Amazon delivered 6.7 billion parcels in 2025 — more than the U.S. Postal Service.

WHY THIS MATTERS — AE INDUSTRY NOTE — IR50 — E-Commerce Logistics & Last-Mile Delivery April 2026

A reasonable assumption until recently: UPS and FedEx own last-mile delivery. They have the routes, the drivers, the brand, the infrastructure. Amazon ships packages. It doesn’t run the postal system. That assumption broke in 2025. Amazon overtook USPS by volume and is now offering logistics services to companies that don’t even sell on its platform. Its parcel volume grew 9.8% while UPS volume fell 2.6%.

The reason is structural, not cyclical. Labour accounts for 50–60% of last-mile delivery costs. Amazon has spent a decade building a parallel system designed to shrink that number. It now operates over one million warehouse robots. Its DeepFleet AI coordinates their movement. Its 85+ same-day fulfilment centres sit within miles of customers. In March 2026, it acquired RIVR, a robotics company building wheeled-legged machines that climb stairs and navigate gates — automating the final steps of doorstep delivery itself. UPS and FedEx are optimising a human-driven delivery model. Amazon is replacing it.

The stakes are clearest for investors pricing traditional carriers as stable dividend utilities. UPS trades at $87 billion. FedEx at $85 billion. Both generate thin margins — 5.9% to 6.3% — in an industry where the dominant cost input is the one most exposed to automation. The autonomous last-mile delivery market is growing at 23–25% annually. The question is not whether automation reaches commercial scale, but which of the incumbents can adapt before the cost gap becomes irreversible.

The full report maps the cost structure, identifies which carriers retain pricing power, and assesses the timeline for autonomous delivery to become competitively decisive.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

Labour accounts for 50–60% of last-mile delivery costs, and Amazon just acquired a doorstep-delivery robotics company.

AE INDUSTRY NOTE — IR50 April 2026

Ir50 Ecommerce Logistics Last Mile Delivery

112KB ∙ PDF file

Download

Download

KEY FINDINGS

• Labour accounts for 50–60% of last-mile delivery costs, making this industry among the most exposed to automation in the logistics value chain.

• Amazon has deployed over one million warehouse robots and acquired robotics company RIVR in March 2026, opening a structural cost gap with UPS and FedEx.

• The autonomous last-mile delivery market is projected to grow from $1.1–$1.3 billion in 2025 to $8.8–$11.5 billion by 2035.

• UPS volume declined 2.6% in 2025 while Amazon’s parcel volume grew 9.8% — a divergence likely to widen.

• Markets are pricing UPS and FedEx as stable logistics utilities while Amazon builds a robotics-native delivery network that is already the largest U.S. parcel carrier by volume.

OVERVIEW

The last mile accounts for 53% of total shipping costs, up from 41% in 2018, and the largest component of that cost is human labour. This report examines how AI-driven route optimisation, warehouse robotics, and autonomous delivery vehicles are restructuring the cost base of the $178–$200 billion last-mile delivery industry. The competitive dynamics are bifurcating: Amazon is building an integrated robotics-native logistics network — one million robots, 85+ same-day fulfilment centres, drone delivery reaching 30 million customers — while traditional carriers pursue incremental cost savings within a labour-intensive model. The cost crossover point for autonomous delivery has not yet been reached at scale, but the autonomous last-mile market’s 23–25% CAGR signals the transition from pilot programmes to commercial deployment. Investors using current labour cost assumptions in UPS and FedEx valuations may be overstating terminal margins.

Download the full report below.

Ir50 Ecommerce Logistics Last Mile Delivery

112KB ∙ PDF file

Download

Download

Related:

Share

Adobe spends $4.3 billion a year on R&D to build tools that AI can now replicate from a text prompt.

WHY THIS MATTERS — AE COMPANY NOTE — CN68 — Adobe Inc. April 2026

A reasonable investor would look at Adobe and see a company that benefits from AI. Cheaper development costs flowing into a 96% subscription revenue base with annual contracts and 37% operating margins. The standard read is that AI makes Adobe’s tools better and its margins wider. That read is wrong — or at least dangerously incomplete. The cost savings are real but irrelevant to Adobe’s competitive position, because every software company gets the same tailwind. The threat is on the revenue side.

Adobe’s pricing model assumes creative work requires specialised, multi-application workflows. A designer opens Photoshop, moves to Illustrator, edits in Premiere, and pays for access to all of them through a bundled subscription. That assumption breaks when a foundation model can produce the same output from a single prompt. Claude Design launched in April 2026 and triggered a 7% drop in Figma shares. It is bundled into existing AI subscriptions at no extra cost. It does not need to match Photoshop’s full capability. It needs to serve the 80% of creative tasks that never required that capability in the first place.

