Bullish on AI—so why bearish on the big winners in U.S. stocks?

The contrast within the same report

The contrast in Vanguard’s economic and market outlook report is that AI capital investment is expected to be substantial, while U.S. stocks are expected to deliver annualized returns of just 4% to 5% over the next 5 to 10 years.

This isn’t a bearish view of AI’s prospects; it’s about assessing the economy and stocks separately. AI may genuinely boost productivity and could bring a wave of large-scale investment, but if stock buyers pay too high a price, valuations will eat up a large chunk of future returns.

Good economic news does not necessarily mean good news for stocks.

It’s easy to conflate two ideas: if AI makes companies more efficient, their profits should rise; and if profits rise, their stocks should go up. But stock returns depend not only on profits, but also on the price paid to buy in. This is especially true when major U.S. stocks have already priced in much of the good news: subsequent good news must be even bigger and more certain to keep supporting high returns.

CAPE valuation is used to describe this pressure. CAPE smooths out cyclical fluctuations by using earnings over a longer period, rather than focusing only on how much companies earned this year. U.S. stock valuations are already approaching levels in the top 10% of historical readings, which means that even if the economy remains healthy, future stock-market returns could still be held back by starting valuations.

This is also why it can be bullish on AI while taking a bearish view of the biggest U.S. stock winners: technological progress affects the economy, but stock returns also depend on the price paid. AI may raise productivity without necessarily boosting subsequent stock returns.

The Other Side of $2.1 Trillion

AI investment is far from small. By 2027, AI capital investment could reach $2.1 trillion, yet that would represent only 30% to 40% of historical technology investment cycles. This suggests that companies are still investing, the infrastructure buildout is not yet complete, and the investment cycle may not have reached its midpoint.

But capital expenditures consume cash upfront. Data centers, chips, energy, software, and supporting infrastructure all require spending. For any individual company, the larger the investment, the faster future revenue must materialize to keep profit margins from being squeezed. The risk is that massive spending comes first while profits take longer to recover, leaving shareholders with thinner net returns.

This uncertainty can be viewed through a net present value framework: across different moats and risk scenarios, the aggregate macro NPV generated by AI is not inherently positive; it may even approach zero or fall into negative territory. That does not mean AI is useless; after investment, competition, and distribution are factored in, the value that ultimately accrues to investors may be less substantial than imagined.

The winners aren’t necessarily the hottest stocks.

One often-overlooked consequence of AI diffusion is that its benefits may be shared across more industries rather than staying in the hands of a few tech giants. If AI tools become widespread, users can cut costs and improve efficiency, while competitors can use the same tools to catch up. AI net present value comparisons therefore need to factor in the strength of companies’ moats, not just who rallies first or whose market capitalization is largest.

This will change how the stock market separates winners from losers. Companies once bought as “certain winners” will need to prove not only that their businesses continue to grow, but also that the growth is sufficient to justify high valuations and heavy capital spending. By contrast, lower-valued assets with less fully priced-in expectations may find it easier to deliver higher expected returns, as long as their fundamentals are not too weak.

So, cheaper assets imply higher expected future returns: about 7% for value stocks, around 6% for developed-market stocks outside the US, and just 4% to 5% for US large-cap stocks. This does not simply mean that “tech stocks are no good”; when prices are already very high, even good companies can become mediocre investments.

The Middle Scenario Beyond the Bubble

Discussions about AI and the stock market are often pushed to two extremes: either the revolution pays off immediately, or the bubble bursts immediately. This is closer to a middle-ground scenario. AI investment is still rising, and productivity could also improve; at the same time, capital spending, valuations, and profit margins will weigh on stock returns.

There is also a timing point here: the relevant productivity effects are framed as occurring around 2028, alongside the figure 7.5. This suggests that AI’s contribution to the economy has not yet fully materialized and will require investment, deployment, diffusion, and competition.

This process may boost economic activity, but it may also depress stock prices. At the economic level, the focus is on total output, total investment, and efficiency; at the stock-market level, it is on who provides the money, who earns the profits, and whether purchase prices have already become too high. Putting these two points together leads to the conclusion that “AI is important, but major U.S. stock-market winners may not necessarily be worth the price.”

---

Source institutions：Vanguard

This content is for reading and understanding research reports. It does not constitute investment advice or trading signals.