The White House is framing artificial intelligence as the infrastructure of a new economic divergence. Its paper, “Artificial Intelligence and the Great Divergence,” compares today’s investment in compute and data centres with railway construction during the Industrial Revolution.
The argument is that long-term growth comes from productivity, and AI can raise it through both capital investment and widespread use. The paper says AI investment added 1.3 percent to US GDP in the first half of 2025, while AI-related infrastructure represented about one quarter of all US investment that year.
The comparison is powerful, but railways also remind policymakers that infrastructure booms create constraints, uneven gains, and expensive assumptions.
Capital Spending Is Already Driving the Growth Story
Investment in data-processing equipment, buildings, infrastructure, and software grew 28 percent in early 2025. Training compute has expanded roughly fourfold each year since 2010, while the cost of model output has fallen rapidly across tasks and systems.
The report combines those trends with rising adoption. By late 2025, 78 percent of organisations reported using AI, up from 55 percent in 2024. It says 40 percent of US workers use generative AI at work and nearly half of businesses pay for subscriptions.
These figures support the claim that AI is moving beyond experimentation, but activity is not identical to productivity. Data centres contribute to GDP while they are being built; the larger long-term return depends on whether organisations use the resulting capacity to produce better goods, services, and decisions.
The paper presents estimates ranging from single-digit GDP gains to 20 percent productivity growth within a decade. More extreme scenarios exceed 45 percent if AI substitutes for human labour over the longer term. That range reveals deep uncertainty rather than a single forecast.
Energy and the Grid Set a Physical Limit
AI infrastructure is electricity-intensive. The White House paper projects that data centres could consume as much as 12 percent of domestic electricity by 2028 and treats energy availability as a prerequisite for leadership.
That makes the grid part of AI policy. Faster permitting can accelerate construction, but generation, transmission, local reliability, water, equipment supply, and community acceptance determine whether projects can operate sustainably.
The One Big Beautiful Bill Act provides financial incentives for data centres and IT infrastructure and seeks to speed facility construction. The report also argues that deregulation can lower cost and increase competition.
Policy must distinguish useful acceleration from transferring costs. If grid upgrades are socialised while the benefits concentrate among a few firms or regions, the infrastructure boom may produce political resistance. Transparent demand forecasts, cost allocation, efficiency standards, and regional planning are economic necessities, not peripheral environmental concerns.
International Leadership Is More Than Compute Capacity
The paper presents the United States as leading in private investment, model development, and compute, while Europe trails across several metrics and China remains dependent in part on US-designed hardware. Trade agreements and foreign-policy commitments to purchase American chips and infrastructure reinforce the strategy.
Compute leadership can attract talent, capital, and application development. Yet economic advantage also depends on skills, competition, diffusion into smaller firms, research, reliable institutions, and whether workers can move into productive new roles.
The Industrial Revolution produced a “great divergence” because technology interacted with capital, energy, trade, and institutions. AI may do the same, but outcomes will not follow automatically from server capacity.
The White House vision gives businesses a clear signal: US policy intends to align investment incentives, energy, construction, and international commerce around AI growth. Companies will respond, adding momentum to the infrastructure cycle.
The harder public question is how to judge success. Construction spending, model benchmarks, and subscription rates are early indicators. Sustained productivity, competitive markets, broadly shared income gains, resilient grids, and new business creation will determine whether AI becomes economic infrastructure rather than an expensive asset boom.
Railways changed where people and goods could go. AI’s comparable achievement would be expanding what organisations and workers can produce. The tracks are being laid quickly; the destination remains a policy choice.