The consequences land unevenly. Adobe’s PDF and document business has regulatory entrenchment that AI tools cannot bypass. Its enterprise marketing platform is locked in by data integration and multi-year contracts. But the creative tools — the business that built the company — sit directly in the path of AI-native alternatives entering from Canva, Google, Anthropic, and OpenAI simultaneously. Whether Adobe’s defensive integrations and Firefly AI hold the professional tier, or whether the seat-based model compresses from both ends, depends on company-level positioning the pattern alone does not reveal.

The full report maps Adobe’s exposure across four dimensions and identifies which segments defend and which drain.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

Claude Design’s launch triggered a 7% drop in Figma shares and renewed pressure on Adobe — signalling that foundation models now compete directly with creative software.

AE COMPANY NOTE — CN68 April 2026

Cn68 Adobe (1)

132KB ∙ PDF file

Download

Download

KEY FINDINGS

• Adobe’s 96% subscription revenue and annual contracts buffer near-term disruption but do not address the category-level threat from prompt-driven design.

• AI-native tools collapse multi-step creative workflows into single interfaces, weakening the bundle logic behind $19.2 billion in Digital Media ARR.

• With 34% of revenue in SG&A and 18% in R&D, Adobe’s cost base is heavily weighted toward labor now being repriced by AI on a declining cost curve.

• Adobe’s Firefly integration and Anthropic partnership are defensive plays whose strategic value depends on whether professional users still need to “reach into” specialised tools.

• A CEO transition during the AI inflection point introduces execution risk at the moment strategic direction matters most.

OVERVIEW

Adobe generates $23.77 billion in annual revenue from creative tools, document services, and enterprise marketing platforms — nearly all of it through per-seat subscriptions. That pricing model assumes creative work requires specialised, multi-application workflows. AI-native design tools challenge this assumption directly, because they allow users to produce layouts, interfaces, and marketing assets from a text prompt without entering Adobe’s product surface at all. The competitive threat is not a single rival but a swarm: Canva from below with 260 million users, Figma in collaborative design, and foundation model developers — Anthropic, Google, and OpenAI — entering creative workflows from the AI infrastructure layer. Adobe’s financial strength provides time to adapt, but the structural tension between embracing AI and protecting seat-based economics has no costless resolution. The strongest defensive positions are in PDF document workflows and enterprise experience management, where data integration and regulatory entrenchment create moats that AI tools cannot bypass.

Download the full report below.

Cn68 Adobe (1)

132KB ∙ PDF file

Download

Download

Related:

Share

The technology saving oil and gas companies money today is building the oversupply that destroys their margins tomorrow.

WHY THIS MATTERS — AE SECTOR REPORT — SR45 — Oil and Gas April 2026

A reasonable view of AI in oil and gas runs something like this: automation reduces costs, predictive maintenance cuts downtime, and the industry becomes leaner and more profitable. The technology is a tailwind. It helps incumbents. This view is wrong — not because the cost savings are imaginary, but because of where they lead.

AI-driven autonomous drilling, closed-loop fracturing, and digital platform integration are already cutting well-delivery costs by 10–20%. That makes production economic at lower oil prices. Operators drill more. Supply increases. At the same time, manufactured energy — solar, wind, batteries — continues down cost curves that have delivered 97% reductions in solar and 89% in batteries over the past decade. Demand for oil grows more slowly, then peaks, then declines. The cost savings that made drilling cheaper also made overproduction easier, precisely when end-use demand was turning.

This is a compound cascade. From within the sector, AI compresses the cost of producing a barrel. From outside the sector, manufactured energy compresses the number of barrels the world needs. The two forces interact: cheaper drilling supports production volumes that delay the price signal, which delays capital discipline, which leads to a sharper correction when the signal finally arrives.

The consequences concentrate unevenly. Oilfield services firms that sell the automation are cutting into their own future revenue base. Midstream operators with 20-year pipeline contracts face counterparties whose economics are degrading under them. Integrated majors with trillion-dollar asset bases confront a question of timing and balance-sheet structure. But which companies within each industry are positioned on the right side of this divide — and which are building the trap they will fall into — depends on competitive structure, cost position, and asset duration that vary company by company.

