation Risk: This is cyclical and capacity-timed. Chinese competition, glass or panel-level packaging substitution, delayed accelerator demand, food-segment weakness and yen exposure can overwhelm the bottleneck thesis. The thesis is invalidated when 2028-plus capacity additions become visible, ABF pricing rolls over or order growth decelerates.
Labour-Denominated Revenue
ShortIT services, BPO, staffing and hourly professional work. Anything denominated in hours, heads or seats is structurally impaired. Size for convexity and event risk rather than carry, and expect gaps rather than grinds.
Role in Thesis: Wipro is the archetype of internally denominated revenue in heads and hours. We are short, or use it to fund the long portfolio, expressed in the weakest large-cap of the FTE-denominated cohort. The bet is that billable-head revenue means AI gains are contractually passed to clients.
Valuation Context: The natural experiment already ran: productivity gains become client savings, not vendor margin. Revenue per employee rises while total revenue stalls, making the 9–13x multiple a value trap. Caveat: It is a consensual and partly discounted short. Because it is cheap, cash-generative and dividend-paying, it is a convexity and event trade rather than a carry short.
Invalidation Risk: Watch headcount versus revenue and rate-card resets on run-work renewals. The thesis is invalidated if the cohort converts successfully to outcome pricing with proprietary IP, or if a mega-deal wave and buybacks squeeze a crowded short.
Role in Thesis: Teleperformance is the per-agent-hour BPO archetype, where the pricing unit itself dissolves. Short, sized small and preferably expressed with defined-risk options given the low multiple. The bet is that deflection and agentic handling remove interaction volume.
Valuation Context: This model has the highest sensitivity to substitution and the least ability to re-denominate. Caveat: It is extremely crowded and optically cheap, with high squeeze and buyout risk. It is already substantially de-rated.
Invalidation Risk: Watch like-for-like volume, revenue per employee and any shift to outcome pricing. The thesis is invalidated on a take-private or strategic bid, or if services-Jevons volume growth genuinely offsets the deflection.
Role in Thesis: Robert Half places hours in accounting, legal and administration—exactly the functions agentic workflows compress. Short as the cleanest displaced-function staffing exposure, with revenue denominated in placed hours and permanent-placement fees.
Valuation Context: Cycle and structure are hard to separate here, but the structural headwind is undeniable. Caveat: The balance sheet is strong and supported by a dividend. It is a crowded trade and already de-rated.
Invalidation Risk: Watch temporary hours, permanent-placement fee mix and whether white-collar job postings in these functions recover with the cycle or stay structurally lower. The thesis is invalidated if this proves purely cyclical and job postings snap back, producing a violent counter-rally.
Role in Thesis: CoreWeave represents the financing-dependent middle: GPU-backed credit and SPVs. Short, preferably via defined-risk puts, as the concentrated expression of the reality that today’s cheap inference sits on someone else’s balance sheet. The bet is that GPU-backed credit, long depreciation lives and customer concentration make this the first thing to break in 2027.
Valuation Context: This isolates the subsidy risk. It pays out in the branch that correlates the rest of the portfolio first. Caveat: Very high volatility, expensive borrow and option premia, and high squeeze risk. Being right on the mechanism but wrong on timing is ruinous.
Invalidation Risk: Watch credit spreads on GPU-backed debt, older-generation rental prices, useful-life assumptions and contract concentration. The thesis is invalidated if hyperscaler demand keeps outrunning supply and take-or-pay backlogs convert cleanly to cash.
Role in Thesis: This is a basket for seat erosion with defined risk. Use it as a liquid instrument to short the narrow per-seat software layer via put spreads, timed into the 2027–28 renewal cycle and December 2027 EU compliance cluster.
Valuation Context: The bet is that gross margins settle at 65–72% as inference enters COGS, while seat quantity erodes and list prices rise to mask it. Caveat: Index composition works against the thesis because it also owns the toll collectors on the ledger and platform side. Timing decay makes this an options trade, not a static position, and it is heavily anticipated by others.
