Brooker Worldview — Signature Thesis #2

Every economic era is defined by what is scarce.

Land was scarce, so agrarian wealth meant acreage. Capital was scarce, so industrial wealth meant factories, machines, banks, and finance. Oil was scarce, so the twentieth century revolved around energy security. Information was scarce, so the late twentieth century rewarded those who controlled data, media, research, and distribution.

For all of human history, one input has been scarce without exception: intelligence.

Skilled analysis, expert judgment, written work, legal drafting, software, design, research, translation, tutoring, diagnosis, forecasting, and business synthesis all required a trained human mind. Trained human minds were expensive, slow to produce, unevenly distributed, and limited in number. Entire industries, salary structures, consulting models, national strategies, and corporate hierarchies were built on that scarcity.

That scarcity is now ending.

Artificial intelligence is making cognitive output — analysis, code, content, research synthesis, professional drafting, first-pass legal review, financial modeling, language translation, software testing, customer support, market monitoring — abundant and nearly free at the margin. This is the single most important economic shift of our lifetimes, and most people are still analyzing it at the wrong level.

The mainstream conversation treats AI as a productivity tool: chatbots, copilots, faster emails, automated summaries. That framing is too small. AI is not merely improving office work. AI redraws the scarcity map of the economy. And in economics, value does not live in what becomes abundant. Value migrates to what remains scarce.

So the essential question of the next decade is not “what can AI do?” That question changes every month and produces shallow thinking. The sharper question is:

When intelligence becomes cheap, where does scarcity move?

AI is not software. It is infrastructure.

Begin with a correction to the popular framing. Software historically became one of the best businesses in the world because the marginal cost of serving one more user was close to zero. Write the code once, distribute it globally, collect recurring revenue, enjoy high gross margins.

AI breaks that model.

Every AI query consumes real compute. Real compute consumes real chips. Those chips sit inside real data centers. Data centers draw real electricity, require cooling, connect through fiber, depend on transformers, substations, turbines, semiconductors, land, water, and permitting. Intelligence has become an industrial product. It is manufactured.

That is the key mental shift. AI is not just software eating the world. AI is electricity, semiconductors, data centers, networks, cooling systems, and model architectures manufacturing intelligence at scale.

The International Energy Agency projects global data-center electricity consumption to roughly double by 2030 to around 945 TWh, with data-center power demand growing at roughly 15% per year from 2024 to 2030 — more than four times faster than other electricity demand. AI-specific accelerated servers are projected to grow even faster. Hyperscale AI centers can exceed 100 MW of power demand, equivalent to the annual electricity use of roughly 100,000 households.

That is why the capital expenditure cycle now underway is not a normal software cycle. It is an infrastructure cycle. Hyperscalers are spending like governments. Governments are treating compute as a sovereign asset. Chip access has become foreign policy. Power purchase agreements have become strategic weapons. Data center location has become a geopolitical and local political issue. Grid capacity has become a technology bottleneck.

Data centers are the new factories. That is not a metaphor. They take raw inputs — energy, data, chips, and capital — and manufacture the most economically valuable product in the world: intelligence.

The new scarcity map

When intelligence becomes abundant, scarcity migrates to the layers that intelligence depends on and the layers intelligence cannot easily replicate. We see seven destinations.

1. Compute. The chips and data-center capacity that manufacture intelligence are supply-constrained, capital intensive, geopolitically contested, and years away from equilibrium. Whoever controls compute controls the means of cognitive production.

2. Energy. Every serious AI forecast is also an electricity forecast. Energy is the ceiling on intelligence. Grid capacity, generation, nuclear revival, gas turbines, transformers, batteries, data-center power contracts, and energy trading all become technology investments.

3. Data. Models are becoming more widely available. Proprietary, high-quality, rights-cleared, domain-specific data is not. Unique data assets — clinical records, transaction flows, industrial telemetry, local-language corpora, customer behavior, regulated datasets — become compounding moats.

4. Trust and verification. When AI can generate infinite convincing content, the scarce commodity becomes proof: proof of authorship, proof of authenticity, proof of humanity, proof of provenance, proof of ownership, proof of transaction history. Cryptographic verification moves from niche to necessity.

5. Distribution. When anyone can generate content, software, analysis, and marketing collateral, production loses value and attention becomes the bottleneck. The ability to reach an audience, customer base, investor network, or community becomes more valuable, not less.

6. Liquidity. Capital that can move at the speed of AI-driven markets — programmable, always-on, globally settled — commands a premium over capital trapped inside banking hours, settlement delays, local rails, and manual processes.

7. Physical embodiment. Robotics — intelligence with hands — remains scarce, hard, and early. The gap between what AI can decide and what machines can physically do is one of the largest investment frontiers of the decade.

This is the new scarcity map. The discipline for investors is simple to state and hard to practice: stop pricing the layer becoming abundant and start pricing the bottlenecks it depends on.

The most dangerous place to invest blindly is the layer that looks magical but is becoming commoditized. Generic content, generic software, generic analysis, generic copilots, generic automation tools, and generic wrappers around foundation models will face margin compression. Scarce-layer assets — compute, energy, data, verification, distribution, liquidity, robotics — sit on the right side of the value migration.

Why “AI productivity” is too narrow

Most corporate AI conversations begin with productivity: how many hours can we save, how many documents can we draft, how many customer emails can we automate? That is useful, but incomplete.