The full report maps seven industries across the oil and gas value chain, ranks their exposure to both manufactured labor and manufactured energy, traces the cascade sequence, and identifies where the market has mispriced the cross-industry dynamics.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

The oil and gas sector’s defining disruption is a compound cascade: AI cuts well-delivery costs while manufactured energy erodes demand for the product those wells produce.

AE SECTOR REPORT — SR45 April 2026

Sr45 Oil And Gas

161KB ∙ PDF file

Download

Download

KEY FINDINGS

• Oilfield services sits earliest on the S-curve, with autonomous drilling and AI-enabled completions already delivering 10–20% cost reductions across field operations.

• The sector’s cascade runs from OFS automation through upstream oversupply to midstream stranding risk — a sequence the market has not priced.

• Midstream infrastructure carries the strongest friction moats but faces long-duration stranding as throughput assumptions in 20–30 year contracts erode.

• Downstream refining faces a compound squeeze as AI automates process control while manufactured energy displaces transportation fuel demand.

• The competition assumption holds unevenly: oilfield services is fragmented enough for rapid cost pass-through, while national oil companies block the mechanism entirely.

• The market prices each industry in isolation but misses the cross-industry feedback loop in which cheaper drilling accelerates the oversupply that destroys upstream margins.

OVERVIEW

This sector report maps seven industries within oil and gas — upstream E&P, oilfield services, midstream infrastructure, downstream refining, LNG, petrochemicals, and fuel retail — and traces how manufactured labor and manufactured energy reshape each one. The sector faces simultaneous disruption from within and without: AI-driven automation enters through oilfield services, reducing workforce requirements and well-delivery costs, while declining solar, battery, and wind costs substitute for hydrocarbons in end-use markets. The critical finding is a compound cascade in which OFS automation temporarily supports production volumes by lowering break-even costs, which accelerates the oversupply that demand erosion creates, ultimately collapsing revenue for the service providers that enabled the cost reduction. Midstream infrastructure, valued on distributable cash flow from long-duration contracts, is the sector’s most likely mispricing because contract coverage obscures the counterparty solvency risk that the cascade creates.

Download the full report below.

Sr45 Oil And Gas

161KB ∙ PDF file

Download

Download

Related:

Share

Four of the five largest solar module manufacturers reported losses in 2024 despite shipping record volumes into a market that grew 33%.

WHY THIS MATTERS — AE INDUSTRY NOTE — IR49 — Solar Panel & Module Manufacturing April 2026

Solar is the fastest-growing energy technology on earth. Installations hit 600 GW in 2024. A smart observer would expect the companies making solar panels to be thriving. More demand, more shipments, more profit. That is how growth industries are supposed to work.

It is not how this one works. Global manufacturing capacity reached 1.8 TW by end-2025 — three times what the market actually installed. That 3:1 overcapacity ratio turned the module market into a price war where the product sells below production cost. Module prices fell below $0.10 per watt in 2024, down from $0.25 per watt in 2020, because the same learning curve that made solar panels cheap — a 25–30% cost decline per cumulative doubling — never stops operating. Every new gigawatt shipped pushes the price lower. The manufacturers cannot outrun the deflation they produce.

LONGi lost $1.2 billion. Trina Solar lost $473 million. JinkoSolar, the global shipment leader at 93 GW, scraped through with a marginal $7.9 million net income on $12.6 billion in revenue. The sole exception was First Solar, which earned $1.29 billion — a 31% net margin — on roughly one-seventh of JinkoSolar’s volume. First Solar’s advantage is not cost efficiency. It is U.S. industrial policy: Section 45X manufacturing tax credits and tariffs that block Chinese competition from the American market.

That creates two separate questions for investors. The first is whether the policy protection that makes First Solar profitable will hold. The second is which of the major Chinese manufacturers will survive a shakeout that analysts expect to last into 2027. Scale alone does not determine the answer — balance-sheet structure, technology bets, and access to protected markets all matter, and the ranking of survivors is not what volume share would suggest.

The full report breaks down each major manufacturer’s cost exposure, competitive position, and structural vulnerability.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

Global module manufacturing capacity is three times larger than installation demand, and the resulting price war has pushed most of the world’s largest manufacturers into loss.

AE INDUSTRY NOTE — IR49 April 2026

Ir49 Solar Panel Module Manufacturing

114KB ∙ PDF file

Download

Download

KEY FINDINGS

• Manufacturing capacity of ~1.8 TW exceeds ~600 GW of annual installations, driving module prices below cost for most producers.

• First Solar earned $1.29 billion net income in 2024; LONGi, Trina Solar, and JA Solar all reported significant losses.

