AI financing markets are showing signs of divergence, with highly leveraged companies under pressure while large technology firms continue to invest heavily. But a well-known asset management company believes the market has not yet entered a full-blown bubble phase, and infrastructure expansion still has several years of room to grow.

The artificial intelligence investment boom has recently faced market skepticism, with data center financing pressures and technology stock volatility intensifying. However, Steve Chiavarone, Chief Investment Officer of Global Equities at Federated Hermes, believes that investor concerns about an AI bubble may be premature, and that the massive infrastructure buildout cycle still has three to four years to complete.

According to a Bloomberg report on October 21, Chiavarone said in a Bloomberg Television interview that day that market concerns about AI investment overheating are reasonable, but the current stage is not yet one of full-blown worry about a bubble bursting. In his view, AI infrastructure is still in a period of continuous expansion, and companies will need several more years to build data centers and related computing facilities. Only when this round of construction nears completion will investors need to pay closer attention to the potential fallout from slowing capital expenditure, overcapacity, and insufficient investment returns.

This assessment stands in sharp contrast to the AI bubble warning recently issued by Citadel founder Ken Griffin.

Bubble risk does exist, but may be priced in too early

Chiavarone believes there are clear differences between the current AI capital market and typical bubble periods. In his view, a genuine bubble often manifests as overly loose financing conditions, where nearly all companies can obtain capital at extremely low cost, and investors no longer seriously differentiate project quality. But the current AI industry financing environment is not like that. The market is deciding whether to provide funding and what financing costs to demand based on companies' financial strength, debt-servicing capacity, and business prospects.

Chiavarone pointed out that most AI capital expenditure is currently still funded by companies' free cash flow, with debt financing accounting for a relatively limited share. At the same time, companies with higher leverage ratios have already begun to come under market pressure. He believes this difference in capital allocation precisely demonstrates that investors have not completely lost their sense of risk.

Recent volatility in the AI infrastructure financing market reflects this divergence. According to a report this week by tech media outlet The Information, shares of AI cloud service provider CoreWeave fell about 20% over the past week, and yields on some of its bonds had previously risen above 15%, reflecting investor concerns about the company's financing and debt repayment capacity. The Financial Times also noted that bonds related to some large data center projects have recently fallen below face value, as the market begins to reassess whether massive construction spending can generate sufficient revenue returns.

These changes show that AI infrastructure investment is not without risk, but the pressure is currently more concentrated on companies that rely on external borrowing and need continuous fundraising.

Large tech companies and AI model developers face different risks

Chiavarone also emphasized that not all AI-related companies should be viewed as the same type of investment target. He believes that the long-term business models of AI model development companies such as OpenAI and Anthropic face commoditization risk. As model technology continues to advance, competitors increase, and performance gaps between different products narrow, AI model services may find it increasingly difficult to maintain high pricing power in the future. This means that even if AI technology becomes widely adopted, some model development companies may not be able to consistently generate returns commensurate with the enormous scale of investment.

By contrast, Chiavarone believes that some large technology companies are using AI to strengthen the competitive advantages of their existing businesses. These companies already have mature customer bases, distribution channels, and profitability, and can embed AI into existing products and services to improve operating efficiency or expand business scale. In his view, such companies not only have the financial capacity to continue investing in AI infrastructure, but also have greater opportunities to obtain actual returns from related investments.

The recent debate in the market about an AI bubble therefore involves not only how much infrastructure has actually been built, but also which companies will ultimately capture the commercial returns.

Wall Street is divided on AI's outlook

Chiavarone's views contrast with Griffin's remarks on Tuesday this week. Speaking at a Citadel Securities conference in New York, Griffin said investors can now be almost certain that they are in an AI bubble. He believes that major technological changes capable of profoundly transforming how the economy operates are often accompanied by excessive asset price increases and overvalued business models.

However, Griffin did not deny the long-term value of AI itself. Citing Amazon during the internet bubble era as an example, he pointed out that even if the industry undergoes a severe correction, companies with lasting competitive advantages can still emerge.

Unlike Griffin, Deutsche Bank CEO Christian Sewing said on Wednesday this week that he does not believe the AI industry is in an imminent, broad-based bubble about to burst, though investment losses or operational setbacks at individual companies can still occur.

BlackRock CEO Larry Fink offered another perspective from the angle of infrastructure supply. He believes that a slowdown in data center construction could keep computing power persistently scarce, thereby driving up computing costs and affecting the economic growth opportunities brought by AI technology.

These differing judgments reflect that the focus of Wall Street's debate is no longer just whether AI can create long-term value, but also the pace of investment expansion, companies' financing capacity, and whether future returns can cover current massive costs.

After a three-to-four-year construction cycle, will the real test arrive?

Chiavarone compared the current AI infrastructure buildout to the historical construction of the U.S. highway system and Europe's post-World War II reconstruction. He believes that large-scale infrastructure investment typically goes through different stages: during construction, it creates jobs and economic activity; after the project is completed, related investment demand may temporarily slow; subsequently, new technologies and infrastructure change the way work is done and drive the formation of new industries.

In his view, the economic adjustment brought by AI may follow a similar process, only with potentially more significant impact.

Regarding U.S. political risk, Chiavarone is also relatively optimistic. He believes the upcoming November midterm elections are unlikely to fundamentally change the policy direction of U.S. AI infrastructure construction. His reasoning is that regardless of how control of Congress changes, major political forces will find it difficult to secure majorities sufficient to easily overcome Senate procedural rules or override a presidential veto. He further pointed out that by 2029, when the next U.S. president takes office, many of the data center facilities currently planned may already be built or nearing completion, so the impact of subsequent policy changes on existing construction projects may be limited.

However, he did not completely rule out the possibility that the AI investment cycle could eventually undergo a correction. On the contrary, Chiavarone believes that as data center construction gradually completes, the sustainability of corporate capital expenditure, the profitability of AI models, and whether infrastructure investment can generate expected returns will become more important market questions going forward. His core judgment is that the investment cycle for AI infrastructure construction is not yet over, and investors' attention to future excess capital expenditure and bubble risk may need to distinguish between the current financing divergence and the industry-wide adjustment several years from now.