Who gets to use the future

Within a few decades of Gutenberg’s work in Mainz, printing presses had spread across Europe. The technology could reproduce knowledge cheaply; societies still needed readers, schools, publishers and administrations capable of turning cheaper information into something economically useful.

Technology travelled faster than the capacity to absorb it. Artificial intelligence may be doing the same. New data from Anthropic, an American AI company, show how unevenly its Claude model is being used relative to population. The figures are partial: they describe one provider, not artificial intelligence as a whole.

Anthropic’s latest figures suggest that AI use is not converging evenly. The 20 countries with the highest per-capita Claude use accounted for 48% of population-adjusted global usage in February 2026, up from 45% in the previous report. 

Within America, by contrast, the five highest-usage states’ share of per-person use fell from 30% to 24% between August 2025 and February 2026. AI diffusion, in other words, can spread within one economy while becoming more concentrated across countries. That suggests a useful distinction.

There is an access gap, an adoption gap and an absorption gap. The first concerns who can obtain the technology. The second concerns who actually uses it. The third concerns who can reorganize work, capital and institutions around it well enough to generate lasting economic gains. The third may matter most.

Economists have seen versions of this before. Electricity, computing and other general-purpose technologies produced their largest gains only alongside complementary capital: new skills, infrastructure, processes and forms of organization.

Electricity did not transform manufacturing merely because factories replaced steam engines with electric motors. The larger gains came when factories themselves were redesigned around distributed power.

Computing followed a similar path: firms bought machines first and spent years changing software, skills and workflows around them. The machine, in other words, is only part of the capital stock. The rest is institutional.

That matters because institutions adjust more slowly than technologies improve. A new model can be deployed quickly. Reporting lines, incentives, procurement rules, skills and habits may take years to change. Call it the absorption gap.

Economic history offers plenty of examples. Imperial China developed important technologies long before industrialization in north-western Europe, yet technological precedence did not automatically produce an industrial revolution. 

Japan after the Meiji Restoration offers the opposite case. It did not invent the technologies powering nineteenth-century industrialization; it built the educational, administrative and physical infrastructure needed to absorb them.

The lesson is not that institutions matter more than technology. It is that technology rarely generates its largest gains by itself.

That distinction is especially useful for AI because capable models are unusually portable. For countries and firms able to obtain them, access may become progressively less informative as a measure of advantage. 

The harder things to reproduce are the complements: reliable data, skilled workers, managerial competence, redesigned workflows and systems of accountability. The same logic applies inside companies.

Giving employees access to a chatbot is adoption. Redesigning work so that AI performs part of the analysis, people verify what matters, information moves differently through the organization, and responsibility remains clear is absorption.

The first changes the tool. The second changes the firm. Two companies may therefore use the same model and obtain radically different results. One may bolt it onto existing workflows and gain a little efficiency. 

Another may redesign entire processes around it, change the division of labor between people and machines and shorten decision cycles. The technology is identical. The organization is not.

AI therefore creates an unusual economic race. The supply side may move at software speed; the demand side still moves at institutional speed.

Today’s usage statistics show where experimentation is taking place. They tell us much less about who will capture the largest gains. Those gains will depend on which firms redesign themselves, which workers acquire new skills, and which institutions learn to change before the technology changes again.

Capable models may not be scarce for long. Absorptive institutions could be.

Ilija Tozija is Vice President of Technology & Innovation at The Group Hospitality, a New York City hospitality group, where he leads technology and AI transformation. He has more than 15 years of experience in governance and risk across large capital programmes — over $1 billion in aggregate value spanning energy, technology, and infrastructure, across the EU, UK, US, and Canada, including major capital works with Mace Group. 

He has advised public and private stakeholders on IFI-financed projects. He studied at the University of Oxford and the Vienna University of Economics and Business.

Disclaimer: The opinions expressed by the author are his own and do not in any way reflect those of Paraluman News.