• Labour is only 10–15% of operational costs; the primary exposure is on the revenue side, where the product itself deflates at 25–30% per doubling.

• U.S. tariffs and IRA manufacturing credits have created the only structurally profitable niche in the global module market.

• Industry consolidation is underway and expected to persist into 2026–2027 as financially weaker manufacturers exit.

OVERVIEW

Solar panel manufacturing occupies a unique structural position: it produces the technology that defines falling energy costs, yet the same cost curve that drives demand growth simultaneously compresses the revenue earned on every unit sold. Module prices fell below $0.10 per watt in 2024, down from approximately $0.25 per watt in 2020, because Wright’s Law learning rates of 25–30% per cumulative doubling continue to operate on an industry with massive and growing production volumes. The competitive structure is intensely fragmented, with hundreds of manufacturers and overcapacity of roughly 3:1, ensuring that cost savings pass almost entirely to customers rather than expanding producer margins. JinkoSolar shipped a record 93 GW of modules in 2024 yet reported an operating loss. First Solar, shielded by U.S. industrial policy, posted a 31% net margin on the same product category. The investable value creation from falling module costs sits primarily downstream — with project developers, utilities, and storage integrators — rather than with the manufacturers who produce it.

Download the full report below.

Ir49 Solar Panel Module Manufacturing

114KB ∙ PDF file

Download

Download

Related:

Share

Global AI spending is forecast to reach $2.5 trillion in 2026, already exceeding the dot-com telecom buildout as a share of GDP.

AE BRIEFING — BR7 April 2026

Br7 Ai Capital Rewards Users Not Builders

95.6KB ∙ PDF file

Download

Download

KEY FINDINGS

• Hyperscaler capex is approaching $560–680 billion in 2026, rivalling the railroad boom of the 1880s as a share of U.S. GDP.

• Per-token inference costs have fallen 9x to 900x per year, yet total enterprise AI spending is rising as usage outpaces price decline.

• Firms with the highest capital expenditure growth have underperformed peers by approximately 8.4% annually since 1963.

• Railroad construction-phase investors were wiped out; monopoly rents accrued to post-consolidation operators who bought distressed assets at a fraction of construction cost.

• The AI buildout’s pre-consolidated structure may alter the capital cycle mechanism, but the directional pressure on returns remains consistent with historical precedent.

OVERVIEW

This briefing examines the largest private-sector infrastructure investment since the railroads and asks who captures the value. Hyperscaler capital expenditure has grown from $107 billion in 2020 to consensus estimates of $560–680 billion for 2026, funded primarily from operating cash flow. The capital cycle evidence from railroads, fiber optics, and telecom is consistent: construction-phase capital earned poor returns while users of the surplus infrastructure captured the economic value. The AI buildout differs structurally because it is pre-consolidated in five to seven firms rather than dispersed across hundreds of speculative entrants, which may shift the correction mechanism from bankruptcy to earnings and shareholder pressure. But the underlying dynamic — massive capital deployed into rapidly depreciating assets on a declining cost curve — creates the same directional force. The briefing traces the mechanism, identifies where the thesis could break, and outlines what it means for capital allocation.

Download the full report below.

Br7 Ai Capital Rewards Users Not Builders

95.6KB ∙ PDF file

Download

Download

Related:

Share

Airbnb runs an 83% gross margin on 8,200 employees — and none of them clean a room or check in a guest.

WHY THIS MATTERS — AE COMPANY NOTE — CN67 — Airbnb April 2026

A reasonable investor looking at Airbnb’s exposure to AI would conclude the company is well protected. It owns no properties. It employs no housekeepers. Hosts bear the labour and energy costs that AI-driven automation is repricing across the hospitality industry. That assessment is correct — on the cost side. It misses the exposure that matters.

Airbnb’s most valuable competitive advantages are friction-based. Network effects, trust infrastructure, review systems, supply breadth — each depends on it being expensive for hosts and guests to find each other outside the platform. AI is dissolving exactly that friction. Travel agents powered by large language models can already search across platforms, compare listings on price and quality, and assemble bookings without ever opening Airbnb’s app. As these tools gain consumer trust, they erode the discovery cost that keeps both sides of the marketplace locked in.

The cost tailwind is real but competitively neutral. Airbnb, Booking, and Expedia all access the same AI tools for customer support, content moderation, and marketing optimisation. The savings flow to whichever platform implements fastest — then get competed away. The demand-side threat is different. It attacks the mechanism that justifies Airbnb’s take rate. If guests stop searching on Airbnb and start delegating to AI agents that route across all platforms, the 14–16% service fee faces pressure from a channel that didn’t exist two years ago. The most exposed segment is professional multi-property hosts, who will shift to whatever platform maximises occupancy — and they generate disproportionate revenue.

