"It's different from the dot-com era"…Wall Street veteran moves to counter AI bubble controversy

BNP Paribas holds investment outlook meeting in New York on the 10th

"Power and equipment bottlenecks act as a buffer against AI overheating"

"AI companies are investing with their own cash, not debt"

"Internet, cloud and other AI infrastructure are already in place"

"AI companies also fall under the 'too important to fail' principle"

Investments funded by the cash held by hyperscalers, rapid corporate adoption, and AI infrastructure already in place such as smartphones and laptops — BNP Paribas cited these examples on the 10th (local time) at a '2026 Global Investment Outlook' briefing to argue that recent U.S. investment in artificial intelligence (AI) is not a bubble. In particular, Pam Haggerty, a BNP Paribas lead portfolio manager recognized on Wall Street as a veteran technology sector analyst with 30 years of experience, said at the event, "If you look at the interconnected structure of the AI ecosystem like a map, you can see that some companies are, in fact, in positions that are 'too important to fail'," explaining that some AI bubbles are unlikely to spread into system-wide risk.

AI investment unfolding like an arms race

Haggerty, who has analyzed the technology sector since 1995 and experienced the entire early internet and dot-com collapse period, identified and detailed key recent controversial issues such as AI capital expenditure (CapEx), the expansion of private credit, opaque financing structures, and infrastructure bottlenecks.

Haggerty first pointed out the bubble risk appearing in the AI field. He said, "The AI infrastructure build-out competition is unfolding like an 'arms race'," and added, "Massive CapEx is concentrated over a short period, raising the possibility of overbuild." He also noted that the uncertainty of return on investment (ROI) is a structural similarity to the dot-com era.

He also cited the risk that some companies are raising funds through off-balance-sheet financing structures — such as private credit, joint ventures (JVs), and special purpose vehicles (SPVs) — where a company's assets and liabilities are not fully reflected in official financial statements. He said, "Vendor financing was a spark for the dot-com collapse in the past," and "a similar structure is appearing in the recent AI sector."

He mentioned circular dependency structures in which certain companies simultaneously connect supply chains and demand networks. In such structures, a shock to one company can spread throughout the ecosystem.

According to McKinsey, global AI infrastructure investment needed by 2030 amounts to $6.7 trillion. Haggerty said, "This is about 6–7 times the nature of U.S. telecom infrastructure investment from 1996–2001 (approximately $1 trillion in present value)," adding, "At that scale, concerns about a bubble naturally arise."

"Different structure from the dot-com era…premature to label it a bubble"

Haggerty drew a line, however, saying it cannot be considered a bubble. He explained, "Currently, most big tech companies are funding the majority of their CapEx with internal cash flow rather than debt," and "the financial structure is far healthier compared to the dot-com era."

Corporate AI adoption is also proceeding faster than expected. He said, "AI is the most fundamental technological innovation since the internet," and cited expanding application areas such as autonomous driving, robotics, industrial automation, and new drug development.

He also said, "Unlike the dot-com era, global internet and cloud infrastructure are already in place," noting that the speed of technology diffusion is incomparable. In fact, ChatGPT reached 800 million weekly users in less than three years.

Valuations are also more moderate compared to the dot-com era. Haggerty said, "PSRs for tech IPOs in 1999–2000 were 40–50x revenue, whereas recent IPO averages are 11x." The forward P/E of the MSCI World IT Index has fallen from 60x in 2000 to below 30x today. Forward P/E is a valuation metric calculated based on expected earnings over the next 12 months.

He added that most extreme valuation cases are formed in the private market, so they do not directly impact the stock market of listed companies.

Haggerty also assessed factors that act to slow excessive investment pace. He said, "AI data center construction is heavily constrained by physical limits such as power, cooling, and equipment supply," and "these bottlenecks naturally moderate the CapEx cycle."

AI leverage unlikely to spread into systemic risk

Haggerty acknowledged that there are clearly companies in the AI ecosystem that are 'too important to fail.' Given that the financial structures of large hyperscalers are currently very robust, he judged that the likelihood of the situation spreading into systemic risk in the short term is low. He added, "Risks exist, but at this stage individual failures do not appear likely to immediately translate into system-wide problems," and "we are closely monitoring the pace of debt expansion at some unlisted AI companies."

Daniel Morris, a market strategist, offered an index- and macro-level perspective. He said, "Although Nasdaq-listed companies' CapEx has increased significantly, analysts' two-year-ahead earnings forecasts are not at overheated levels." This suggests that, unlike the 2000 dot-com bubble, earnings expectations themselves are not abnormally inflated. Morris expects the growth rate of AI-related CapEx to gradually slow and analyzed, "Looking at current indicators, the risk of a systemic collapse does not stand out."

Chris, the panelist on credit and fixed income, diagnosed that concerns about systemic risk from the credit market side are limited. He emphasized that the companies currently deploying the most capital are hyperscalers and top-tier tech firms, which have low leverage ratios and very high credit ratings. He explained that the private lending market is small relative to the overall capital markets and has a diversified investor base, so risk is not concentrated in specific institutions.

New York = Park Shin-young, correspondent nyusos@hankyung.com