In From Monopoly to Age of Empires1, we described a world fragmenting into competing technological and economic blocs. Technology, we argued, would not end scarcity. It would transform it: intelligence becomes abundant, while the power, chips and materials behind it become the things nations compete for.
For eighteen days in June, that argument became literal: access to one of the world’s most capable AI systems was effectively export-controlled. For emerging market investors, the implication is clear: AI is changing the basis of competitive advantage, shifting it towards countries and companies that control scarce inputs, infrastructure and industrial capabilities, and away from those whose edge has depended primarily on exporting low-cost labour. This follow-up examines what that shift means for emerging markets: who owns the inputs required to manufacture intelligence, and who exports the work AI is learning to perform.
Eighteen days in June
On 12 June, a US government export-control directive required Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals, including the company’s own employees. Because Anthropic could not verify nationality in real time, it suspended both models for all users. Eighteen days later the controls were lifted: Fable 5 returned globally, while Mythos 5 initially returned only to approved US organisations. These actions had little precedent in cloud software but were entirely familiar in the worlds of uranhium enrichment, encryption and advanced weaponry.
A product that looked like cloud software was handled like a geostrategic material. That is the regime change we described in our first paper.
In the Monopoly world, technology was a product, sold to anyone willing to pay. In the Age of Empires world, decisive technologies become instruments of leverage – rationed, licensed, traded for allegiance – and access to the frontier is secured with power, semiconductors, capital and, critically, political alignment. Semiconductors crossed the line in 2022; in June, frontier intelligence appeared to cross it too. For governments relying on frontier models they cannot operate independently, June made the dependency explicit: access can be conditioned, restricted or withdrawn.
Three things have changed
Our first paper argued that technology would accelerate the new regime. The June episode shows how AI is doing so in three ways.
1. Usable intelligence has become a manufactured good
Traditional software is written once and replicated at near-zero marginal cost. Intelligence is different. It must be produced continually from energy, compute, data and capital – and its critical inputs increasingly sit in the physical world.
This is not the economics of software. AI may feel asset-light to the user, but it is asset-heavy for the system – an industry governed by capacity cycles, cost curves, supply-chain security and state involvement.
The interfaces may be branded by American platforms, but much of the system that manufactures intelligence runs through Asia: logic in Taiwan, memory in Korea, substrates, packaging and equipment across Taiwan and Japan, and power and thermal systems across the region. In the Monopoly era, the world’s most profitable companies converted software scale into cash flow. In the Age of Empires era, those cash flows are being redirected into Asian industrial capacity to secure intelligence’s scarce inputs.
2. Moats are getting thinner, shorter and more physical.
When intelligence gets cheaper, advantages built mainly on it become easier to reproduce. Model leads can be copied, distilled or undercut. What machines cannot easily replicate rises in relative value: power contracts, chokepoint hardware, embedded customer relationships and even the licence to operate at all. As June showed, a more durable moat than an algorithmic lead is permissioned access to infrastructure, customers and technology – access that Washington and Beijing can grant, restrict or withdraw.
3. The race is increasingly decided in permissions and gigawatts.
Every contender permitted by export controls is competing for the same scarce stack of chips, models, grid capacity and energy. China added more than 500GW of generating capacity in 2025, while more than 2,000GW of prospective generation and storage sit in US interconnection queues. Not all gigawatts are equal, but the strategic contrast is clear: sovereign AI requires sovereign infrastructure.
Bubble or not?
That physical build-out brings us to the question most client conversations begin with: “Is AI a bubble?” The scale of investment has made it impossible to ignore. In emerging markets, three semiconductor manufacturers – TSMC, SK Hynix and Samsung Electronics – had reached roughly 30% of the MSCI EM index by end-June 2026, approaching the Magnificent Seven’s share of the S&P 500. To question whether AI is a bubble is increasingly to ask whether global equities are one.
