Cathie Wood: talk of an AI bubble is misleading investors

Talk of a bubble in the artificial intelligence market is most often reduced to parallels with the dotcom crash of the late 1990s. ARK Invest suggests assessing the situation not by quotations, but by the classic signs of a bubble: excess supply and lack of real demand. Cathie Wood, CEO of the investment company, Tom Stout, President of ARK, and Brett Winton, Chief Futurologist of the fund, discussed in the podcast the stage of development the industry is actually at and why in the new cycle a reliance on traditional protective assets can turn into "smoldering craters" in an investor's portfolio.

"If everyone is asking if it's a bubble or not - it's hardly a bubble," Winton says. The futurologist relies on audience dynamics. About a billion people now use chatbots - that's only 15% of smartphone owners. The potential, he estimates, is several times higher: by the end of the decade, the audience could grow to 4-5 billion. "The growth rate of the user base will accelerate because we are only at the very early stage of the engagement curve," he emphasizes. ARK's conclusion: in terms of penetration, AI is now closer to the Internet of the mid-1990s than to the peak of the dotcom boom.

This argument reinforces the state of infrastructure. Companies are already having to limit access to their data centers, and Winton describes the market itself as under supplied: capacity supply physically lags behind the demand for usage models. This structure is characteristic of the beginning of the cycle, not the end of it. Unlike the late 1990s, when IT companies built excess infrastructure to meet uncertain future demand, the high utilization and under-supply of capacity now indicate that technology is already mainstream.

ARK believes that the direct parallel with the dot-com crisis is misleading for investors. The key difference is the level of maturity of the technology: whereas in the late 90s the market evaluated a dream of the future Internet, now it evaluates solutions that are already in operation.

Cathie Wood emphasizes that the technological base necessary for the current leap was not in place 20 years ago. Cloud platforms became part of the infrastructure only after 2006, the breakthrough in deep neural networks occurred in 2012, and the Transformer architecture for data processing appeared only in 2017.

According to Winton, the current stage of AI development is closer to the mid-1990s: a phase of low penetration with a high growth trajectory. "We're closer to 1995 than 1999," summarizes the futurologist. ARK's main conclusion is that the AI market is developing not according to the dotcom model, where valuations outpaced reality, but according to the trajectory set by years of technology accumulation and its transition to mass application.

The differences with the dotcom era are not only about technology, but also about what exactly is put into the quotes. ARK does not deny that the current valuations of many companies look overvalued. But Wood emphasizes that the premium to the market that investors pay today will not last forever. ARK's baseline scenario assumes that within five years, the overvalued multiples will return to average values - the so-called compression will occur. For an asset to remain interesting, the growth rate of the business must be so high that it can cover this process with a margin. That is why the fund's internal criterion for entering a position is an expected return of at least 15% per annum over a five-year horizon: this figure already takes into account the risk of a decline in multiples.

Wood cites Palantir as an example. The company looks "very expensive" by current metrics, but 123% year-over-year revenue growth in the U.S. commercial segment justifies the premium. Winton summarizes this approach as, "A high multiple by itself means nothing if the scale of the market is changing faster than expected."

Brett Winton describes the AI market as a hierarchical structure where each level performs a different economic function.

At the base of this pyramid are the fundamental models that define the technical architecture (OpenAI, Anthropic, Google, xAI). This is the most capital-intensive layer: learning requires industrial-scale infrastructure. ARK estimates that the total potential of this segment could reach $15-20 trillion by 2030, reflecting both the cost of future services and global shifts in corporate costs.

Platforms are built on top of the base layer to turn raw models into application output. This is where the payoff for companies that can't rebuild the IT landscape from scratch comes in. Winton believes that the solvent demand at this layer is comparable to the market for the models themselves, as platforms (like Palantir) give businesses access to technology without huge capital investments. Closing the chain are vertical applications - solutions for specific industries, from logistics to medicine. According to the futurologist, this segment is growing fastest, and it is this segment that forms the primary demand that fuels the rest of the ecosystem.

However, the potential of AI is not limited to software. ARK emphasizes Embodied AI - the intersection of the digital and physical worlds: robotaxis, autonomous systems, and bioengineering - as a separate, even larger area. Cathie Wood cites the example of robotaxis: currently, revenue in this segment is less than $1 billion a year, but in 5-10 years it could grow to $8-10 trillion. In healthcare, AI can reduce the cost of drug development by about 75%, launching a new cycle of efficiency for pharmaceutical companies.

The potential of humanoid robots, according to Winton, surpasses almost all existing markets. The futurologist calls longevity technologies the next big goal: in a world of abundance, people will channel freed-up resources into life extension. It is this circuit - autonomous systems, robotics and biotechnology - that ARK estimates will provide the largest value gains in the coming decades.

The technological shift, which ARK describes as accelerating, poses serious threats to companies whose economies are built on last generation infrastructure. The pressures are already manifesting themselves in the transportation industry: the foundation estimates that the cost of autonomous trucking could fall to three cents per ton-mile, compared to four cents for rail. For businesses with extremely low margins, such a difference means that entire segments will lose competitiveness.

Cathie Wood emphasizes that other industries will be affected by similar shifts. Autonomous systems, robotization and algorithmic planning can redistribute profits within sectors much faster than traditional giants have time to restructure their processes.

Winton phrases this risk more starkly: investors' portfolios could be filled with "smoldering craters" - companies whose returns fall precipitously as new solutions set a higher standard of performance. ARK warns that in the face of such a technological shift, non-innovation assets risk declining in value by the end of the decade, even if the overall economy continues to grow.

This article was AI-translated and verified by a human editor