Whether Airbnb’s $2.4B annual product development spend and low balance-sheet rigidity are enough to outrun this shift depends on specifics the article can’t resolve. The full report maps each of Airbnb’s moats, classifies which are structural and which are draining, and assesses the competitive timeline against three named peers.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

Subscribe now

Airbnb earns an 83% gross margin because hosts bear every property and labour cost — but that insulation has a blind spot.

AE COMPANY NOTE — CN67 April 2026

Cn67 Airbnb

131KB ∙ PDF file

Download

Download

KEY FINDINGS

• Airbnb’s asset-light model makes its cost structure more resilient to AI-driven disruption than hotel chains or competing travel platforms.

• AI-native travel agents threaten Airbnb’s demand side by routing bookings without passing through its interface.

• Product development at 19.2% of revenue ($2.4B) is high for travel, but competitors access the same AI tools within months.

• Millions of uncoordinated individual hosts become a vulnerability if AI-driven property management lets new platforms aggregate supply more efficiently.

• Annual free cash flow of $4.6B and minimal fixed obligations give Airbnb substantial room to restructure.

OVERVIEW

Airbnb’s marketplace model shifts property ownership, maintenance, and guest-facing labour to millions of individual hosts, leaving the platform with a cost base dominated by engineering, marketing, and customer support. Declining AI costs reduce these operating expenses while leaving the service-fee revenue model intact — a tailwind, not a threat — but one available equally to Booking, Expedia, and smaller competitors, meaning competitive dynamics are likely to force much of the saving through to users rather than expanding margins permanently. The structural question sits on the demand side. Airbnb’s most valuable moats — network effects, trust infrastructure, supply breadth — are friction-based, holding only as long as it remains expensive for hosts and guests to find each other outside the platform. AI-powered travel agents that compare and book across platforms erode exactly that friction. The company’s strong balance sheet and low adaptation rigidity buy time, but the speed of AI agent adoption is the variable the market may be mispricing.

Download the full report below.

Cn67 Airbnb

131KB ∙ PDF file

Download

Download

Related:

Share

The UK’s 6.18 million public-sector workers represent the single largest concentration of labour-cost exposure in the British economy.

AE SECTOR REPORT — SR44 April 2026

Sr44 Uk Public Sector

178KB ∙ PDF file

Download

Download

KEY FINDINGS

• Administrative functions in central government and welfare processing face the highest manufactured labour exposure, with AI task substitution already underway.

• The NHS’s £202 billion budget is 70% staffing costs, creating the sector’s largest pool of labour-cost reduction potential.

• UK government AI procurement reached £1.17 billion in 2025, up 102% year-on-year, yet operational deployment ranks sixth out of ten comparator nations.

• The incumbent trap is structurally more severe in the public sector because legacy systems, union agreements, and political accountability block market-style restructuring.

• Defence capital intensity is reaching 43% of total MoD spend — the highest since at least 2002 — creating significant exposure to hardware cost curves.

OVERVIEW

This report maps the UK public sector’s eight major industries against the cost pressures created by declining AI and energy costs. The central finding is that disruption follows the task structure of each industry, not its political priority. Administrative processing — tax collection, benefits assessment, correspondence handling — reaches the adoption threshold first because these tasks are almost entirely labour-based and face no physical-presence constraint. The cascade runs from there: efficiency savings in back-office functions generate fiscal headroom that political pressure redirects to the NHS, defence, or education rather than returning it to Treasury. This creates a distinctive dynamic where cost reduction in one industry funds expanded activity in another. For investors, the primary exposure runs through the private firms supplying IT services, AI tools, and managed services to government — a market growing at over 100% annually from a still-small base.

Download the full report below.

Sr44 Uk Public Sector

178KB ∙ PDF file

Download

Download

Related:

Share

CrowdStrike earns 95% of its revenue from subscriptions, yet spends 78% of that revenue on labour-intensive sales, marketing, and R&D.

AE COMPANY NOTE — CN66 April 2026

Cn66 Crowdstrike

131KB ∙ PDF file

Download

Download

KEY FINDINGS

• CrowdStrike’s cloud-native architecture and AI-driven detection position it as a structural beneficiary of cheaper AI-powered labour.

• Stock-based compensation of $841 million — 21% of revenue — masks a biological-labour dependence that AI has not yet displaced.