The history of infrastructure booms is a warning. From canals and railways to electricity and fibre, they have often ended in overbuild, shifting assets from financiers to operators before the real deployment began. AI may not be exempt. But this cycle has two unusual features. First, demand is recursive: AI is already being used to write code, design chips and improve the systems that follow it. That adds additional layers of incremental demand for the infrastructure producing it. Intelligence is becoming a metered input into the economy, produced continuously and billed by usage, more like electricity than software. Second, sovereigns are now in the order book, investing not only for returns but against dependency on other nations.
The cycle is also broader, and potentially earlier, than the market shorthand suggests. What began with chips is spreading through data centres, cooling, networking, power and financing. In its June 2026 midyear outlook, JPMorgan estimated that global AI-related capital expenditure could reach $5.5 trillion through 2030. This is becoming a global infrastructure build-out, not simply a technology product cycle.
None of that guarantees returns. The pace can slow, parts of the value chain can over-earn, and financial disappointment need not stop deployment. We doubt the building stops. The railway crash still left behind the railways. Our answer is to be selective, not to exit – and recent price action has made being selective more attractive. Some equity prices now imply a build-out that ends early, while the constraints that govern it are still measured in years: grid queues, transformer lead times, fab qualification, mine permitting. Where demand and bottlenecks remain intact, lower prices improve prospective returns.
The investment question is not simply whether returns justify the capital now being committed. It is where those returns accrue and what kind of economy the build-out creates: where does AI create abundance, where does it preserve or intensify scarcity, and who controls the bottlenecks between the two? For emerging markets, that distinction matters.
What this means for emerging markets
The AI era does not eliminate traditional advantages; it changes their relative value. For economies that import the hardware and energy required to manufacture intelligence while exporting the work intelligence is learning to perform, the risk is a terms-of-trade shock in the strictest sense: what they sell loses value relative to what they must buy. Increasingly, advantage lies with economies that own, build and finance the systems AI depends on, not those whose edge rests on supplying labour to tasks AI is learning to perform.
The most exposed business models are those built primarily on labour arbitrage. The development ladder that carried Japan, Taiwan, Korea and China toward the technological frontier had a first rung: sell your labour cheaper than the incumbent. That rung does not disappear, but AI weakens its pricing power. It impacts the office first because knowledge work is already digital. The competitor for a Manila call-centre seat or a Bangalore coding contract is no longer only a cheaper worker; it is an AI model whose capability keeps improving while its cost per task falls.
Offshoring was a way to import labour without importing workers. AI extends that logic: skilled work delivered at the price of compute. When Alex Karp, Palantir’s chief executive, argued at Davos in January that AI could reduce the need for large-scale immigration into Western labour markets, he was describing a potential dividend for developed economies. Seen from Manila or Bangalore, the same logic sounds less benign.
The factory rung comes under pressure more slowly, but in the same direction. As automation reduces labour’s share of production costs, wage savings must compensate for distance, inventory and supply-chain risk. That shifts the production calculus toward proximity, resilience, power availability and political alignment. A company that can automate its back office also has less reason to offshore it.
Taken together, the old labour-arbitrage model that propelled many lower-income economies is becoming less powerful at the margin. The pressure will not first appear as a collapse in trade volumes, but in wages, graduate employment, utilisation and margins. Technology is driving most of these changes; policy is reinforcing the same direction. The political tolerance for large, persistent trade surpluses is narrowing.
AI impact: India as a test case
India’s premium valuation rests not only on demographics and domestic demand, but on a three-decade compounding machine in white-collar services employment. The listed IT companies may not feel the pressure first. They can use AI to automate delivery, migrate client workflows and emerge leaner and more profitable.
Their problem is growth. Indian IT compounded for decades by adding billable people. It must now grow while using fewer of them. The market has recognised part of this through a multiple derating. The harder question is terminal value. Cheap against history is not necessarily cheap against the future.
