Abundance Economics Research

Latest article on e27. Anthropic may look stronger than OpenAI on the usual pre-IPO scoreboard. It has reported stronger private-market momentum. It appears closer to near-term operating profit. Its gross margin is reported above OpenAI’s. It has deep enterprise relationships, a strong reputation with developers, and Claude Code has become one of the clearest examples of an AI product that customers already pay for at scale. On the surface, that looks like the cleaner business. But the more useful question is not which company looks better today. It is which company gets stronger as AI does what everyone expects it to do: improve and get cheaper. On that question, OpenAI and Anthropic are not the same kind of company. Anthropic mainly sells access to frontier intelligence. OpenAI sells that too, but it also controls a mass consumer interface used by hundreds of millions of people. That difference matters because falling AI costs do not affect both businesses in the same way. For a frontier model seller, cheaper AI erodes the price of the thing being sold. For a consumer platform, cheaper AI lowers the cost of serving users whose attention can be monetised through advertising, commerce, subscriptions, and referrals. The same cost curve can damage one business model and strengthen another. Link in comment

U.S. nursing homes face a margin trap — while the pricing mechanism trapping them is about to flip. In most industries, when one firm cuts costs, competition forces prices down and the advantage disappears. Nursing homes are different. Roughly two-thirds of revenue comes from government payers at administratively fixed rates that don't move when an individual operator gets more efficient. Because rates are set centrally, an operator that cuts documentation time or reduces agency staffing keeps the full saving. AI-driven labour reduction translates directly into permanent margin expansion — with no competitive price erosion. The December 2025 repeal of the CMS minimum staffing mandate removed the main regulatory barrier. Which operators have the capital to capture this, and which are already running negative margins, is where the gap widens. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

The UK's 70% workforce AI-exposure rate is the highest in the G7 — and its employment structure explains why. The UK's 83% services employment share is not just a GDP statistic. It means the economy is disproportionately concentrated in professional and financial services, where human labour accounts for 60–80% of total costs — the exact cost structure that makes substitution economically compelling as AI task costs fall. Job postings in high-exposure occupations have already fallen 38% since 2022, nearly double the decline in low-exposure roles. Because aggregate employment remains stable, the contraction is invisible in headline data. The Employment Rights Act 2025 raises the cost of human headcount from April 2026, widening the gap against falling AI costs. Which UK-listed services sectors absorb the pressure — and which pass it on? Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

The US electrical grid equipment shortage is funding the engineering automation that will end it. Data centers are projected to reach 40% of US electrical-equipment demand by 2030, and the supply crunch that followed has pushed manufacturer margins high enough to finance something the market hasn't priced: the standardization and automation that erode bespoke production economics. Changshu Switchgear reports 60% shorter design cycles and 50% lower development cost through digital engineering reuse alone. Because elevated margins finance the transition, the shortage accelerates its own resolution. Customization shifts from competitive barrier to variant-cost burden as configuration becomes reusable software. Which manufacturers exit the shortage with structural cost advantage — and which exit with a fixed-cost liability? Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

Toyota's lean manufacturing — the envy of every automaker — was built to optimise the input now being automated away. The company employs 383,853 people across 63 plants. Competitors like BYD aren't retrofitting legacy facilities — they're building new ones designed with 60–80% fewer assembly workers per vehicle from the start. Because those factories begin without inherited labour structures, their cost base compounds downward as automation improves. Toyota's does not. FY2026 guidance shows ¥995.5 billion in operating income disappearing in a single year. The hybrid portfolio and ¥21.5 trillion equity base extend the runway — but they don't resolve the underlying cost divergence in volume segments. Whether Toyota's financial strength converts into structural repositioning fast enough is the question the guidance doesn't answer. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

PayPal's 439 million active accounts are shrinking in value per account as AI removes the friction that made them sticky. Transactions per active account declined in Q1 2026. That decline matters because PayPal's moat was never purely about scale — it was about the cost of switching. AI agents eliminate exactly that cost, converting stored credentials from lock-in into one option among many that an agent can route around. Card-network fees, regulatory obligations and credit-loss exposure remain. Those costs don't compress just because routing becomes automated. Whether PayPal's identity and protection layer earns pricing power inside agentic commerce — or becomes commodity processing selected by someone else's platform — is the question the account count cannot answer. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

The industrial gases oligopoly controls 75–80% of a $94–110 billion market — and its moat is becoming the threat. Pipeline infrastructure that costs billions to replicate has kept three firms dominant for decades. That infrastructure advantage only holds when production stays centralised. Electrolyser costs for producing hydrogen have fallen approximately 70% since 2012, because of a 12–15% learning rate per doubling of cumulative production. That decline is shifting production toward distributed, on-site generation — which means the pipeline network incumbents built to lock in customers is now the asset buyers can route around. Which cost structures survive that bypass — and which balance sheets have the timing to adapt — the post doesn't settle. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

Hyundai Motor is buying robot capability — and may be blocked from using it where it matters most. The pending SoftBank stake acquisition frames the debate around price. The more consequential question is whether Hyundai can deploy Atlas humanoids across its own assembly lines before deployment learning commoditizes — and Korea's Metal Workers' Union has already stated it will block domestic rollout without a labor-management agreement. This matters because Atlas unit costs are publicly estimated and available industry-wide, meaning hardware economics alone confer no durable advantage. The edge, if it exists, comes from accumulating safety-validated procedures and failure libraries on live production lines ahead of peers. Does deployment learning compound fast enough — and broadly enough — to move Hyundai's per-vehicle cost structure before that window closes? Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

McDonald's has 45,356 restaurants and a wage bill that belongs almost entirely to someone else. That combination breaks how most fast-food labour models are built. Roughly 95% of those restaurants are franchised, meaning the people inside them are employed by franchisees, not by McDonald's corporate. McDonald's earns royalties on sales and rent on real estate — neither line moves directly with restaurant labour cost. When AI scheduling or automated ordering cuts franchisee wage bills, the margin gain lands on the franchisee first. It reaches McDonald's only indirectly, because healthier franchisees can sustain royalties, cover rent, and open new locations. The same structure runs through Restaurant Brands International, Yum! Brands, and Starbucks. Which entity in that chain actually retains AI-driven labour savings as durable margin — and which competes it away through value menus — is the question the franchise model alone cannot answer. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.

IBM's mainframe moat is eroding and expanding from the same source — cheap AI tools that cut COBOL migration costs by up to 94%. That tension is not a paradox. Because COBOL expertise has always been the friction, not the technology itself, AI lowers the cost of leaving and the cost of building new workloads simultaneously. Customers gain renewal leverage before any workload moves; IBM gains the ability to run inference on hardware previously gated by specialist-labor scarcity. IBM's Software segment posted a 33% profit margin in FY2025. Whether Transaction Processing revenue tracks mainframe capacity upward or decouples from it will determine whether that margin is deepening or draining. Which direction the annuity moves is not yet visible in disclosed metrics. Follow me for mechanism-based analysis of AI, markets, and corporate strategy. Full AER reports are linked from my profile.