What is the economic cost of not using AI?
Every significant technology wave in economic history has produced two groups of people: early adopters who captured the productivity gains, and late movers who competed against those gains for years after. The divide between the two groups is usually measured in hindsight, and in the case of AI it is already being measured. The cost of not adopting AI is leading to structural disadvantage between those who adopt and those who take the backseat. The earliest estimates suggest that such cost runs into trillions of dollars at the national level and into hefty margin compression at the firms level.
## The historical precedent
When electrification arrived in American factories, adoption was neither instant nor uniform. Economic historians have documented how firms that converted from steam power to electric motors in the 1890s often saw only modest productivity improvements at first. The reason was that they retained factory layouts designed around a central steam shaft. The larger gains emerged over the following decades, particularly during the 1910s and 1920s, as factories were redesigned around the flexibility of electric power. Firms that postponed those organisational changes fell further behind, competing against businesses whose operations had already been rebuilt around a more productive model.
Paul David, the Stanford economic historian, made this case in his 1990 paper on the dynamo and the computer. His argument was that general-purpose technologies take time to diffuse, and that the diffusion gap itself creates lasting productivity divergence between early and late adopters. AI fits the general-purpose technology template well: it is applicable across sectors, it improves over time, and its gains compound through complementary investments in workflow, training, and data infrastructure.
The other obvious comparison is the internet. Firms that built e-commerce infrastructure in the mid-1990s, when it was operationally difficult and strategically uncertain, entered the 2000s with customer datasets, logistics systems, and institutional knowledge that newer entrants spent years trying to replicate. Amazon's dominance in 2025 is partly a function of decisions made in 1995. The lesson is not that first movers always win. The lesson is that first movers in general-purpose technologies often build structural advantages that persist for much longer than the initial technology gap.
## What the numbers say at the national level
The IMF's January 2024 analysis estimated that AI will affect 40 percent of jobs globally and 60 percent of jobs in advanced economies, with the effects running to both displacement and augmentation. The analytical implication for national competitiveness is direct: countries that build the workforce infrastructure and regulatory conditions for rapid AI adoption will capture the augmentation gains. Countries that move slowly will absorb the displacement effects without the corresponding productivity uplift.
The Stanford AI Index 2025 documents that private investment in AI reached $252 billion globally in 2024, with the United States accounting for the majority. That investment concentration signals where productive capacity is being built. Nations that are not attracting or generating comparable investment are not keeping pace in the infrastructure layer that determines which firms and workers will be most productive five years from now.
McKinsey's modeling has estimated that generative AI could add between $2.6 trillion and $4.4 trillion annually to the global economy across use cases. McKinsey's 2025 State of AI found that 71 percent of organizations now regularly use generative AI, up from 33 percent two years prior. The firms in that 71 percent are building the institutional knowledge, the workflow integrations, and the data infrastructure that makes subsequent AI deployment faster and more effective. The remaining 29 percent are not catching up to a target in waiting. They are chasing organizations whose capabilities continue to compound with every deployment.
## What the numbers say at the firm level
The OECD's report on the adoption of artificial intelligence in firms found significant variation in adoption rates across firm size, sector, and country. The firms that adopt AI tend to be larger, more productive to begin with, and more likely to invest in complementary skills. This creates a reinforcing dynamic, or, call it a paradox if you wish: AI adoption accelerates the productivity of firms already running ahead, while firms running behind face the compounding disadvantage of both lower baseline productivity and slower adoption.
There is a sector dimension here that deserves direct attention. McKinsey's research has consistently found the highest near-term AI value concentrations in software engineering, customer operations, and marketing. A firm in any of these sectors that has not yet deployed AI in these functions is already competing against firms that have been running AI-assisted operations long enough to have completed a first cycle of iteration and improvement.
## The turn in this argument
This is where the optimistic version of the AI productivity story encounters its most serious complication. McKinsey's March 2025 State of AI reported that while 71 percent of organizations use generative AI, more than 80 percent report no measurable EBIT impact. At face value, this looks like evidence that the cost of non-adoption is overstated. Read carefully, it reveals something different.
The firms reporting no EBIT impact are, in many cases, using AI at the margin: automating a task here, drafting a document there, running a chatbot that handles the easy customer queries. They have adopted the tool without redesigning the work. The productivity gains from electricity did not arrive when factories installed electric lights. They arrived when the factory floor was rebuilt around what electricity actually made possible. The firms that will pay the real cost of non-adoption are not only those that skipped AI entirely. They also include firms that adopted AI without changing the workflows around it, and whose competitors figured that redesign out first.
So in the end, the current EBIT question does not weaken the economic case for adoption. It simply further explains it. The firms that will capture the gains are those that adopt and redesign simultaneously. The firms that will suffer the structural disadvantage are those that do neither. Adoption without redesign is a cost with a deferred return but non-adoption is a cost with no return at all.
## For governments, this means
The economic cost of delayed AI adoption shows up in total factor productivity, in the competitive position of domestically headquartered firms in global markets, and in wage growth for workers in sectors where AI augmentation would otherwise increase output per hour. The OECD's AI adoption research points to three levers that correlate with faster and broader adoption: digital infrastructure investment, workforce training at scale, and regulatory conditions that permit experimentation without creating legal uncertainty for deploying firms.
That last point matters in the European context. Regulatory clarity accelerates adoption by reducing the legal risk premium that firms otherwise factor into AI investment decisions. Regulatory ambiguity does the opposite. For European companies currently assessing their AI obligations, understanding what compliance requires before key regulatory deadlines is on similar footing with the adoption decision itself.
The economic cost of not using AI is not a warning about the future. It is an account of what is already accruing. Countries and firms that adopt thoughtfully, redesign deliberately, and invest in the workforce and data infrastructure that makes adoption compound will separate from those that do not. The separation has already begun, and history suggests it will widen long before it stabilises.