The AI boom has uncanny echoes of 1998, but one crucial metric shows why it won't crash
AI capital expenditure could reach $1.6 trillion next year, says a fund manager.
When the dot-com boom and bust of 1998-2002 started, Dom Rizzo, the global technology equity fund manager at T. Rowe Price, was only five years old.
However, as a student of market bubbles and economic cycles, Rizzo has opinions on the comparison that can be drawn with the current AI bonanza.
Interviewed on the Excess Returns podcast that was published Tuesday, Rizzo observed the echoes of 1998 are uncanny: the crisis triggered by the near-collapse of Long Term Capital Management then, has a mirror image now in the Situational Awareness blow-up of last month; the high-volume, high-momentum trading factor generating daily and monthly volatility is also similar. The 40% drawdown witnessed in 1998 was almost matched by a 22% drawdown in chip stocks SOX in July.
However, where Rizzo finds the most fundamental difference is in the fundamentals themselves. The semiconductor industry experienced an 8% revenue decline in 1998, mostly caused by a decline in pricing. In 2026, though, "industry sources have something like a 64% revenue increase" in their forecasts.
This time around, and this may surprise some observers, Rizzo thinks companies are halfway through the spending cycle, and are on the cusp of an acceleration in capex. After the 75% jump in 2026, Rizzo expects something like $1.5 billion to $1.6 trillion in capex from the hyperscalers in 2027. This is considerably higher than the more ambitious forecasts out there, like Bank of America's $1.2 trillion, predicted by analyst Vivek Arya.
Morgan Stanley recently forecasted roughly $1.1 trillion of AI capex in 2027. Rizzo is much more bullish.
Rizzo believes the hyperscalers, like Amazon (AMZN), Alphabet $(GOOG)$ and Microsoft $(MSFT)$, are incentivized to maximize their investment spending because the returns on invested capital are so appealing. While many AI skeptics worry about the payback on all this capex, Rizzo cites the Amazon guidance and earnings call in July in which executives laid out the returns they expect. "Two or three years to get your money back, then two to three years of great cash flow" is how Rizzo puts it.
Hyperscaler capex is not rolling over, in Rizzo's opinion. "They are at an inflection point."
For Rizzo the critical issue in the whole debate about the profitability of AI capex comes down to this: "What happens to the chip ecosystem, what happens to the hyperscaler ecosystem, what happens to the application software ecosystem, to infrastructure... every single question is really downstream from this concept of what do the frontier labs look like in a few years?"
Rizzo expects what he calls an "80:20 world" wherein 80% of the AI tokens used by corporates are open weight or open source and 20% of them are frontier like Anthropic or OpenAI. The key aspect, however, is that "the majority of the value will accrue to the frontier models as they orchestrate other models, as they make them more efficient."
The implications of what Rizzo anticipates are potentially threatening for the traditional software-as-a-service giants like Workday (WDAY) or Salesforce (CRM) as they are disintermediated or turned into "dumb data pipes".
However, the AI capex trends mean that in the short-term "chips lead us higher." Most parts of the chip ecosystem from design, equipment and production look set to benefit from strong pricing but logic semiconductors are the spot that Rizzo finds most attractive right now. This would include manufacturers like Taiwan Semiconductor Manufacturing Company $(TSM)$, Samsung (KR:005930) and Intel $(INTC)$.
The last reassurance, Rizzo offers investors is this: the concerns about funding all this capex is overdone."The funding gap between how much [the hyperscalers] need versus their equity is really not that big." Rizzo is confident that while the absolute numbers sound big, they really aren't in relation to their debt to cashflow ratios.
-Jules Rimmer
Comments