• Platform consolidation with 48% of customers on six or more modules creates switching costs, but AI also lowers barriers for niche entrants.

• The July 2024 global outage created a lasting concentration-of-risk liability that competitors continue to exploit in enterprise procurement.

• At 27x trailing revenue, the market prices CrowdStrike for sustained premium growth with limited margin for error.

OVERVIEW

CrowdStrike occupies an unusual structural position: it sells AI-powered security while remaining heavily dependent on human labour to build and distribute that product. The Falcon platform automates threat detection and response tasks that previously required human analysts, giving CrowdStrike a direct role in deploying manufactured labour across its customers’ security operations. But the same cost dynamics that empower CrowdStrike’s product also enable AI-native competitors and hyperscaler security bundles from Microsoft, Google, and Amazon. The company’s asset-light balance sheet, $4.3 billion cash position, and management’s demonstrated AI-first orientation reduce adaptation risk. The core question is whether platform switching costs and data network effects can sustain pricing premiums as AI compresses the cost of building comparable detection capability. This note assesses CrowdStrike’s structural exposure across cost, competitive, and swarm dimensions.

Download the full report below.

Cn66 Crowdstrike

131KB ∙ PDF file

Download

Download

Related:

Share

Anthropic’s Mythos model can find software vulnerabilities that human experts missed for decades, but the April cybersecurity selloff confused which companies benefit versus which face disruption.

AE BRIEFING — BR6 April 2026

Br6 Cybersecurity After Mythos

80.7KB ∙ PDF file

Download

Download

KEY FINDINGS

• AI gives attackers and defenders the same tools, but attackers need to succeed only once while defenders must succeed everywhere.

• The April cybersecurity selloff confused vulnerability scanning firms that are disrupted with attack detection firms that benefit.

• The USD 75 billion cybersecurity services market faces direct compression because AI performs these tasks at near-zero cost.

• Companies with accumulated network data, identity platforms, and data recovery systems retain strong defensive moats.

• Anthropic’s restricted Mythos access to 50 major firms creates regulatory advantages for large vendors.

• Offensive AI capabilities will become widely available within 18 months, but defensive capabilities remain concentrated.

OVERVIEW

AI automation affects cybersecurity differently than other professional services because both attackers and defenders gain the same capability simultaneously. But there’s a crucial mathematical imbalance: an attacker only needs to find one exploitable vulnerability to succeed, while a defender needs to find and fix every vulnerability to stay secure. This asymmetry means offense benefits more from automation than defense. The result isn’t industry-wide commoditization but redistribution of spending away from hourly services toward real-time detection platforms, identity management, and data recovery systems. Total cybersecurity spending will actually increase as AI expands the attack surface faster than it improves defenses.

Download the full report below.

Br6 Cybersecurity After Mythos

80.7KB ∙ PDF file

Download

Download

Related:

Share

The global advertising market crossed $1 trillion in 2025 — the same year AI agents began completing purchases without consumers ever seeing an ad.

AE BRIEFING — BR6 April 2026

Br6 Agentic Commerce Advertising Threat

87.5KB ∙ PDF file

Download

Download

KEY FINDINGS

• Agentic buying attacks advertising’s foundational assumption: that a human sees and responds to an ad before transacting.

• Alphabet, Meta, and Amazon collectively control over half of global ad spend but face structurally different variants of the threat.

• AI-attributed orders on Shopify rose elevenfold between January 2025 and early 2026 while AI chatbot traffic to US retail sites increased 670% year-over-year.

• Incumbent platforms face an Incumbent Trap: enabling agentic commerce cannibalises their advertising revenue, resisting it cedes the agentic layer to new entrants.

• The competition assumption holds unevenly — open protocols distribute agentic power, but dominant platforms retain enough control over identity and data to slow the transition.

OVERVIEW

Digital advertising monetises human attention at the point of decision. Agentic buying — where AI agents autonomously discover, compare, and purchase products — removes the consumer from the environments where advertisements are displayed. This briefing traces the mechanism through each of the three dominant advertising platforms. Google Search is most directly exposed because its value depends on capturing declared consumer intent; agents that bypass search eliminate the impression entirely. Meta faces erosion of browsing-based purchasing as routine transactions shift to agent-mediated channels. Amazon confronts a paradox: its marketplace advertising depends on consumer presence, but external agents can route purchases to competing merchants. The infrastructure for this transition is already in production — Google’s Universal Commerce Protocol, Shopify’s Agentic Storefronts, and OpenAI’s Instant Checkout are live, not experimental. The timing is uncertain, but the mechanism is observable and the direction is structural.