The deeper transformation is not productivity inside the old organization. It is the redesign of the organization around machine intelligence.

A company that uses AI to write faster emails has not changed its operating model. A company that deploys AI agents to monitor markets, draft reports, run diligence checklists, update CRM records, summarize legal documents, reconcile data, prepare board packs, and escalate exceptions has changed its workflow architecture. The first company buys tools. The second company manufactures intelligence internally.

This distinction matters because the bottleneck shifts from access to AI tools to the ability to design workflows that AI can execute safely. The scarce skill becomes not “prompting” in the superficial sense, but process engineering: breaking work into repeatable decisions, data inputs, escalation points, quality checks, and governance rules.

In the old economy, headcount was capability. In the AI economy, workflow architecture is capability.

A ten-person firm with well-designed agents, clean data, clear SOPs, and strong governance may outperform a hundred-person firm where knowledge sits in scattered inboxes, undocumented habits, and individual memory.

Robots and AI agents are the new demographics

Now connect this to the most underestimated macro force in the world: demographics.

Nearly every major economy — Japan, Korea, China, Europe, and increasingly Thailand — faces the same arithmetic. Working-age populations are shrinking or stagnating. GDP growth is, at its core, workers multiplied by productivity. When the worker count falls, countries face a brutal choice: accept lower growth, increase immigration, raise participation, borrow more, or deploy new forms of labor.

For decades, governments filled the gap with debt. Borrowing sustained demand that demographics could no longer produce organically. But debt is a painkiller, not a cure. The only durable cure for a shrinking workforce is a new source of workers or a large jump in productivity.

AI agents and robots are that source.

An AI agent that performs the workflow of an analyst, bookkeeper, customer service representative, compliance officer, research assistant, or software tester is a new worker added to the economy. It never ages, scales instantly, operates across languages, improves annually, and can be copied across workflows. A humanoid or industrial robot is the same logic applied to physical labor.

This is why demographics are no longer destiny. Deployment is.

Aging societies that aggressively deploy machine intelligence can sustain or even accelerate growth despite shrinking populations. Aging societies that do not will discover that their demographic curve is also their GDP curve.

For Thailand, this is not a technology question. It is a national economic survival question. The World Bank has projected that Thailand’s share of population aged 65 and over could rise from roughly 13% to 31% by 2060, while the working-age share declines from about 71% in 2020 to 56% in 2060. That implies an economy that must produce more output with fewer workers, fewer taxpayers, more elderly dependents, and more fiscal pressure.

Thailand cannot solve that arithmetic through slogans. It must solve it through productivity, automation, capital formation, and AI deployment.

The counterargument: AI may destroy jobs and weaken institutions

A serious thesis must face the strongest objection. The objection to abundant intelligence is not that AI is weak. The objection is that AI may be too strong in the wrong places.

If AI automates entry-level work, how do young professionals learn? If AI reduces the need for junior analysts, who becomes the next senior analyst? If AI floods the world with synthetic content, how does trust survive? If every company can produce infinite marketing, how does anyone know what is real? If software can be generated cheaply, what happens to SaaS margins? If AI systems require constant supervision, do companies merely shift labor from doing work to “botsitting” work?

These objections are valid. They do not refute the thesis. They sharpen it.

Abundant intelligence does not automatically create prosperity. It creates leverage for those with the architecture to use it and disorder for those without it. Companies that dump AI tools onto messy processes may get noise, risk, hallucination, and employee frustration. Companies that redesign workflows, governance, and data around AI can get speed, scale, and structural cost advantage.

The same is true for countries. AI deployment without education, reskilling, energy strategy, and institutional trust can deepen inequality. AI deployment with deliberate architecture can offset demographic decline and expand productive capacity.

So the right conclusion is not “AI replaces humans.” It is more precise: AI replaces poorly structured cognitive work, amplifies well-structured human judgment, and punishes institutions that cannot redesign themselves.

What this means in practice

For corporates, AI adoption is not an IT project. It is workforce strategy and operating-model redesign. The right unit of analysis is not “which tools should we buy?” but “which workflows can be performed by digital workers, what data do they require, what authority do they have, and how do humans supervise exceptions?”

For investors, the scarcity map should sit beside every allocation decision. Abundant-layer businesses may look impressive today but face future margin collapse if they lack data, distribution, trust, or workflow lock-in. Scarce-layer businesses — compute, power, chips, proprietary data, verification rails, robotics, and distribution networks — deserve a different analytical lens.

For policymakers, the implication is national. AI readiness is not a ministry-level digital program. It is energy policy, education policy, industrial policy, capital-market policy, and demographic policy fused together.

The Brooker position

Our investment worldview is also shaped by the scarcity map. We are less interested in undifferentiated AI wrappers and more interested in the layers that remain scarce: compute, infrastructure, robotics, verification, programmable finance, data, and distribution. That is where pricing power migrates when intelligence becomes abundant.

Intelligence was the last great scarcity. Its abundance does not end the game. It moves the board.

The winners of the next decade will be those who looked up from the dazzling abundance of machine intelligence and asked the older, sharper question:

What is scarce now?

This is the second essay in the six-part Brooker Worldview 2026 series. The next essay turns to money itself — and why the loudest macro narrative in the world, de-dollarization, is almost exactly backwards.