AI Capital Expenditure to Hit $1.6 Trillion Next Year, Fund Manager: This Wave Resembles 1998 Turmoil, Not Dot-com Bubble

Dom Rizzo, global technology equity fund manager at T. Rowe Price, forecasts that capital expenditure by hyperscale cloud providers will reach $1.6 trillion by 2027—significantly surpassing even the most optimistic market projections. He believes the current artificial intelligence (AI) boom, while reminiscent of the dot-com bubble, is more akin to the market turbulence of 1998 than the subsequent tech crash, with a key difference being robust semiconductor fundamentals and sustained capital spending.

Rizzo was only five years old during the 1998–2002 dot-com bubble’s rise and fall. However, as an investor who has long studied market bubbles and economic cycles, he brings a distinct perspective on how today’s AI frenzy compares to that era.

During a Tuesday episode of the 'Excess Returns' podcast, Rizzo highlighted striking similarities between today’s market and 1998. Back then, the near-collapse of Long-Term Capital Management (LTCM) triggered a financial crisis—comparable to last month’s Situational Awareness blow-up. Additionally, high trading volume and momentum-driven factors causing sharp intraday and monthly market swings are also closely mirrored today. In 1998, markets briefly pulled back by 40%; this July, the Philadelphia Semiconductor Index (SOX) dropped 22%—a smaller magnitude but similarly dramatic.

Yet, Rizzo stresses the most fundamental difference lies in underlying fundamentals. In 1998, semiconductor industry revenue declined by 8%, dragged down by falling product prices. In contrast, for 2026, industry estimates suggest semiconductor revenues could surge approximately 64%.

Rizzo believes companies are only halfway through their capital expenditure cycle, with a new acceleration phase likely ahead—contrary to some market expectations. After a projected 75% increase in capex in 2026, he anticipates hyperscalers’ spending could hit $1.5–1.6 trillion in 2027, well above aggressive forecasts like Bank of America analyst Vivek Arya’s $1.2 trillion estimate.

He notes that hyperscalers such as Amazon (AMZN-US), Alphabet (GOOGL-US), and Microsoft (MSFT-US) have strong incentives to maximize investment due to highly attractive returns on invested capital. While many AI skeptics question when such massive spending will pay off, Rizzo cites Amazon’s July earnings guidance and conference call, where executives outlined expected returns. He summarizes it as: “Payback in two to three years, followed by two to three years of strong cash flow.”

In Rizzo’s view, hyperscalers’ capex hasn’t peaked and reversed—it’s “at an inflection point.” For him, the core debate around AI capex profitability ultimately hinges on one question: “How will the chip ecosystem evolve? How will the hyperscaler ecosystem change? How will applications and infrastructure develop? All these questions depend on one thing: What will frontier AI labs look like in a few years?”

Rizzo predicts a future “80-20 rule world,” where 80% of enterprise AI tokens will be open-weight or open-source, while 20% will belong to cutting-edge models like Anthropic or OpenAI. If this comes true, it could threaten traditional SaaS giants like Workday (WDAY-US) and Salesforce (CRM-US), potentially disintermediating them or reducing them to mere “data pipelines.”

Nevertheless, from the standpoint of AI capex trends, “chip stocks will continue leading in the short term.” Most segments of the chip ecosystem—from design and equipment to manufacturing—are poised to benefit from strong pricing power. Logic semiconductors, in particular, stand out as Rizzo’s top pick, including major players like TSMC (2330-TW), Samsung Electronics (KR005930), and Intel (INTC-US).

Rizzo’s final reassurance to investors: concerns about funding all this capex are overblown. He argues that fears over financing gaps between required capital and internal resources are exaggerated. While the absolute numbers sound enormous, relative to debt-to-cash flow ratios, the burden is manageable, he says.