By Naina, 22nd May 2026

For the first time in nearly two decades, the question dominating finance ministries, central banks and corporate boardrooms is not whether artificial intelligence will alter the trajectory of global economic growth. It is by how much, for whom, and at what cost. The phase of speculation is over. AI and automation have moved from the margins of macroeconomic discussion to its centre, and the numbers arriving on policymakers' desks this year carry a weight that earlier cycles of technological enthusiasm never quite produced.

The International Monetary Fund's January 2026 World Economic Outlook update lifted its global growth forecast to 3.3 percent for the year and 3.2 percent for 2027, a modest but pointed upward revision over its October 2025 projections. The Fund attributed the resilience to a combination of accommodative financial conditions, monetary easing and, most significantly, a surge in AI-related investment that has pushed information-technology spending in the United States to its highest share of capital formation in more than two decades. Strip the AI build-out out of the American GDP figure for the third quarter of 2025, and growth would have been markedly weaker. That single observation, made by the IMF and corroborated by the US Bureau of Economic Analysis, reframes the entire conversation. Artificial intelligence is no longer a sectoral story. It has become the marginal driver of headline growth in the world's largest economy.

A Capital Cycle Without Modern Parallel

The most visible manifestation of AI's economic footprint is the scale of capital expenditure now being committed by a small cluster of hyperscale technology companies. Amazon, Microsoft, Alphabet, Meta and Oracle are collectively projected to spend somewhere between 600 and 720 billion US dollars on capital projects in 2026, up roughly 36 percent over 2025 and more than triple their combined outlay in 2024. Approximately 450 billion dollars of that 2026 figure, by independent estimates, is destined for AI-specific infrastructure: graphics processing units, data centres, networking gear, cooling systems and the power-generation assets needed to feed them. Goldman Sachs Research now projects cumulative AI-related capital expenditure of roughly 7.6 trillion dollars between 2026 and 2031. McKinsey's earlier modelling suggested the figure could climb to 6.7 trillion by 2030 just to keep pace with compute demand.

These numbers carry macroeconomic weight. Big technology capex now exceeds two percent of US GDP, a share that begins to rival the largest infrastructure programmes of the twentieth century. The hyperscalers are committing nearly 90 percent of their operating cash flow to capital spending in 2026, against roughly 65 percent in 2025. Debt issuance by the same cohort has already surpassed 100 billion dollars in the first quarter alone, exceeding their entire 2025 bond programme. Capital intensity, measured as capex divided by revenue, has climbed to between 45 and 57 percent across the leading firms — a level traditional industry analysts would have once associated with utilities or heavy manufacturing, not software businesses.

What this means in plain language is that the AI build-out is functioning as a fiscal stimulus in private clothing. Power equipment makers, semiconductor fabricators, electrical contractors, transformer manufacturers, real-estate developers and engineering services firms are all riding a wave of demand that did not exist three years ago. Asian technology export sectors, particularly in Taiwan, South Korea and increasingly Malaysia and Vietnam, are absorbing a significant share of the spend. The capex cycle alone is doing work that loose fiscal policy used to do in earlier eras.

The Productivity Question, Still Unresolved

If capital spending is the easy part of the AI story, the productivity dividend is the contested part. The range of professional forecasts is uncomfortably wide. Daron Acemoglu of MIT, working with conservative assumptions about how many tasks can be profitably automated in the near term, estimates that generative AI will add roughly 0.7 percent to US productivity and 1.1 percent to GDP over the next decade. Goldman Sachs places its central estimate at 1.5 percentage points of additional annual productivity growth, with a fifteen percent eventual lift to labour productivity in developed markets once the technology is fully embedded. McKinsey's range runs from 1.5 to 3.4 percentage points of additional annual GDP growth in advanced economies. The Penn Wharton Budget Model projects a more measured 1.5 percent boost to total factor productivity by 2035, rising to 3.7 percent by 2075.

Goldman Sachs equity strategists have, however, introduced a sobering caveat. As of early 2026, the bank's research team cannot find a meaningful statistical relationship between AI adoption and economy-wide productivity. The aggregate data does not yet show what the forecasts insist must be coming. At firm level, the picture is dramatically different: companies that have measured AI-driven productivity gains on specific tasks report a median improvement of around thirty percent in narrowly defined use cases. This gap — between robust microeconomic gains and an absent macroeconomic signal — echoes the famous Solow paradox of the late 1980s, when computers were said to be visible everywhere except in the productivity statistics.

