Learning From History: The AI Investment Wave Likely To End This Year, But The Fed May Shield U.S. Stocks

Major hyperscale data‑center operators such as Amazon Web Services (AWS) have pledged billions more in capital expenditure for artificial intelligence (AI), and most analysts remain optimistic that large technology and energy firms will continue to benefit from elevated capex in the near term. Nevertheless, Dhaval Joshi, Chief Strategist at Counterpoint within BCA Research, expresses skepticism. In his recent report he argues that, by historical analogy, the current surge in AI investment is likely to conclude this year, while stressing that the eventual adjustment should not be as severe as the internet bust of 2000.

Joshi notes that U.S. technology capital expenditure has reached a record 7.2 percent of GDP, exceeding the 7.1 percent peak seen during the work‑from‑home surge after the pandemic and materially above the 6.4 percent level recorded around the internet bubble. He adds that if hyperscale operators follow through on commitments to spend roughly $1 trillion on AI‑related capex in 2026–27, the share could rise further. At the same time, he cautions that any comparison with prior technology spending booms must account for a long‑term structural increase in tech investment. Over the past century, total U.S. gross capital formation has hovered near 15 percent of GDP, but the economy’s shift from 20th‑century heavy industry to 21st‑century lighter, tech‑intensive sectors has steadily raised the share attributable to technology. In 1950, tech capex represented only about 1 percent of GDP, less than one‑tenth of total investment, and that share has increased by roughly 0.7 percentage points per decade since.

To evaluate the limits of the current AI spending cycle, Joshi recommends assessing technology capex relative to its structural trend rather than on an absolute basis. He proposes examining the ten‑year change in tech capex as a share of GDP and subtracting the structural 0.7 percent per‑decade growth to isolate the cyclical component. Drawing on prior relative peaks—namely the personal computer boom of the 1980s, the internet bubble around 2000, and the pandemic‑era work‑from‑home surge in 2020–21—he derives three key observations. First, each prior episode ended after the ten‑year change in tech capex exceeded the structural trend by roughly 1 to 1.5 percentage points. Second, the current acceleration has not yet reached those historical excesses, implying there may still be room for further expansion. Third, given the spending commitments from hyperscale operators, the empirical ceiling for the AI capex surge is likely to be reached by the end of 2026.

Expectations about that trajectory, Joshi argues, will shape investor behavior because equity markets are forward‑looking and tend to price in changes before they fully materialize. Crucially, he highlights a key difference from past tech busts: the prevailing interest‑rate backdrop is more supportive. Historical collapses in technology stocks—such as those in 1984, 2000 and 2022—occurred during Federal Reserve tightening cycles, whereas the current outlook anticipates further Fed easing. If U.S. inflation remains near 3 percent as Joshi expects, real interest rates should continue to decline, a dynamic that previously helped delay a broader tech correction in 2021.

On this basis, Joshi contends that 2026 will resemble 2021 more closely than 2000. Barring an external shock, he judges a U.S. recession to be unlikely, anticipates a moderation of bull‑market momentum while equities still outperform cash, and expects technology stocks to cede leadership—resulting in the S&P 500 outperforming the Nasdaq.