- Goldman Sachs expects the five largest US hyperscalers to accelerate AI infrastructure spending next year, raising fresh questions about revenue growth, data-center demand, power requirements and the returns on the industry’s massive capital outlays.
The five largest US hyperscalers could spend about $1.2 trillion on AI infrastructure in 2027, according to Goldman Sachs strategists led by Ryan Hammond, in a forecast reported by Bloomberg. The estimate represents an increase of more than 50% from roughly $800 billion expected in 2026 and is above the current Wall Street consensus of about $1.1 trillion, according to Goldman.
The projection highlights how quickly the AI infrastructure cycle is expanding. Spending is flowing into advanced processors, networking equipment, data centers, cooling systems and electricity capacity as companies including Microsoft Corp. (NASDAQ: MSFT), Amazon.com Inc. (NASDAQ: AMZN), Alphabet Inc. (NASDAQ: GOOGL), Meta Platforms Inc. (NASDAQ: META) and Oracle Corp. (NYSE: ORCL) build out computing capacity.
Goldman expects the spending to remain elevated beyond 2027, with hyperscaler capital expenditures potentially reaching $1.4 trillion in 2028. The firm's broader AI infrastructure work estimates that roughly $7.6 trillion could be invested across compute, data centers and power between 2026 and 2031.
AI Infrastructure Spending Is Expanding Beyond Chips
The scale of the spending means the AI investment cycle is increasingly becoming an infrastructure story rather than simply a semiconductor story. Goldman estimates that global AI investment could exceed $1 trillion in 2026, although that broader figure includes spending beyond the five major US hyperscalers and adjusts for investment by private and non-US companies.
For hyperscalers, the immediate requirement is additional computing capacity. AI models require large numbers of accelerators, but those processors also need high-speed networking, specialized cooling, and increasingly large amounts of electricity.
That is creating opportunities throughout the infrastructure chain. NVIDIA Corp. (NASDAQ: NVDA), for example, remains a major supplier of AI accelerators, while memory, networking, server and power-equipment companies are also benefiting from rising data-center investment.
NVIDIA's latest quarterly results illustrate the scale of demand. The company reported $89 billion in Data Center revenue, up 117% year over year, as total quarterly revenue reached $96.2 billion.
The physical constraints around that growth are becoming increasingly important. Goldman has identified data-center construction costs, chip replacement cycles, architecture choices, and power, labor, and equipment bottlenecks as major variables that can change the ultimate amount of capital required for the AI buildout.
Recent developments at Oracle also show how infrastructure bottlenecks can affect the timetable. The company's New Mexico data-center project has faced construction and power-related issues, underscoring the practical challenges involved in converting AI demand into operating capacity.
The spending requirements are also reaching the electricity market. Goldman previously estimated that US data-center power demand could rise sharply as AI facilities consume more electricity, making grid connections and new generation capacity an increasingly important part of the investment cycle.
The $1.2 Trillion Spending Cycle Faces a Revenue Test
The central question for investors is whether AI revenue can grow quickly enough to support the enormous infrastructure investment. Goldman estimates the hyperscalers would need approximately $300 billion in annual AI revenue in the coming years to break even on the investment, according to Bloomberg's report.
Microsoft is already seeing substantial growth in Azure. The company reported 43% growth in Azure and other cloud services revenue in its latest quarter, while Microsoft Cloud revenue reached $59.3 billion. Its Azure AI economics and infrastructure spending have therefore become an important part of the broader debate over whether hyperscaler spending can translate into sufficient returns.
Amazon is facing a similar equation through AWS. The cloud division generated $42.2 billion in second-quarter revenue, up 37% year over year, while Amazon expects roughly $220 billion in total capital expenditures during 2026. Its AWS AI infrastructure expansion is therefore directly tied to the company's ability to convert additional capacity into cloud revenue.
The financing of that infrastructure is another emerging issue. Goldman expects debt to account for about 35% of hyperscaler capital expenditures in 2027. The firm has said the delay between infrastructure investment and monetization is helping drive greater reliance on debt markets.
Reuters separately reported that gross debt issuance from hyperscalers could reach $420 billion in 2027, up about 60% from 2026 estimates. The increase reflects the financing requirements created by data-center and AI infrastructure expansion.
The result is a more complicated AI investment cycle. Stronger spending supports demand for chips, servers, networking equipment, construction and power infrastructure, but it also increases the amount of revenue and cash flow that hyperscalers ultimately need to generate from AI.
Goldman's $1.2 trillion forecast therefore represents more than another increase in technology spending. It signals that the physical buildout behind artificial intelligence could continue expanding at an extraordinary pace in 2027, while the financial test increasingly shifts toward AI revenue, utilization, free cash flow, and returns on capital.