Goldman Sachs forecasts $1 trillion in hyperscaler capex through 2030, saying Wall Street consensus is 'still too low' on AI infrastructure spending.
Goldman Sachs Global Investment Research projects that combined capital expenditure from Amazon, Microsoft, Google, Meta, and Oracle/OpenAI-tier spenders will surge from $156 billion in 2022 to approximately $1 trillion in 2027, with the research house arguing that Wall Street consensus significantly underestimates this scale. Goldman's base case for 2027 sits at approximately $1.1 trillion, with its bull case reaching $1.4 trillion—a range reflecting genuine uncertainty about AI monetization speed rather than the direction of spend.
The data traces a clear inflection. From $156B in 2022 (post-pandemic cloud normalization), spending jumped to $254B in 2024 (ChatGPT moment, GPU allocation wars), nearly doubled to $443B in 2025 (HBM shortages, data center land grabs, gigawatt power deals), and projects to $764B in 2026E and $1,018B in 2027E—each increment reflecting accelerating hardware demand and infrastructure buildout.
Two critical qualifications merit emphasis. The 2026E and 2027E figures are estimates subject to capital budget shifts and timing adjustments. ByteDance is explicitly listed as "not covered" in Goldman's research; its figure is extrapolated rather than directly sourced from company disclosures. The base-to-bull range of $1.1T–$1.4T reflects this uncertainty honestly.
China's trajectory is equally striking in relative scale. Alibaba, Tencent, ByteDance, and Baidu are projected to increase capex from roughly $8B in 2022 to $123B in 2027E—a 15x surge in absolute terms. At these projections, US hyperscalers will spend approximately eight times what China's four largest AI infrastructure investors spend in 2027—compute bifurcation rendered as a single ratio.
The underlying structural reality extends far beyond individual company budgets. This represents a new category of sovereign-scale infrastructure commitment—one that, if realized, will consume a share of US GDP last seen during railroad construction and electrification in the 20th century. Every physical bottleneck in AI—HBM memory, advanced packaging, grid power, data center land—is a direct output of this capex trajectory. These are not temporary supply mismatches but multi-year structural demand commitments. Micron's $250B AI memory buildout and SK Hynix's record earnings reflect the supply side racing to keep pace; Meta's gas-fired generation projects in Alberta exemplify how data center demand has rendered grid expansion inadequate; the entire AI stack undergoes verticalization because at this capex scale, dependency on external suppliers presents existential risk.
The US-China gap is equally structural. An approximately 8x spending differential by 2027E is not a policy gap or talent gap but a compute gap made quantitative. China does not attempt to match US hyperscaler spend; it rationally pursues efficiency within a lower absolute base—a constraint that shapes architectural innovation, as DeepSeek's efficiency-first approach demonstrates. Two distinct AI economies are forming: one competing on raw capital deployment, one on compute efficiency per dollar.
Goldman's historical comparison to railroads in the 1870s, electrification in the early 20th century, and the internet in the late 1990s carries weight. Each consumed 2–3% of GDP at peak investment; each was eventually justified by its productivity unlock but first produced brutal overbuild phases before monetization caught up. At $1T+ in annual capex, AI infrastructure approaches that GDP share threshold. The central question is whether AI monetization accelerates sufficiently by 2027–2028 to justify the capital deployment or whether financial pressure on hyperscaler balance sheets becomes acute.
This capex trajectory defines Layer 0 of the AI infrastructure stack—the raw capital that makes every layer above it possible or scarce. When this layer compounds at 6–7x over five years, every layer simultaneously becomes more valuable and more constrained. Scarcity is structural, not cyclical; the US-China gap represents the geopolitical fact of the decade; and a trillion dollars annually has to earn a return. That monetization race remains the central unresolved question in the AI economy.