Invalidation Risk: The thesis is invalidated as a naked short because the basket contains the ledger and platform winners this analysis refuses to short; it must therefore be traded as a spread.
Dates
| When | What happens | Why it matters |
|---|---|---|
| 11 September 2026 | EU Cyber Resilience Act exploited-vulnerability reporting obligations bite. | First test of whether provider-side compliance is a real, quantifiable cost line. |
| 9 December 2026 | EU product-liability directive transposition deadline. | Strict-liability exposure for software and AI-enabled products begins to shape contracts and indemnity caps. |
| 1 January 2027 | Mainstream carriers’ absolute generative-AI exclusions push serious deployers into specialty placements. | Premia, sublimits and governance warranties become a visible cost of deployment. |
| 1 December 2027 | EU Annex III high-risk deployer obligations plus full CRA application land. | The mechanical inflection for conformity assessment and audit evidence. |
| No fixed date yet | Enterprise renewal cycle with seat audits, agent-driven headcount compression and in-house rebuilds. | NRR decomposition becomes legible and shows whether recurring revenue still comes from seats. |
| No fixed date yet | Political consultations on statutory rents, including index licences, market-data fees and statutory data access. | The earliest warning that the legally protected earnings this analysis favours are about to be regulated. |
Sequence
| Window | What happens | What it means for the portfolio |
|---|---|---|
| Now to December 2026 | CRA reporting obligations and product-liability transposition begin to harden the liability regime. Mainstream carriers push absolute generative-AI exclusions into 1 Jan 2027 renewals. | Serious deployers move into specialty placements. Seat growth slows everywhere, masked by price and mix, while mandated intermediaries can be built in weakness. |
| First to third quarter 2027 | First genuine financing stress test of the compute subsidy, including depreciation lives versus capex guidance and GPU-backed credit spreads. | This is the correlation-shock window. The mandated-intermediary long leg sells off with everything AI-labelled for two to four quarters before dispersion pays. |
| Fourth quarter 2027 to end-2028 | EU Annex III deployer obligations and full CRA application meet the enterprise renewal cycle. Seat audits, in-house rebuilds and agent-action pricing make the economics visible. | Conformity assessment and audit evidence inflect. The cleanest mandated intermediaries print cost-to-income improvement with unchanged pricing; the re-rating arrives in 2028–29. |
| 2029 to 2030 | The compute subsidy resolves through financing stress or excess capacity. Adopter validation becomes statistically undeniable as a six-to-ten-quarter revenue-per-employee series with stable gross margins. | Services disinflation becomes measurable and opens the 2028+ consumer and duration leg. Both hardware paths favour the same relative position after a panic, not after a gentle transition. |
| 2031 to 2035 | Regime consolidation into three to five metered cognition utilities plus sovereign enclaves. Scarcity has migrated fully to permission, proof, tail capital and megawatts. | Equity value concentrates in mandated trust infrastructure, ledger-plus-identity, risk capital and energy. The political fight shifts to who keeps the surplus. |
Scenario Probabilities
| Path | Weight | What happens |
|---|---|---|
| Base Case (48%) — Dispersed validation, steady grind | 48% | The cost-out is real but arrives via attrition, is partly reinvested and partly absorbed by verification and compliance spend. Analysts attribute margin expansion at mandated intermediaries to ordinary cost discipline, closing the discount slowly over 2028–29. Seat erosion grinds instead of gapping. Systems of record successfully re-denominate to agent actions. The pair earns modestly, the alpha is dispersion rather than direction, and the household keeps the largest share of the gain. |
| Bear Case (30%) — Taxed, competed away or absent | 30% | Agentic reliability plateaus, verification overhead tracks labour savings and no cost pool moves. Alternatively, the surplus is clawed back: frontier providers make sublimit-capped outcome pricing stick, politics regulates the statutory rents this analysis favours, or a financing shock de-rates the whole stack together. Adopters get neither margin nor multiple. |
| Bull Case (22%) — Legible, early validation | 22% | 2027–28 reporting shows mandated intermediaries growing revenue with flat-to-down headcount and unchanged pricing. Per-seat vendors disclose negative seat contribution. Conformity and affirmative-AI insurance bookings inflect ahead of December 2027 deadlines. Supervision hours per completed task finally fall, proving the surplus stays with buyers. The mandated cell re-rates aggressively toward mid-twenties multiples. |
Counter-arguments and Risks
If agent authorisation consolidates into two or three stacks—hyperscaler identity plus core ledger plus attestation—before an open, portable standard emerges, a massive rent exists that needs no statute and no capital. If this happens, agent-action metering is a genuine take-rate on the adopter’s labour savings, and the systems-of-record gatekeepers will capture the multiple expansion, forcing mandated intermediaries to pay the toll rather than collect it.
