Catenaa, Thursday, November 12, 2026- Six major US hyperscalers need to see $1.42 trillion in cumulative AI revenue between 2028 and 2030 to achieve a 15% ROI on their 2026–2027 investments.
In an interview with Goldman Sachs, Venture capitalist and CEO of Social Capital Chamath Palihapitiya estimates that Amazon, Alphabet, Meta, Microsoft, Oracle and SpaceX would need to generate approximately $1.42 trillion in cumulative AI infrastructure revenue between 2028 and 2030 to achieve a 15% return on invested capital from their 2026–2027 AI-compute investments.
According to Palihapitiya, this $500 billion represents the training compute and the inference cost of goods sold for their AI customers, who in turn need to generate enough revenue from differentiated AI services to build sustainable businesses.
“So, there must be demand for those AI services. We think this demand will materialize as enterprises roll out AI more broadly and adopt more sophisticated agentic processes,” he said.
However, he said that if demand slows, they would be more concerned about the economics of the entire infrastructure software supply chain.
Meanwhile, Eric Sheridan, US internet and interactive entertainment research analyst at Goldman Sachs, said that the hyperscalers have materially larger revenue backlogs now than they did a few years ago.
As of 3Q26, according to Sheridan, the three largest cloud service providers, AWS, Azure, and Google Cloud, reported a combined revenue backlog of $1.8 trillion, representing about 140% year-over-year growth amid growing enterprise demand for AI training and inference workloads.
He said that some hyperscalers have publicly disclosed that they expect roughly 50% of their backlogs to convert to revenue over the next two years.
Goldman Sachs estimates that the six hyperscalers need roughly $500 billion in annual revenue over 2028–2030 to support the targeted return.
Ryan Hammond, Senior US Portfolio Strategist at Goldman Sachs, said the six hyperscalers are on track to spend $890 billion in capex in 2026, and FactSet consensus estimates imply a further $1.4 trillion in spending next year.
“Our equity analysts estimate that the hyperscalers would need to generate annual AI revenues of roughly $350 billion in the next few years to break even on these investments,” Hammond said.
He said that their estimate of required hyperscaler revenues is even larger when factoring in a return on invested capital. “Even so, the hyperscalers continue to show increasing signs of monetization, and cloud revenue backlogs totaled $1.8 trillion during Q3 earnings season,” he added.
Moreover, Hammond said that capex in 2026 and 2027 implies even larger required revenues for the application layer, the companies buying much of the compute from the hyperscalers, given application companies’ additional costs and profit margins on top of the required revenues for the hyperscalers.
He said that scenarios for application layer EBIT margins and hyperscaler ROIC suggest a wide range of required revenues for the application layer, depending partly on which companies make up the application layer and how their costs are structured1.
For example, Hammond said that the application layer would need to generate over $1 trillion in annual revenues if those companies had a 20% EBIT margin and the hyperscalers generated a 20% return on invested capital.
