Dom Rizzo, the global technology equity fund manager at T. Rowe Price, expects hyperscaler AI spending to reach something like $1.5 billion to $1.6 trillion in 2027, above rival forecasts from Bank of America and Morgan Stanley. He sees the AI boom's volatility echoing 1998, but argues the underlying revenue growth makes a repeat of the dot-com bust unlikely.
Rizzo, interviewed on the Excess Returns podcast published Tuesday, argues hyperscalers are only halfway through their spending cycle and on the cusp of an acceleration, not a slowdown. After a 75% jump in capex spending in 2026, he expects that pace to keep building into next year.
A forecast above Wall Street's numbers
Rizzo's estimate tops other Wall Street calls. Bank of America analyst Vivek Arya has predicted $1.2 trillion in 2027 capex. Morgan Stanley, meanwhile, has forecast roughly $1.1 trillion. Rizzo points to Amazon's July earnings call, where executives laid out an expected payback timeline of roughly two to three years, followed by two to three years of strong cash flow, as evidence hyperscalers see the returns as worth the outlay. He says the spending is not rolling over: "They are at an inflection point."
Shares of the hyperscalers he cites also moved lower. Amazon fell 2.09%. Alphabet dropped 3.61%. Microsoft slipped 0.44%.
Uncanny echoes of 1998, but different fundamentals
Rizzo compares the current AI run to the 1998 buildup that preceded the dot-com bust: the near-collapse of Long Term Capital Management then has a mirror in last month's Situational Awareness blow-up now. The 40% drawdown of 1998 was almost matched by a 22% drawdown in chip stocks in July. But the fundamentals diverge sharply: the semiconductor industry's revenue fell 8% in 1998. Industry sources now forecast something like a 64% revenue increase for 2026.
He also downplays funding worries, saying the gap between hyperscalers' capex needs and their equity is not that big relative to their debt-to-cash-flow ratios.
Where Rizzo sees the money flowing
Rizzo expects an 80:20 split, in which 80% of corporate AI tokens run on open-weight or open-source models, while the remaining 20% stay on frontier models like Anthropic or OpenAI — yet most of the value still accrues to the frontier labs as they orchestrate other models. That shift, he argues, threatens traditional software-as-a-service providers such as Workday and Salesforce, which risk being reduced to a low-value connective layer between users and the frontier models built by others.
In the short term, Rizzo favors logic semiconductor makers, naming Taiwan Semiconductor Manufacturing Company, Samsung and Intel as best positioned to benefit from strong chip pricing.
Source: MarketWatch
Trading involves risk.
Risk Disclaimer
