Goldman Sachs warns Wall Street is underestimating 2027 hyperscaler AI spending

Goldman Sachs is pushing back against Wall Street’s prevailing view on how much money major hyperscalers will spend on AI infrastructure by 2027. In market coverage of a Goldman analysis, the bank argued that consensus capital expenditure forecasts are underestimating the scale of buildouts needed for advanced artificial intelligence workloads.

The market report frames Goldman’s position as a broad assessment rather than a single company-specific call. Goldman Sachs said its own estimates run higher than the market consensus for hyperscaler capital expenditures in 2027, implying that spending on data centers, computing hardware, and related infrastructure may accelerate more than investors currently expect.

Hyperscalers are the large cloud and data-center operators that provide on-demand computing, storage, and networking to enterprises and developers, and increasingly, to AI model builders. Their capital expenditure plans are a key input for markets because they often determine how quickly additional capacity, such as power-hungry data center expansion and high-performance compute clusters, comes online.

Goldman’s caution matters for the broader AI supply chain because hyperscaler capex tends to ripple through multiple segments. More spending can translate into higher demand for servers and networking equipment, power and cooling infrastructure, data center construction and engineering, and transport and storage capacity that supports AI training and inference.

Even with that linkage, the publicly available market post did not provide the specific figures or the underlying assumptions Goldman used to arrive at its higher 2027 estimate. It also did not break out whether the difference versus consensus came primarily from larger data-center expansions, faster hardware refresh cycles, increased networking intensity, or higher overall capacity utilization targets.

For investors and corporate buyers, the core question is timing and magnitude. If hyperscalers spend more than consensus expects, that can raise expectations for near-to-medium-term demand across the infrastructure stack and potentially shift what companies in related industries view as the next capex cycle. It can also affect how quickly cloud customers receive additional compute capacity, which is increasingly central to AI deployment roadmaps.

What remains unclear from the market coverage is how Goldman views risk factors such as regulatory permitting timelines, electricity supply constraints, and potential optimization of training and inference costs. The bank also did not outline, in the report summary available here, whether it expects hyperscaler spending to plateau after 2027 or continue rising beyond that horizon.

Investors will likely look for additional detail in subsequent Goldman commentary and in hyperscaler disclosure. Next, the key indicates to watch are updated capex guidance from major cloud providers, as well as quarterly commentary on AI infrastructure buildouts, data-center power availability, and the pace of new high-performance compute deployments.