Meta Pixel
Insights

AI Must Save 40 Minutes of Work a Week to Pay for Big Tech’s Investments

Bradesco study estimates $768 billion in additional capital spending by major US technology companies.

The artificial intelligence boom has unleashed a wave of capital spending by the world’s largest technology companies, with hundreds of billions of dollars flowing into data centers, advanced chips and computing infrastructure. But as spending accelerates, investors face a fundamental question: will AI generate enough economic value to justify the investment? The answer matters not only for Big Tech’s profitability but also for whether the technology can deliver productivity gains commensurate with the capital being deployed.

To put that challenge into perspective, Vitor de Holanda Jó, an economist at Brazilian bank Bradesco’s economic research department, calculated how much AI would need to boost US productivity to justify the investment. His conclusion: the technology would have to generate economic benefits equivalent to roughly 40 minutes of work per week for every American worker.

The study estimates that Microsoft, Alphabet, Amazon, Meta and Oracle will have invested an additional $768 billion between 2023 and 2026, compared with a hypothetical scenario in which their capital spending had grown in line with US nominal GDP.

That does not mean American workers would necessarily leave work 40 minutes earlier each week. The figure represents the economic value of the productivity improvements AI would need to deliver, averaged across the US workforce, to generate an adequate return on invested capital. Those gains could come from completing tasks faster, cutting costs or producing more with the same workforce, without necessarily reducing working hours or eliminating jobs.

According to the study, the five companies are expected to spend approximately $732 billion on capital expenditures in 2026 alone, much of it on data centers and AI-related equipment. That amounts to 2.25% of US GDP, up from 0.6% in 2022.

Drawing on a recent National Bureau of Economic Research (NBER) working paper, What Investment Data Implies about the AI Transition, Bradesco approached the question in reverse: rather than forecasting how much productivity AI might generate, it started with the scale of investment and calculated the productivity gains needed to make that spending economically worthwhile.

Under the bank’s central assumptions — an 8% cost of capital and a six-year useful life for equipment — the additional $768 billion in investment would require an annual economic return equivalent to approximately $166 billion.

But AI providers do not capture all the benefits their technology generates. Some accrue to businesses using AI, while others may be passed on to consumers through lower prices. Assuming technology providers capture half the economic gains, AI would need to generate approximately $332 billion in annual benefits across the US economy.

That represents about 2% of total US labor compensation, estimated at $16.5 trillion annually. Based on an average 34-hour workweek, the equivalent productivity gain is approximately 40 minutes.

Higher productivity, fewer jobs?

Expressing productivity gains in minutes of work also raises a question that extends beyond investment returns: what will AI mean for employment?

If companies can produce the same amount of goods and services using fewer working hours, they may slow hiring or replace certain functions with automated systems. Conversely, greater efficiency could lower prices, stimulate demand, support business expansion and create new occupations. The net impact will depend on how quickly those changes unfold and whether the economy can absorb displaced workers.

There is also an important distinction between what benefits an individual company and what benefits the broader economy. A business may increase profits by automating tasks and reducing payroll expenses. But if widespread automation leads to persistent unemployment or downward pressure on wages, household consumption could weaken.

In that scenario, some corporate productivity gains could come at the expense of household income, potentially constraining demand growth.

Bradesco’s study does not estimate AI’s impact on employment or conclude that the technology will cause job losses. Its calculations focus on the economic benefits needed to justify investment, regardless of how those gains are distributed among businesses, workers and consumers.

The results also depend heavily on the assumptions used.

In its central scenario, Bradesco assumes that, without the AI boom, the five companies’ capital expenditures would have grown in line with US nominal GDP from their 2022 levels.

Under a more favorable scenario for AI, spending would instead have continued growing at roughly 30% annually, as it did between 2015 and 2022. That would reduce the estimated additional investment from $768 billion to $162 billion.

Keeping the assumptions of a six-year equipment lifespan and 50% capture of economic benefits, the required productivity gain would fall to just nine minutes per worker per week.

Bradesco considers this alternative less plausible because it implies the companies would have spent nearly half a trillion dollars in 2026 even without the expansion of artificial intelligence.

Chip obsolescence poses the biggest risk

One of the study’s main concerns is the potential gap between the accounting and economic useful lives of AI equipment.

In recent years, major technology companies have extended the depreciation periods for servers from three or four years to five or six years. This reduces depreciation expenses recognized in their financial statements and boosts reported earnings.

The risk is that successive generations of more powerful and efficient processors could make existing equipment economically obsolete before the end of its accounting life.

If the useful life of AI equipment is only four years, the required productivity gain rises to approximately 58 minutes per worker per week, assuming providers capture half the benefits. If they capture just 25%, the requirement jumps to 116 minutes — nearly two hours.

Bradesco also notes that the 2% productivity improvement required under its central scenario is equivalent to roughly one or two years of normal US labor productivity growth.

And that calculation covers only investment through 2026, including projected spending for the remainder of the year. It excludes plans for 2027, which the study says already exceed estimated 2026 expenditures.

The exercise has limitations. Not all capital spending by the five companies is directly related to AI: Amazon, for example, also invests in warehouses and logistics infrastructure. Moreover, AI could generate economic benefits beyond labor savings, reducing the productivity improvement needed to justify the investment.

On the other hand, excluding certain finance leases may understate the amount of capital actually committed.

The closest historical comparison, according to Bradesco, is the telecommunications and internet infrastructure boom of the late 1990s.

US investment in information-processing equipment and software reached 8.7% of GDP in 2026, the highest level since the 9.1% recorded in 2000.

The fiber-optic networks built during that earlier boom ultimately supported decades of digital economic expansion. Yet many of the companies that financed those networks did not survive long enough to capture the returns.

A similar distinction may prove crucial today: artificial intelligence could transform global productivity without necessarily delivering sufficient returns, or delivering them quickly enough, to the companies financing its infrastructure.

Leave a Reply

More in Insights

Advertisement

The BSG Morning

Your daily briefing on Brazil’s economy, companies and markets, the curated read portfolio managers and investors rely on.