The more consequential pressure point may be graduate intake. AI tends to automate junior work first because those tasks are structured, supervised and easier to verify. Firms seeking to protect margins do not need mass redundancies to reshape the labour market; they can simply hire fewer people at the bottom of the pyramid. Companies have that option. Countries do not. For two decades, the IT sector provided one of the developing world’s largest white-collar escalators, carrying Indian graduates into urban incomes, mortgages and monthly investment plans.
That ladder is not only shrinking; it is relocating. Global capability centres (GCCs) are bringing higher-value product, engineering and AI work to India, but hiring more selectively than the old services model. India may retain technological value without recreating the broad employment engine that supported its domestic economy.
This is why the underweight is not an IT-sector call. The five largest listed Indian IT-services companies have already been repriced and, as at the end of June 2026, represented roughly 0.75% of the EM index. Part of India’s broader valuation premium, however, was a consumption-and-credit premium underwritten by remarkable white-collar job creation. The more important question for investors is whether the downstream effects on employment, consumption and credit have been repriced to the same degree.
India’s strengths are real. The question is whether they justified the premium the market placed on them. We became more demanding about the price we are willing to pay.
What would change our view? A sustained re-acceleration in both white-collar hiring and revenue growth across IT-services vendors and GCCs. That combination would indicate that AI is expanding India’s delivery capacity, rather than simply reducing the number of workers required to generate each unit of revenue.
Elsewhere, the picture is different. Southeast Asia’s assembly and manufacturing economies sit more on the factory than office rung, so pressure on wage arbitrage should arrive more slowly. But as labour becomes a smaller share of production costs, proximity, automation, energy availability and supply-chain resilience matter more. Other emerging markets, including parts of Latin America, sit outside both the main path of disruption and the core AI build-out. For investors, their attraction lies elsewhere: domestic franchises, resources, financial deepening, infrastructure, tourism – and valuation. Greece is a useful example: recapitalised banks offer exposure to improving domestic balance sheets, while tourism derives its value from physical experiences that AI cannot replicate. That distance from the AI cycle may itself make some of these markets useful diversifiers.
The map is changing, not disappearing
The AI model winners remain uncertain. The physical requirements of intelligence are more durable. As intelligence becomes abundant, value migrates towards the scarce systems that produce, move and sustain it: leading-edge manufacturing, memory, packaging, power equipment, grid capacity, critical materials and trusted infrastructure.
That leaves emerging markets with a larger role, not a smaller one. East Asia controls critical layers of the AI hardware stack. China remains deeply embedded in manufacturing, electrification and power equipment. Resource-rich economies supply the metals and energy required by a more electrified, capital-intensive world. Other manufacturing hubs can gain as supply chains move closer to customers, reliable power and political alignment.
The counterpoint to labour-arbitrage risk is therefore straightforward. Emerging markets do not stop mattering; the basis of advantage changes. The old winners often supplied labour more cheaply. The new winners are more likely to supply something difficult to replicate.
Scarcity, however, is not permanent. Today’s shortage funds tomorrow’s capacity. The opportunity is not to own every AI beneficiary, but to identify where scarcity will endure longer than the market expects because supply remains constrained by technical complexity, qualification, infrastructure lead times, resource quality or political permission. For investors, the question is not only what is scarce, but how long it will remain scarce, who controls it and whether its value will reach shareholders.
From pieces on the board to players in the game
Across our portfolios, the same variable keeps reappearing: scarcity. The countries and companies that matter more are defined less by the price of their labour than by the sophistication of what they can make, build, operate and finance, and by how difficult those capabilities are to replicate.
The question is simple: what scarce product, service or capability can you provide that is difficult to replicate?
In the Monopoly world, an emerging market could prosper as a low-cost piece on someone else’s board: a cheaper place for a factory, call centre or service hub. In the Age of Empires world, that is no longer enough. Success belongs increasingly to the players: economies with something difficult to replace, whether energy, resources, industrial capability, technological depth, financial capacity, trusted infrastructure or physical experiences machines cannot replicate.
AI does not make emerging markets irrelevant. It changes the terms on which they win, and therefore the terms on which we invest.