Download the full report below.

Br6 Agentic Commerce Advertising Threat

87.5KB ∙ PDF file

Download

Download

Related:

Share

Over two million people work in McDonald’s restaurants. Almost none of them work for McDonald’s.

WHY THIS MATTERS — AE COMPANY NOTE — CN63 — McDonald’s Corporation April 2026

The assumption runs through most investor models of fast food: McDonald’s bears the restaurant wage bill, and every minimum-wage increase in California, France, or the UK compresses corporate margin. That belief is structurally wrong. About 95% of McDonald’s 45,356 restaurants are franchised. The people inside those restaurants are employed by the franchisee, not by the corporate entity.

McDonald’s earns money two ways: a royalty on franchisee sales and rent on real estate it owns or controls. Neither of those lines moves with restaurant labour cost. When AI voice ordering, automated scheduling, and computer-vision order checking lower franchisee wage bills, the gain lands on the franchisee P&L first. It reaches McDonald’s corporate margin only indirectly, through healthier franchisees who can pay royalties, cover rent, and finance new restaurants.

This changes how the sector should be modelled. Every investor writing minimum-wage exposure into a McDonald’s thesis is modelling the wrong entity. The same logic runs across the quick-service restaurant majors — Restaurant Brands International, Yum! Brands, and Starbucks all operate variants of the same asset-light franchise model. But who actually retains the AI-labour cost decline as margin, and who ends up competing it back to customers through deeper value menus, is not a question the franchise structure alone answers.

The full report maps McDonald’s structural exposure across cost, competitive, swarm, and adaptation dimensions — and explains why the answer for McDonald’s is not the answer for its closest peers.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

McDonald’s corporate P&L doesn’t bear the restaurant labour cost that AI is about to compress — 95% of its 45,356 restaurants are franchised.

AE COMPANY NOTE — CN63 April 2026

Cn63 Mcdonalds

147KB ∙ PDF file

Download

Download

KEY FINDINGS

• Approximately 95% of McDonald’s 45,356 restaurants are franchised, so restaurant labour sits on franchisee P&Ls rather than the corporate income statement.

• Drive-thru voice AI, computer-vision order checking, and AI scheduling protect franchisee unit economics first, because franchisee profitability governs royalty capacity.

• A 46.1% operating margin and 31.8% net margin in FY2025 put McDonald’s at the top of quick-service restaurant economics globally.

• The same Google Cloud AI stack is available to Burger King, Yum, and regional peers, so industry savings likely pass through to consumer value menus.

• The binding adaptation constraint is not balance sheet — it is propagation speed across roughly 43,000 independent franchisee capital decisions.

OVERVIEW

McDonald’s structural position under AI-driven cost deflation is unusual. The company operates as a franchisor rather than a restaurant operator: royalties, rent from franchisee-occupied real estate, and licence fees dominate a US$26.9 billion corporate revenue base. The roughly two million people who work in McDonald’s restaurants are employed by franchisees, not by the corporate entity. When AI voice ordering, computer-vision order accuracy, and AI-driven scheduling lower restaurant labour cost, the immediate beneficiary is the franchisee P&L, not the corporate margin. The corporate parent captures the gain indirectly, because healthier franchisee unit economics sustain royalty capacity, rent coverage, and new-restaurant development. The harder question is whether McDonald’s can retain this benefit as Burger King, KFC, Taco Bell, and regional competitors deploy the same AI stack — or whether competitive pass-through converts an industry-wide tailwind into deeper consumer value menus.

Download the full report below.

Cn63 Mcdonalds

147KB ∙ PDF file

Download

Download

Related:

Share

SLB’s revenue fell 2% in 2025 but its adjusted EBITDA margin held near 24% — a trough-year profile that usually takes a price recovery.

WHY THIS MATTERS — AE INDUSTRY NOTE — IR48 — Oilfield Services April 2026

The assumption was straightforward. Oilfield services is cyclical. Revenue declines, margins compress with it, operators extract price concessions, and EBITDA follows the rig count down. That did not happen in 2025. Baker Hughes posted record annual adjusted EBITDA of $4.83 billion, up 5%, on flat revenue. Halliburton held adjusted operating margin at 14% through a 3% revenue decline.

The cost structure is being reshaped mid-downturn. SLB’s autonomous geosteering system completed 25 trajectory changes on a single well in seconds — work that used to require a geologist, a directional driller, and a well-plan engineer in constant communication. Baker Hughes reports 20% reductions in unplanned downtime from AI-driven predictive maintenance. The savings are large enough that cost structure is improving faster than price is compressing.

