In the AI boom, which companies benefit more?

With the dust settling on the third quarter earnings season, Goldman Sachs released a report stating that the market's investment in AI infrastructure has not slowed down, but is instead accelerating upward.

Although investors have concerns about whether the credit market can absorb this investment boom and whether spending will exceed free cash flow, data indicates that the balance sheets of tech giants, known as Hyperscalers, still have significant borrowing capacity.

For investors, key signals have emerged: AI investments are transitioning from pure infrastructure development to a stage where returns are more differentiated. Current capital expenditure estimates may still be underestimated, with potential upside reaching $200 billion.

Meanwhile, the investment logic is gradually shifting from "selling shovels" (infrastructure) to AI platform stocks and "productivity beneficiaries" that can significantly improve efficiency through AI.

Capital Expenditure Estimates Significantly Upgraded, Return Differentiation Intensifies

The third quarter earnings season not only catalyzed further increases in AI capital expenditure estimates but also reignited investor concerns about the risks behind the AI investment boom.

Goldman Sachs' portfolio strategy team pointed out that the market consensus for the capital expenditure estimates of the five major AI hyperscale computing companies (Hyperscalers: Amazon AMZN, Google GOOGL, Meta, Microsoft MSFT, and Oracle ORCL) for 2026 has soared from $467 billion at the beginning of the earnings season (a year-on-year increase of 20%) to $533 billion now (a year-on-year increase of 34%).

Although the "AI trade" is currently mainly focused on the infrastructure sector, the return dispersion within this sector is increasing. This differentiation is primarily driven by two factors: the level of investor confidence in whether AI investments can truly generate revenue benefits, and the scale of leverage used to fund these investments. This differentiation may provide investors with potential opportunities to gain AI-driven returns at a "cheaper" price.

$200 Billion Upside Potential in 2026 Capital Expenditure

Despite the current expenditure figures being astonishing, analysts' forecasts may still be overly conservative.

The bottom-up market consensus suggests that the capital expenditure growth rate for hyperscale computing companies will slow from a recent 76% to 25% by the end of 2026. This means that analysts expect a significant deceleration in the growth of capital expenditures for these giants in 2026.

However, Goldman Sachs reviewed data from the past two years and found that analysts' estimates at each stage have not only been conservative but "overly" conservative. Based on historical technology investment cycle expenditure scale analysis, Goldman Sachs believes that the current capital expenditure estimates for hyperscale computing companies in 2026 still have an upside potential of up to $200 billion.

The market generally worries that cash flow and balance sheet capacity will limit expenditures in 2026, but data disproves this view.

The vast majority of hyperscale computing companies' capital expenditures have so far been funded through cash flow, but these companies also possess significant debt financing capacity. Since 2021, although these tech giants have collectively increased their net debt by $295 billion on their balance sheets, due to strong profitability growth, their collective net debt/EBITDA leverage ratio is only +0.2 times According to Goldman Sachs' estimates, these five companies can easily increase their net debt by $700 billion on their balance sheets, and even so, their net leverage ratio would not exceed 1x. Therefore, compared to cash flow or balance sheet capacity, supply chain bottlenecks or investor appetite are more likely to become constraints on recent capital expenditures.

Leverage Pressure and Feedback Loop

While the largest AI infrastructure companies (the aforementioned Hyperscalers) have strong balance sheets, many other publicly listed companies and private enterprises involved in AI infrastructure development face more severe challenges.

The recent growth rate of balance sheet debt and alternative financing has raised concerns among investors. The close ties between the largest U.S. publicly listed companies and smaller AI firms create a feedback loop. This means that pressure in one part of the AI ecosystem (especially smaller private companies) can easily transmit risks and affect investors across the entire AI sector. In fact, during the third-quarter earnings season, several hyperscale computing companies reported that changes in private equity investment values had a significant impact on their earnings.

Next Stage Winners: Platform Stocks and Productivity Beneficiaries

As corporate AI adoption continues to rise and concerns about the pure infrastructure sector grow, investors' focus is shifting. Goldman Sachs highlights the following two categories of future beneficiaries:

1. AI platform stocks: As more companies actually adopt AI technology, these companies will gain direct revenue tailwinds.
2. AI productivity beneficiaries: Goldman Sachs has identified a group of companies with high labor costs and significant wage exposure that explicitly mentioned leveraging AI automation to improve efficiency in their recent earnings calls.

As the report coldly points out at the end, while these names may mean good news for Wall Street, the implied employment "replacement" opportunities are certainly not good news for the general public