Unresolved Market Factors
Will frontier models de-commoditise in regulated verticals and claw back the surplus via per-claim outcome pricing? If scaling returns flatten and the binding input becomes expert-generated environments and audited deployment, providers might charge for outcomes. However, charging for outcomes means signing for outcomes, which imposes insurance economics. This only holds if liability is genuinely transferred rather than allocated by tight sublimits.
Will governments attack the very statutory rents this thesis favours? Market-data fee reviews, consolidated tapes and credit-file access reform could regulate these intermediaries the way medical-loss ratios regulate payer overhead. A fiscal response to eroding payroll-tax bases may also land first on the power bill and consumption rather than on profits.
What to Watch: The 11 Vital Signals
- Cost per completed and verified task, plus supervision hours. Production coding agents cost roughly $0.80–$2.60 per successful task against $80–$150 per hour for human labour. The tell is whether supervision hours fall alongside cost.
- NRR decomposition at large per-seat vendors. Look specifically for negative seat contribution masked by list-price and tier increases, since seat growth is decelerating everywhere.
- Headcount versus revenue at mandated intermediaries. Cost-to-income falling with unchanged pricing is the whole long thesis across index, clearing, settlement and statutory data.
- Revenue per employee versus total revenue at FTE-denominated services. The signature of pass-through is productivity up, revenue flat and the multiple static.
- Affirmative AI E&O and performance-market development. Watch for capacity above $25m per risk, premium levels, sublimit structures for model drift, ISO 42001-linked discounts and the first published loss experience.
- AI-assurance line items. Watch whether any listed company begins disclosing AI-assurance or conformity bookings as a separate line. Currently nobody does, which is why the exposure runs through reinsurers and TIC.
- Expense ratio and employee count at concentrated insurers by jurisdiction. Japan versus Germany and France is the natural experiment for whether attrition or works-council consultation governs the cost-out.
- Compute-subsidy stress indicators, including GPU-backed credit spreads, SPV structures, useful-life assumptions and older-generation rental prices.
- Physical-bottleneck pricing for ABF substrate, HBM and advanced packaging, alongside the 2028 capacity clock that signals when the bottleneck trade should end.
- Services-ex-shelter inflation prints in high-services-weight economies from 2029. This is the opening bell for the consumer and duration leg.
- Regulatory consultations on statutory rents. Any consultation touching index licences, market-data fees or statutory data access is the earliest warning that the best cell is about to be regulated.
What would refute it
Three observable events would overturn the assessment.
- Agentic reliability plateaus, and supervision hours do not fall with cost per verified task.
- Frontier providers make outcome pricing stick in regulated verticals while carrying almost no tail.
- An open, portable agent-authorisation standard strips the write-right, or politics regulates index, data and file prices the way a loss ratio regulates payer overhead.
How this analysis is produced
The assessment is produced in several steps. Independent model families answer the same question separately and then attack the results. What you read here has survived several rounds.
- Two independent first theses. The same opening question goes to several model families that cannot see one another. Disagreements are kept, not averaged away.
- Dated evidence. Every claim that depends on facts is broken into individual search questions and answered with dated, sourced web research. Question, answer, sources and timestamp are logged and remain traceable.
