The AI Boom Faces a $300 Billion a Year Test. Goldman Sachs Says Hyperscalers Still Have a Long Way to Go.

As investors pile back into the technology giants spending hundreds of billions of dollars to build the infrastructure behind artificial intelligence, Goldman Sachs says those companies still face an enormous hurdle before the boom can be considered financially successful.

Goldman Sachs strategist Ryan Hammond estimated that major hyperscalers like Amazon, Microsoft and Oracle will need to generate roughly $300 billion in annual AI revenue in the next few years simply to break even on their investments.

That calculation offers a reality check for investors who have pushed back into AI-linked stocks as enthusiasm over the technology accelerates again. The spending behind the boom is enormous.

Goldman estimates that U.S. hyperscalers are on track to spend roughly $800 billion on capital expenditures in 2026. The bank also estimated that total global AI investment will surpass $1 trillion this year, including about $581 billion in the United States.

Cloud growth suggests that monetization is already happening, but not yet at the scale needed to justify the investment. Hyperscaler cloud revenue in the second quarter was running at an annualized rate roughly $70 billion above the trend that existed before the generative AI boom, according to Hammond's analysis.

There is also a massive amount of contracted business waiting in the pipeline, with announced backlogs among hyperscalers exceeding $1.5 trillion. But backlogs do not automatically translate into the returns investors ultimately expect.

The hurdle becomes considerably higher when looking beyond simply recovering the cost of infrastructure. "We estimate that AI users would need to spend roughly $1 trillion annually on AI applications in order for the hyperscalers to generate solid returns on investment and the application layer to generate strong profit margins on their compute expenses," Hammond wrote.

That means the AI investment story increasingly depends on what happens outside the data center. Companies have spent years buying Nvidia chips, building data centers, securing electricity and expanding cloud capacity.

For those investments to produce attractive returns, businesses and consumers eventually have to spend enough on AI applications, agents and services to support the infrastructure underneath them.

Goldman has previously identified enterprise adoption as particularly important. Consumer use of AI has grown rapidly, but many consumers continue to use free versions of AI products. Businesses paying for AI tools and integrating them into everyday workflows could therefore become critical to the economics of the industry.

There are reasons for optimism. Goldman expects AI agents to dramatically increase computing demand, forecasting a 24-fold increase in token consumption by 2030 as businesses and consumers adopt more sophisticated applications. Falling computing costs could also eventually improve hyperscaler margins.

Investors, meanwhile, have shown little willingness lately to abandon the AI trade. The Roundhill Magnificent Seven ETF, which gives exposure to the dominant U.S. megacap technology companies, has rallied sharply as enthusiasm returned to technology shares. The Magnificent Seven collectively reached a record $24.52 trillion in market capitalization earlier this week.