Is AI capital expenditure too crazy? Goldman Sachs: This is just the beginning

Recently, the massive capital expenditure in the AI field has raised market concerns about its sustainability. Goldman Sachs' latest research report clearly reveals that current AI investment levels are far from overheating, and this level of investment is sustainable, indicating that the macro story of AI infrastructure development remains robust. On October 19th, according to news from the Chasing Wind Trading Desk, Goldman Sachs' latest report believes that the scale of AI investment is not excessive, and the current technological backdrop still supports AI capital expenditure. Moreover, the proportion of AI-related investment in the U.S. GDP is currently far lower than in other historical technology cycles. At the same time, they expect that the productivity improvements brought by AI will generate $8 trillion in capital income for U.S. companies, far exceeding the current and foreseeable total AI investment. The AI Investment Boom is Sustainable Since mid-2023, investment in AI infrastructure has continued to accelerate. In 2025 alone, publicly traded U.S. companies are expected to see an incremental revenue of approximately $300 billion in AI-related infrastructure investments. Data from the U.S. national accounts shows that AI-related spending has an annualized growth rate that has increased by $277 billion compared to 2022. Since September, OpenAI has announced a series of significant investment agreements: a $300 billion partnership with Oracle, a $100 billion investment from NVIDIA, a strategic collaboration with AMD to deploy 6GW of GPU computing power, and a partnership with Broadcom to deploy 10GW of custom AI chips. The report points out that the technological backdrop still supports AI capital expenditure, based on two main reasons: On one hand, productivity improvements are significant. Goldman Sachs estimates that once generative AI is fully applied, U.S. labor productivity will increase by 15%, a process that will gradually unfold over the next decade. Academic and corporate case studies show that AI applications can bring about a 25-30% average productivity increase, although currently only 2.5% of jobs face automation risks, mainly concentrated in programming, customer service, and consulting fields. On the other hand, the demand for computing power continues to rise. The growth rate of AI model sizes (400% annually) far exceeds the decline rate of computing power costs (40% annually), with the demand for training queries and cutting-edge models growing at annual rates of 350% and 125%, respectively. Although energy efficiency has improved, it is difficult to offset the expansion of demand. As long as the growth in computing power demand outpaces the decline in costs, investment in AI infrastructure will have sustained momentum. Current AI Investment Levels are Not High The report indicates that although AI infrastructure investment has reached a nominal high, it is not exaggerated compared to historical technology cycles. Historically, the peak investment in technology cycles such as railroads, electrification, and IT accounted for 2-5% of GDP, while the current proportion of AI investment in the U.S. is still less than 1%. Goldman Sachs estimates that the productivity improvements brought by generative AI will create a present value of $20 trillion for the U.S. economy, of which $8 trillion will flow to U.S. companies as capital gains. Even under pessimistic or optimistic assumptions, this range is between $5 trillion and $19 trillion, significantly higher than the current and future total AI investment. More importantly, this estimate does not yet account for overseas profits, emerging profit pools, or the potential gains from AGI (Artificial General Intelligence)