- Adversarial review. Several review roles attack the thesis from different angles: one hunts for the strongest refutation, one for the awkward edge cases, one tests whether a path from thesis to share price exists at all, one checks the timeline for contradictions. Each role raises its own questions, which are again answered with evidence.
- Merge, then the next round. The surviving theses are merged into one and attacked again. The counter-position and the unresolved tension on this page come out of that step. They were not bolted on afterwards to look balanced.
- Back to the start. The process runs again until there is a clear result and a list of tradable companies with structural advantages.
Any analysis can be wrong. That is why the falsification criteria and the counter-position sit on the same page as the thesis, not in the small print.
The numbers shown against individual companies do not come from this process. The Business-Rating scores business-model quality, the Market-Fit-Rating eighteen fundamental figures against the current market, the Cycle-Rating the valuation against the stock’s own history. They are documented under the Leeway scores.
Common questions on permission and proof
Who keeps the AI surplus, the user or the vendor?
In the first round the buyer, because tools monetise only a fraction of the labour they displace and the saving is legible to purchasers. The surplus stays durable only where a complement AI cannot cheapen defends the price.
Why do index and settlement providers sit at the core rather than the software users?
Because their price is set by index membership, settlement finality or statute, while artificial intelligence attacks a large cognitive cost base. The market has sold them as data victims. That is the one genuine revaluation in this analysis.
Are ERP and identity a purchase?
A small position, or simply a company that is not on the avoid list. The shift from seat licences to agent actions defends revenue per customer rather than multiplying it. If the write-right is the scarce asset, the toll sits there. That is why this position underpins the rule that durable capture needs a statute or capital.
Which 2026 and 2027 dates matter?
11 September 2026 for cyber-resilience reporting, 9 December 2026 for product liability, 1 January 2027 for insurance exclusions, December 2027 for operator duties, and the 2027-to-2028 enterprise renewal cycle.
Why avoid IT services and outsourced business processes?
Their revenue is measured in heads, hours or agent-hours. Productivity gains from artificial intelligence are contractually passed to customers. Revenue per employee rises, total revenue stalls, and the valuation remains a trap. The position is now consensus, so keep it small and timed to events, not to a running return.
Is this a bet against AI capex?
No. Physical bottlenecks are a trade with an exit, not a compounder. The financing-dependent middle is the hedge that pays when the 2027 stress test correlates the whole complex first.
Read individual companies through the three ratings
Leeway scores business quality, fundamentals and timing separately and shows the reasoning. Try the linked intermediaries, register providers and companies on the avoid list for fourteen days, or read first how the ratings are built.
Company Valuation and Fundamental Analysis
The data is recalculated on a weekly basis and depends on the current market value of the company and the balance sheet figures of the annual financial statements. The market value changes continuously with price changes, the balance sheets are created annually and change the valuation massively. The time of the annual financial statements and the metrics used can be viewed under "Metrics". Further information on how the analyses work can be found as tooltips directly on the analyses as well as in our explanations.
General
The information is intended to support you in your independent decision-making in implementing investment strategies and serves purely informational purposes. Past performance is not a reliable indication of future performance. No warranty can be given for the completeness, correctness and accuracy of the listed content. The information does not constitute specific investment recommendations. We neither know you nor your financial situation and do not provide investment advice. Only licensed investment advisors with knowledge of your personal circumstances may do this. PWP Leeway UG (limited liability) is not an investment advisor and does not collect any personal data for the purpose of investment optimization. PWP Leeway UG (limited liability) is a provider of investment recommendations and investment strategy recommendations. As such, it is registered with and supervised by the Federal Financial Supervisory Authority (BaFin). The legal notices on the use of the website and the General Terms and Conditions of PWP Leeway UG (limited liability) also apply.
All price data are closing prices of the respective stock exchanges. Price information and master data are provided by an external service provider. Furthermore, public trading data, such as provided by Finra.org, is used to analyze market behavior. No warranty can be given for the completeness, correctness and accuracy of the listed content.
