As the recent technology stock sell-off continued in anticipation of Nvidia’s (NASDAQ: NVDA) critical May 20 earnings, the legendary ‘Big Short’ trader Michael Burry took to X to issue a stark warning about capital overconcentration in the artificial intelligence (AI) sector.
Specifically, the famous investor noted how, by most measurements, investments are excessively exposed to AI even relative to the levels seen at the fever pitch of the Dot-com bubble.
Citing Torsten Slok of Apollo, Michael Burry noted that 87% of venture capital funding is directed at AI, while, in 1999, the share stood below 40%.
Simultaneously, the ‘Big Short’ trader noted that 38% of high-yield bond and as much as 49% of investment-grade debt issuance is linked to the same sector.
Perhaps most notably, Burry highlighted that such a setup creates a severe risk, considering that by 2002, some $100 billion of investment-grade bonds – generally considered high-quality and safe – issued during the Dot-com turned into junk.
Therefore, the legendary investor highlighted that, in addition to potentially dangerously exposing venture capitalists, the prevailing AI activity puts vehicles such as pension funds at risk.
Burry claims that high-yield bond share ‘belies the idea that today’s AI debt issuance is clean.’
Additionally, Burry noted the relative similarity between the issuance of high-yield debt – signaling relative uncertainty in the underlying businesses – during the 2026 stage of the AI boom and its share in the wider tech-media-telecom (TMT) bubble in 1999.
The legendary investor thus concluded that the market appears to contradict the argument that the ongoing technological revolution cannot be compared to the Dot-com due to the fact that many of the companies involved have actual and successful operations in addition to AI.
Indeed, 2026 has seen a notable decline in confidence in the much-touted technological revolution as, on the one hand, capital expenditure (CapEx) appears to be only increasing in size and, on the other, actual revenue tends to be heavily obfuscated in reporting, making the actual return on investment (ROI) unknown.
Furthermore, the concentration of compute demand – an important source of AI-linked revenue for much of big tech – in the hands of OpenAI and Anthropic adds another layer of worry, considering that the two private companies are heavily subsidized by the very companies they are providing revenue to and remain unprofitable.
Why AI infrastructure is not like Dot-com infrastructure
Lastly, another long-standing argument claiming that the CapEx would, ultimately, be worth it precisely due to the similarities with the Dot-com era appears dubious at best.
Specifically, numerous observers noted that while the infrastructure laid at the turn of the century eventually became extraordinarily useful, drawing a direct line of comparison to the ongoing buildout.
The issue with the argument is that GPUs and CPUs turn obsolete far sooner than fiber-optic cables, which, along with the likely hardware advancements and software changes, means that equipment purchased in the mid-2020s will either be utilized quickly or depreciate to uselessness in a handful of years.
While the problem of rapid obsolescence does not translate to the energy infrastructure needed to support the planned data centers, it offers little comfort for the bulls, given that, on the one hand, most countries are long overdue on upgrading their networks and, on the other, relatively few facilities appear to actually get completed even close to the original timetables.
Featured image via Shutterstock