The likely explanation is that diffusion takes time, and measurement takes longer. GDP accounting captures the capital outlay immediately but lags the productivity spillover by years. The Federal Reserve, the European Central Bank and the Reserve Bank of India are all navigating policy decisions on the basis of data that almost certainly understates underlying productivity growth. If AI is quietly raising potential output, the global economy may be running with more slack than headline indicators suggest, which carries direct implications for the interest rate path and for the inflation outlook over the next two to three years.

Labour Markets Under Quiet Pressure

The labour-market consequences of AI and automation are unfolding in a manner that is gradual on aggregate but sharp at the margin. The International Monetary Fund estimates that between 40 and 60 percent of jobs in advanced economies are exposed to AI in some form, with about half facing displacement risk and the other half augmentation. Goldman Sachs places the share of US work hours exposed at roughly 63 percent, of which 25 to 50 percent are directly automatable. The International Labour Organization's refined global index of occupational exposure has begun providing more granular country-level estimates, suggesting that exposure is highest in clerical, customer-service, accounting, paralegal and entry-level analytical roles.

What the data does not yet show in clean fashion is a wave of measurable displacement. The Yale Budget Lab's ongoing tracker finds no systematic relationship, at present, between exposure measures and changes in employment or unemployment. The signal sitting just beneath the surface, however, is a slowdown in hiring at the bottom of the white-collar ladder. Goldman Sachs's labour economists expect that entry-level workers in their twenties and thirties in knowledge and content sectors will absorb the early impact, and that technology-driven productivity gains typically lift unemployment by about 0.3 percentage points for each percentage point of productivity acceleration during the transition years. Their base case sees the US jobless rate edging up to 4.5 percent this year before fiscal tailwinds and AI-driven output growth begin to stabilise the picture in the second half.

The compositional shift is at least as important as the headline figure. The roles being eliminated and the roles being created are not interchangeable. A laid-off customer-service agent does not seamlessly become a prompt engineer or a machine-learning operations specialist. The retraining gap is the central labour-policy question of the next five years, and it is one that very few governments have begun to address with adequate scale or seriousness.

The Cross-Country Divide

A working paper from the IMF research department, updated in early 2026, finds that AI will widen rather than narrow the gap between advanced and developing economies. The growth impact in advanced economies could be more than double that in low-income countries, driven by differences in sectoral exposure, AI preparedness and access to essential data and chips. The Fund's framework hinges on three variables: how much of a country's economy sits in AI-affected sectors, how ready its institutions and infrastructure are to absorb the technology, and whether it has reliable access to the data, models and compute needed to participate in the upside.

Low-income economies dominated by agriculture, with limited digital infrastructure and restricted access to frontier AI technologies, face the most acute risk of falling further behind. Even with best-case improvements in preparedness, a significant growth gap is likely to persist. The geopolitical implication is uncomfortable: AI may consolidate the lead held by the United States and China, give selective middle-power economies including India, the United Arab Emirates, Saudi Arabia, Singapore and South Korea a meaningful upside, and leave most of sub-Saharan Africa, parts of Latin America and several South Asian neighbours of India struggling to participate at all.

India's Position in the New Cycle

For India, the AI transition arrives at a particularly delicate moment. The IMF continues to project that India will remain the fastest-growing large economy in 2026, even as the country navigates a softer global trade environment and persistent pressure on rural consumption. The India Skills Report 2026 finds that national employability has risen to 56.35 percent, the highest figure on record, and that more than 90 percent of Indian employees have begun working with generative AI tools in some capacity. ServiceNow's modelling, commissioned through Pearson, projects that India will add 33.9 million workers to its labour force by 2028, of which roughly 2.73 million will be in technology-intensive roles created or reshaped by AI.

The other side of the ledger is less reassuring. Bernstein Research, in an open letter to the Prime Minister last month, warned of a deepening quality-of-employment crisis as artificial intelligence reduces the routine coding, testing and customer-support work that has underpinned India's IT services sector for two decades. Volume hiring in the large IT exporters has visibly slowed, even as AI-specialised hiring grows at triple-digit rates. The risk for India is not aggregate job loss in the technology sector, but a hollowing-out of the middle of the skills pyramid: roles that paid well, employed millions and supported spillovers into real estate, education and consumer services across tier-one and tier-two cities.

