Goldman Sachs expects Amazon, Alphabet, Microsoft, Oracle and Meta to spend a combined $1.2 trillion on AI infrastructure in 2027. The estimate stands more than 50 percent above roughly $800 billion projected for this year and above the Wall Street consensus of $1.1 trillion. For companies that buy cloud and AI capacity, the scale matters because spending already exceeds cash from operations and still lacks matching AI revenue.
How Goldman frames the $1.2 trillion estimate
The forecast, reported by Bloomberg citing strategist Ryan Hammond, covers five spenders: Amazon, Alphabet, Microsoft, Oracle and Meta. Goldman puts 2027 outlays at $1.2 trillion against about $800 billion this year and a $1.1 trillion market consensus. The bank had already warned in June that consensus estimates were far too low, so the new figure extends that position. Measured against GDP, Goldman describes the cycle as the largest investment wave since railroad construction in the 19th century.
Growth remains steep but decelerates quickly. Goldman sees the pace falling from nearly 100 percent in 2026 to 54 percent in 2027 and then to 12 percent in 2028. Spending has moved beyond what the five companies generate from ongoing operations, which points to additional debt financing. To recoup the outlays, the group would need around $300 billion per year in AI revenue, a level current earnings do not yet reach.
The revenue side shows mixed signals. Cloud revenue growth accelerated from 25 percent in 2024 to 48 percent in the second quarter of 2026, which supports part of the case for expansion. It remains unclear whether revenue growth at AI labs OpenAI and Anthropic is fast enough to justify the infrastructure bets. Both labs sit at the center of the expectations and financial instruments behind the buildout. Power, labor and memory-chip bottlenecks could also restrain the pace of construction.
What the spending wave means for business
For enterprise buyers, the near-term effect is continued expansion of data center capacity and cloud AI services from the five vendors. Large customers with multi-year commitments may secure priority access and steadier terms while capacity is being added. Smaller firms will likely rent the same infrastructure indirectly through cloud platforms and models from OpenAI and Anthropic rather than building their own systems. Their costs and service quality will therefore track how quickly new capacity comes online and how vendors price it under heavier financing pressure.
The forecast does not guarantee returns or uninterrupted construction. Debt-funded expansion makes vendors more sensitive to utilization, pricing and contract length, while power, staffing and memory-chip supply can delay projects in specific regions. Buyers should verify delivery timelines, service-level terms and exit options instead of assuming that announced spending equals available capacity. It is also important to separate current cloud strength from proof that long-term AI revenue can cover the planned outlays.
The marker to watch is whether 2027 spending reaches $1.2 trillion or settles near the $1.1 trillion consensus, alongside the slowdown to 12 percent in 2028. If cloud and AI-lab revenue keeps climbing without new financing stress, the buildout case holds. If bottlenecks bite or revenue lags, vendors will likely stretch timelines and tighten commercial terms.
