Foreign media: Goldman Sachs' latest view is that Wall Street still underestimates the duration of the AI investment cycle. Although the market has been discussing over the past year when large tech companies will slow their spending, Goldman Sachs expects hyperscale cloud providers' capital expenditures to remain above current consensus estimates through 2027.
Expenditures in 2027 may exceed expectations
Goldman Sachs estimates in its report that the capital expenditures of hyperscale cloud providers could reach approximately $1.1 trillion by 2027, higher than the current Wall Street expectation of about $920 billion. Under a more optimistic scenario, this figure could rise to $1.4 trillion.
The core assessment is that demand for AI computing power is still in its early stages. Goldman Sachs expects AI token consumption to grow 24-fold by 2030, primarily driven by the adoption of enterprise AI agents. As token usage increases, demand for data centers, chips, networking equipment, and power infrastructure will rise accordingly.
Cloud provider order backlog is rising rapidly
Goldman Sachs believes an important signal supporting this judgment comes from data disclosed by the cloud service providers themselves. The report notes that, as of the first quarter of this year, the combined backlog of orders for Google Cloud and Amazon Web Services reached $832 billion, up from $358 billion six months earlier.
Goldman Sachs estimates that the supply and demand for AI will not approach balance until at least the second half of 2027, suggesting that capital expenditures in this area may remain elevated longer than investors currently anticipate.
The bottleneck shifts to electricity and labor.
However, Goldman Sachs also noted that future spending expansion may not be constrained by financing capacity; the real limitations are more likely to stem from practical construction conditions. The report highlighted that several data center projects are already experiencing delays, with memory, power, and labor viewed as key constraints on expansion.
From a beneficiary perspective, Goldman Sachs believes that if capital expenditures continue to exceed expectations, earnings growth for companies in the semiconductor, networking equipment, cooling, and power supply chains will remain supported. However, valuations for some AI infrastructure stocks have recently expanded rapidly, with stock price increases outpacing upward revisions to earnings expectations, leading to rising volatility risks.
Corporate return verification is still limited.
Goldman Sachs also noted that, so far, there is limited large-scale, quantifiable validation of AI’s improvements to corporate efficiency and profitability. According to its statistics, while 54% of companies mentioned AI productivity during their first-quarter earnings calls, only 11% provided specific data on efficiency gains, and just 2% quantified the impact on profitability.
This also reflects a current reality in the business world: AI investments continue to rise rapidly, but returns on those investments have not yet been widely proven. Meanwhile, some companies are beginning to account for the token costs associated with AI tools, and the market is reassessing whether productivity gains can offset the costs of running models.
Additional information: The report noted that, amid concerns over the war in Iran and uncertain interest rate prospects, the Nasdaq 100 Index fell 2% on Wednesday, declining a total of 6% since the sell-off began last Friday, but remains up 13% for the year.