This matters for investors holding energy exposure, for operators negotiating service contracts, and for every junior petrotechnical professional whose first role has just been automated. The 2024–2026 layoffs at Chevron, ConocoPhillips, Shell, BP, Halliburton, and NOV now look structural rather than cyclical. The pattern creates winners among the service majors — but which of the five largest retain the margin captured from automation, and which watch it flow through to customers, is not what most energy investors assume.

The report answers that question through segment-level margin mechanics and a competitive breakdown of where pricing power persists and where pass-through is complete.

The full report is available here:

This article is part of the Why This Matters series from Abundance Economics — short translations of our research for readers who want the insight before the full report.

Related:

Share

SLB ran a 24% adjusted EBITDA margin in 2025 while revenue fell — automation is holding margins where pricing cannot.

AE INDUSTRY NOTE — IR48 April 2026

Ir48 Oilfield Services

131KB ∙ PDF file

Download

Download

KEY FINDINGS

• Oilfield services is absorbing a cyclical downturn and structural automation at the same moment, making the 2024–2026 headcount reductions hard to reverse.

• SLB’s revenue fell 2% in 2025 but adjusted EBITDA margin held near 24% — cost structure is improving faster than price compression.

• Below the top three, operators retain bargaining power and capture most automation savings through price rather than service-company margin.

• SLB’s ChampionX acquisition and Baker Hughes’s $32.4B IET backlog move core oilfield work into adjacencies where scale compounds rather than erodes.

• Junior petrophysical and geosteering roles are being absorbed by autonomous systems — the pipeline through which senior expertise was historically built.

• Cheaper manufactured energy helps electrified fleets today but erodes long-run demand for the industry’s end-product, compressing terminal value on subsea backlogs.

OVERVIEW

The oilfield services industry entered 2026 with a contradiction. The majors ran their lowest full-year revenues since the recovery began, yet adjusted EBITDA margins at SLB and Baker Hughes held near cycle highs. The reason: the same cost structure that would normally compress in a downturn is being reshaped by AI and autonomous systems. SLB’s Neuro autonomous geosteering completed 25 trajectory changes on a single lateral in seconds; Baker Hughes reports 20% reductions in unplanned downtime from AI-driven predictive maintenance. Each replaces task bundles that used to justify junior roles at both operators and service companies. The cyclical layoffs of 2024–2026 are layered on top of structural substitution, which is why the headcount reductions exceed what oil-price weakness alone would justify. The full report examines which segments retain pricing power and which face permanent margin compression.

Download the full report below.

Ir48 Oilfield Services

131KB ∙ PDF file

Download

Download

Related:

Share

Inference costs have fallen between 9x and 900x per year across benchmarks since 2023; one week of AI agent launches in February 2026 then removed roughly $285 billion from enterprise software market value.

AE BRIEFING — BR5 April 2026

Br5 Agent Swarm Targeted Attacks

93.5KB ∙ PDF file

Download

Download

KEY FINDINGS

• Per-token inference prices have fallen between 9x and 900x per year across benchmarks, making manufactured reasoning the clearest falling cost curve in the economy.

• The first-order consequence has been software pricing compression rather than labour automation; a single week of agent launches erased roughly $285 billion in market value.

• Cost collapse has triggered swarm entry: median time to $100 million in cumulative funding for agentic startups has fallen from three years to eighteen months.

• Only 11 to 14 percent of enterprises have agents in production, and Gartner expects over 40 percent of agentic AI projects to be cancelled by end of 2027.

• Open-source models, usage-based pricing, and open protocols such as MCP are routing cost savings to buyers rather than letting them be captured as provider rents.

OVERVIEW

The briefing argues that AI agents are arriving at enterprises first as a compression of software pricing rather than as a displacement of workers. The mechanism is straightforward: foundation-model inference costs, the input to any agent, are falling by multiple orders of magnitude per year because of learning-curve effects on GPUs, silicon, and model design, while per-seat software prices were set when knowledge work was bottlenecked by human attention. When that bottleneck breaks, the license-count denominator loses its economic meaning. A swarm of well-funded vertical agent startups then attacks incumbent software from a hundred specialised angles simultaneously, faster than incumbents can refactor. The repricing is not uniform. Categories where the underlying system of record is deterministic and mission-critical retain more defensibility; generic per-seat interfaces to structured data face the steepest displacement.

Download the full report below.

Br5 Agent Swarm Targeted Attacks

93.5KB ∙ PDF file

Download

Download

Related:

Share