Scarcity has shifted, not disappeared: The premise of AI abundance needs sharper evidence

A RECENT commentary in these pages (“The coming AI-driven ‘abundance’ shock”, Apr 14, 2026) argued that artificial intelligence is quietly rendering scarcity obsolete, and that Singapore must redesign its social architecture for a world of boundless machine-generated plenty.

The argument rests on a seductive premise: that AI allows knowledge, analysis and creativity to be replicated at near-zero marginal cost. From this, the authors raise far-reaching concerns about identity crises, cognitive instability and the erosion of democratic deliberation.

These are important questions. But the premise on which their arguments rests deserves closer scrutiny, and the policy conclusions that follow would benefit from sharper empirical grounding.

AI is expensive, and getting more so

The single most important empirical fact about AI in 2026 is that it consumes extraordinary real resources. By late 2025, AI data centres alone drew roughly 29.6 gigawatts of power globally, equivalent to the peak electricity demand of New York State, the AI Index Report of Stanford University finds.

Since 2020, US residential electricity prices have risen by more than 36 per cent, as at February 2026, according to US Energy Information Administration data.

As reported by the Associated Press, Maine is set to become the first US state to pass a moratorium on large data centre construction, banning facilities above 20 megawatts until November 2027, with similar Bills introduced in at least 10 other states.

Algorithmic capability may scale freely, but energy, semiconductor fabrication capacity, water for cooling and talent do not.

Speaking of “abundance” while the binding constraints on AI expansion remain risks overlooking a fundamental distinction: The nature of the product is not the same as the cost of its production.

Scarcity has shifted from one domain to another. And the cost of the output is changing, too: In April 2026, Anthropic ended Claude subscription access for third-party agent frameworks, with the US$200 monthly plan previously covering up to US$5,000 of compute. This is an example of input scarcity propagating directly to output prices, countering the marginal cost argument.

The productivity evidence is modest

The article implies that AI is already fracturing the link between human effort and economic value. The empirical evidence tells a more cautious story.

Nobel laureate Daron Acemoglu’s task-based model estimates that AI will contribute less than 0.53 per cent to total factor productivity cumulatively over the coming decade – which makes the premise of imminent abundance harder to sustain.

These gains are real, but they are concentrated in narrowly defined task categories and strongest among lower-performing workers.

A recent Massachusetts Institute of Technology study mapping 13,275 AI applications against the full ontology of US work activities found that the top 1.6 per cent of activities account for over 60 per cent of all AI market value, confirming that current AI capabilities cluster in a narrow band of tasks.

A controlled study by Model Evaluation & Threat Research of 16 experienced open-source developers found that AI tools actually slowed them down by 19 per cent, despite the developers themselves believing they had been sped up by up to 24 per cent.

The evidence does not yet support the claim that AI has decoupled human effort from economic value, the premise on which the distributional and institutional concerns rest.

Furthermore, lab director Erik Brynjolfsson and colleagues from the Stanford Digital Economy Lab find that aggregate consumer surplus from generative AI already substantially exceeds producer revenues, suggesting that currently the value is accruing primarily to users rather than concentrating among capital owners, complicating the inequality narrative that underpins the abundance thesis.

Singapore has already moved

The call to rethink Singapore’s social architecture is well-intentioned, but it understates the extent to which the nation has already acted.

Budget 2026 established a National AI Council chaired by the prime minister. In January 2026 the country released the world’s first governance framework for agentic AI. These are functioning instruments.

Singapore’s National AI Strategy is framed around “AI for the Public Good”, signalling that the agenda extends beyond economic competitiveness to distribution, social cohesion and values.

Abundance makes the human matter more

One prevailing concern is that AI abundance will erode the link between work and identity.

Alex Imas, professor of behavioural science, economics and applied AI at the University of Chicago, counters that as commodity production cheapens, spending shifts towards relational goods and services where human involvement is the source of value, not a cost to be eliminated.

Starbucks illustrates the point: After years of automating its stores to improve margins, the company reversed course in 2025, hiring more baristas and reintroducing handwritten cups and ceramic mugs after concluding that human hospitality, not efficiency, was what drove customer satisfaction.

If this pattern holds, AI reinforces the meaning-giving function of work.

Broad diagnoses about identity, cohesion and foundations of knowledge may point to potentially real problems, but they do not yet narrow the policy choices. Without sharper questions, even a well-resourced government cannot know which instruments to build.

Sharper questions

Policy questions about AI need to be specific and tractable.

How should electricity pricing reflect the externalities of compute-intensive industries?

What measurable indicators should governments track to detect whether AI adoption is eroding workforce well-being, professional identity, or social cohesion before those effects become entrenched?

And, what safeguards are needed to ensure that the shift to AI-mediated public services does not exclude or disadvantage elderly citizens who lack digital fluency?

The comparative advantage of Singapore in the AI era will continue to rest on the quality of its institutions, the pragmatism of its policy design, the curiosity of its people, and the speed of its execution.

The country does not need to prepare for an abundance that does not exist yet. It needs to keep investing in infrastructure, building workforce capabilities, helping citizens use AI, and governing technology with compassion.

The writer is the founder of Singapore-based Lightbulb Capital, an innovation advisory firm focused on financial services, and a lecturer at Essec Business School’s Master in Finance programme