PwC's earlier estimate that AI could contribute roughly 23.8 percent to India's GDP by 2030 is now widely cited in policy circles. Whether that figure proves directionally correct will depend almost entirely on three variables that lie inside India's control: the speed of digital public infrastructure expansion beyond payments and identity, the success of new skilling programmes in actually moving displaced IT workers into AI-adjacent roles, and the availability of domestic compute capacity at a price that does not force every Indian start-up to rent foreign cloud time. The India AI Mission, sovereign compute initiatives at the state level and the rising private investment in domestic data centres point in the right direction, but the gap between announcement and execution remains the persistent feature of Indian industrial policy.

The Risks Hiding in Plain Sight

For all the optimism embedded in 2026 growth forecasts, three risks deserve more attention than they currently receive. The first is the divergence between AI capital expenditure and AI revenue. Allianz Research notes that the gap between capex growth and sales growth in the AI complex is now running at roughly 46 percent, exceeding the 32 percent divergence observed during the 2001 telecom bubble. Infrastructure is being built in advance of revenue that may or may not arrive on the timetable that valuations assume.

The second is concentration. A handful of hyperscalers account for the bulk of capex, a handful of model developers account for the bulk of frontier capability, and a single chip designer continues to capture an outsized share of the economic rent. Any disruption to that narrow tier — regulatory, geopolitical or financial — would transmit immediately into the broader economy. The IMF, in its January update, flagged that increasing reliance on debt finance to fund the build-out has made the system more sensitive to a tightening of financial conditions.

The third is distributional. The IMF's December 2025 scenario-planning workshop concluded that, in the baseline case, capital owners capture a disproportionate share of AI's gains while labour absorbs the disruption. Wealth inequality widens. Political support for the transition weakens. Without active redistribution through tax policy, public investment in retraining, and stronger competition policy, the political backlash against AI could ultimately constrain the productivity dividend itself.

The Policy Imperative

The policy response globally has been uneven. The United States has chosen to lead through private capital and permissive regulation. The European Union has chosen to lead through rules. China has chosen to lead through state-directed industrial policy and aggressive domestic compute build-out. India sits between these models, with strong digital public infrastructure, growing private investment, and a regulatory framework still under construction. None of these approaches has yet produced a fully convincing answer to the central question of how to redistribute the gains.

Three principles are beginning to emerge as durable. First, AI policy is increasingly indistinguishable from energy policy: a single hyperscale data centre can absorb the output of a small nuclear reactor, and grid capacity has emerged as the binding constraint in several geographies including Ireland, parts of the American Sun Belt and Singapore. Second, skills policy needs to move faster than the technology cycle, which means abandoning the long planning horizons that have defined traditional vocational training. Third, competition policy must adapt to a world in which the most economically valuable assets are concentrated in fewer firms than at any point in modern industrial history.

The Direction of Travel

The most honest summary of where the global economy stands in May 2026 is this: artificial intelligence has begun to reshape growth in ways that are visible in the capital accounts of major economies, partially visible in their growth forecasts, only marginally visible in their productivity statistics, and not yet visible in their unemployment data. All four of those layers will eventually move into alignment, and the order in which they move will determine whether the next decade is remembered as the most productive expansion of the modern era or as the period in which the gap between technological capability and institutional readiness produced economic outcomes that fell well short of what the technology promised.

The choice is not whether to participate in this transition. The choice is how to govern it. For policymakers, that means treating AI as a macroeconomic and social transformation rather than a sectoral technology shock. For corporations, it means building capabilities rather than buying licences. For investors, it means distinguishing between firms that are building the rails of the AI economy and those that are merely renting space on them. For India specifically, it means using the next twenty-four months to convert demographic advantage and digital public infrastructure into durable AI capability before the window of opportunity that is currently open begins to narrow.

The economy of 2030 will be shaped, more than anything else, by decisions taken in the next eighteen months. The data, the capital and the policy debates have all arrived at the table simultaneously. What happens next is a matter of choice, not inevitability.