Nervousness around AI stocks reached a boiling point last week. Michael Burry, the investor who became famous for predicting the 2008 mortgage crisis, placed short positions on Nvidia and Palantir.
Burry’s claim that big tech companies polish their profits with accounting tricks by keeping depreciation artificially low briefly shook the market. But is his analysis correct?
Burry’s point centers on the depreciation timelines of data centers. Where companies used to depreciate their AI hardware over four years, they now stretch this to five or six years. This saves billions in costs in the short term. It sounds like deception, but there is a legitimate reason: the older Nvidia A100 chips from 2020 are still running at full capacity in the cloud. Algorithmic efficiency improvements keep older hardware productive for longer. What seemed impossible four years ago with certain chips is now feasible thanks to smarter software.
At the beginning
The market overreacted. CoreWeave fell 16 percent after delaying a single data center. This panic reveals a fundamental misunderstanding of where we stand in the AI revolution. McKinsey estimates that only 10 to 15 percent of potential AI applications have been implemented so far.
Look at the education sector: it is virtually unchanged despite the transformative potential of personalized AI learning. Or the government sector: AI adoption in the public sector is minimal even though efficiency gains could be astronomical. Healthcare, which accounts for 10 to 20 percent of GDP in developed countries, is still at the threshold of AI integration because regulatory barriers are only now beginning to shift.
The shortage of AI talent remains severe. Companies are fighting over scarce experts, with salaries skyrocketing. In a true bubble peak, talent becomes abundant because everyone reskills. Now the opposite is happening: universities cannot keep up with the demand for AI-trained graduates.
The real revolution has yet to begin
We have not yet seen the “killer apps” of AI. ChatGPT is impressive but remains a tool. The real transformation will come when AI agents autonomously perform complex tasks—from running full marketing campaigns to managing supply chains. This “agentic AI” is still in its infancy.
The multimodal revolution—AI that seamlessly combines text, images, video, and audio—is just beginning. Imagine AI attending a meeting, taking notes, pulling up relevant documents, preparing a presentation, and handling the follow-up. The technology exists in parts, but the integration has yet to be completed.
Robotics with an AI brain is on the verge of breaking through. Boston Dynamics robots can already dance, but robots that autonomously function in factories, hospitals, or homes? Those are coming, but they are not here yet. This convergence could transform entire sectors.
Skeptics are wrong
The productivity paradox—AI is everywhere except in economic statistics—is often cited as bubble evidence. But this is historically normal. With electricity, it took thirty years before it showed up in productivity numbers because factories had to be completely redesigned. We are seeing the same with AI: business processes must be fundamentally rethought, not just automated.
Venture capital continues to flow in—278 billion dollar in 2024. In true bubble peaks, capital dries up. Now we see the opposite: investors afraid of missing the boat. The difference compared to 2000? Back then people invested in companies without a business model. Now they invest in companies building real technology that are still searching for optimal applications.
Brussels is working on AI legislation, Washington is still debating. Historically, bubbles often peak after regulation, when the rules are clear and the market matures. We are still in the wild west of AI.
Underrated factors
Scientific breakthroughs driven by AI are still missing. AlphaFold was impressive for protein folding, but the major breakthroughs in drug development, materials science, or nuclear fusion are still to come. Each breakthrough could trigger a new wave of investment.
New use cases are discovered weekly—from AI reading X-rays better than radiologists to AI analyzing legal contracts in seconds. In bubble peaks, applications are exhausted and companies desperately search for new markets. Now the opposite is true.
The geopolitical dimension guarantees long-term investment. The AI race between the US and China is existential. Both countries will continue investing regardless of market sentiment. This is not a normal market cycle but a technological arms race tied to national security.
Skepticism remains high, from academics warning about AI hype to CEOs who do not see the added value. This skepticism is healthy and typical of the middle phase of technological revolutions, not the end. Only when taxi drivers start giving AI stock tips should you worry.
The current nervousness mainly shows that we are in an in-between phase. The infrastructure has been built, the first applications are running, but the major societal impact is still ahead. With ChatGPT only two years old, corporate adoption still early, and transformative applications in development, we look more like year three of a ten-year revolution than the finish line. The question is not whether AI is a bubble, but how much growth is left before the inevitable correction comes. The arguments suggest: more than many think.
Han Dieperink is chief investment officer at Auréus Vermogensbeheer. Earlier in his career, he was chief investment officer at Rabobank and Schretlen & Co.