Michael Burry has taken his warning on artificial intelligence a step further, using a 145-year-old case study to question whether the technology industry’s focus on large language models is taking AI in the wrong direction. In his latest post, titled “History Rhymes: Large Language Models Off to a Bad Start?”, Burry looked back at an 1880 New York Times story about Melville Ballard, a deaf and mute teacher who described having complex thoughts before he had learned formal language.

“This 145-year-old case study, presented at the Smithsonian Institute no less, provides a potentially devastating critique of today’s Large Language Models and the spending behind them,” Burry wrote. The post comes after Burry’s warning on Monday that he was “moving timelines up” because he believed “the bubble in AI may burst sooner than later.” He said he was “more confident than ever before” that his bearish view would “play out over the next year.”

Burry has been warning that the AI boom is being driven by huge levels of investment, debt and financial commitments. He goes beyond the valuation of AI companies. Burry is questioning the basic architecture behind today’s AI systems.

Burry says AI may have started with the wrong building block

Burry asks if thought can exist without language? The question appeared in an 1880 paper presented by Professor Samuel Porter of the National Deaf-Mute College at Kendall Green. The paper was titled “Is There Thought Without Language? Case of a Deaf Mute.”

ALSO READ

The case involved Melville Ballard, a teacher at the Columbia Institute for the Instruction of the Deaf and Dumb. Ballard had communicated with his family using signs and gestures when he was young. Before he learned written language, he was already asking questions about the world and its origins.

According to the account, Ballard asked himself, “How came the world into being?” He wondered about the origin of human life, the Earth, the Sun, the Moon and the stars. After seeing a large tree stump, he asked, “Is it possible that the first man that ever came into the world rose out of that stump?” He then reasoned through the idea himself. The stump, he understood, had once been part of a tree. The tree had grown from the ground. This made the idea that the first human had emerged from the stump seem absurd.

The account said Ballard was about five when he began to understand the relationship between parents and children. At around eight or nine, he started asking questions about the origin of the universe. He did this before he had acquired formal language.

The question that stayed with Ballard

Ballard did not stop asking questions after learning more about the world. After he learned about the creation of the universe by a ruling Spirit, he began asking where the Creator himself came from. The account said he asked, “Shall we ever know the nature of God and comprehend His infinity after we enter His kingdom?”

He had spent years trying to understand things that he could not yet express through conventional language. Burry uses this story to make his central argument about AI. “By putting language before the capacity for reason, we are not building intelligence; we are building an increasingly sophisticated mirror,” he wrote. His claim is that reasoning and understanding should come before language, rather than emerging from the ability to process language.

What the 1880 presentation said about reason

Professor Porter argued that human reasoning could not simply be treated as a more advanced version of the abilities found in animals. Porter wrote, “That by which we understand all things must be essentially superior to anything else that is understood by it.” Burry also added another observation from the presentation.

ALSO READ

Ballard used signs, gestures and facial expressions to communicate with the audience. His meaning was understood even without conventional spoken language. Burry says that language without the capacity for reason fails at Understanding. He added, “Only with Capacity for Reason does Language unlock Understanding.”

Burry’s criticism of large language models

Burry argues that large language models have taken the opposite route. Instead of developing a capacity for reasoning first and then using language to express it, today’s systems begin with language. “By putting language first, LLMs build a primitive form of reason purely through logical inference,” Burry wrote.

He adds that this type of reasoning has limitations and can produce hallucinations when systems reach the edges of their knowledge. “The capacity for reason never existed. Therefore, language cannot scale through reason to understanding,” he wrote. Burry believes the distinction is important because a system can become increasingly capable of producing language without necessarily developing the kind of understanding humans possess.

“The Large Language Model, as well the Small Language Model, lives permanently in the middle,” he wrote. He also stated that it can simulate reason, but without true capacity for reason, without eyes, without understanding.”

Burry proposes a different test for AI

Burry calls his idea “Ballard’s Test”. He is saying that an entity does not possess capacity for understanding until reason is demonstrated in the absence of language.” His claim is that if a system can demonstrate genuine reasoning without relying on language, there would be stronger evidence that it possesses a deeper capacity for understanding.

Burry says the AI industry originally tried to build systems around reasoning, but that approach was not realised. The industry then moved towards language because it was easier to develop and scale. “This is a known flaw, a bad start,” he wrote. This, according to Burry, has helped create what he calls a “parameter trap”.

The “parameter trap”

Burry argues that the industry has responded to the limitations of language models by increasing their size and computing power This ‘bad start’ has led to a ‘parameter trap,’ where brute-force language processing powered by zillions of power-hungry chips has become an incredibly ironic bottleneck,” he wrote. His criticism is not limited to the technology. It is closely connected to the amount of money being invested in AI infrastructure.

Burry has repeatedly argued that companies are committing enormous amounts of capital on the assumption that AI demand and returns will continue to grow. He believes that if the underlying technology does not develop as expected, the spending could become a problem.

Burry says AI is part of another capital cycle

In his recent warnings, Burry has described the AI boom as part of a larger “capital cycle”. His claim is that companies tend to invest heavily when they see strong demand and attractive returns. As more companies enter the same area, investment can eventually create more capacity than actual demand requires.

Burry pointed to net investment by S&P 500 companies as a share of US nominal GDP. According to his calculation, the figure was around 2.07% as of June 30, its highest level in nearly four decades outside the period around the dot-com boom.

Burry sees a similarity with the late 1990s, when technology companies continued investing heavily before the Nasdaq peaked in March 2000. He says that spending does not necessarily stop when market conditions begin to deteriorate.

Companies can continue building infrastructure because they have already committed to projects, equipment and contracts. Burry believes the current AI cycle could continue creating financial risks even if the market begins to weaken.

Trillions of dollars are tied to AI infrastructure

Burry has also focused on the size of the commitments being made by the biggest technology companies. Microsoft, Amazon, Alphabet, Meta and Oracle have made major commitments linked to AI infrastructure. These include purchases, leases, construction projects and other arrangements.

According to a recent report cited by Burry, Alphabet, Amazon, Meta, Microsoft, Oracle, Nvidia, Broadcom, SpaceX and Advanced Micro Devices had disclosed around $3 trillion in commitments largely connected to AI infrastructure.

The commitments reportedly include around $1.2 trillion in leases that have not yet started and about $1.9 trillion in purchase commitments involving chips, energy and data centre infrastructure. This figure is significantly larger than the roughly $600 billion in capital expenditure reported by these companies over their latest 12-month periods.

Burry’s concern is what happens if AI demand does not grow fast enough to justify that spending. Many of these agreements cannot simply be cancelled. If demand weakens, companies could still be required to pay for data centres, chips and other infrastructure even if those assets generate lower returns than expected.

Burry has also questioned how technology companies account for the useful life of AI chips and servers. He has argued that some large cloud companies are treating Nvidia chips and other AI equipment as having useful lives of five or six years. Burry believes their economic life could be closer to two or three years. The difference matters because companies record depreciation expenses over the expected useful life of an asset.

A shorter useful life would mean companies recognising higher depreciation costs earlier. Burry has estimated that the difference could reduce reported depreciation expenses by around $176 billion between 2026 and 2028.

He has described the accounting practice as “one of the more common frauds of the modern era.” Burry has estimated that, if his assumptions prove correct, Oracle’s 2028 earnings could be overstated by 26.9%, while Meta’s could be overstated by 20.8%. These are Burry’s estimates and remain his interpretation of the accounting treatment rather than an established finding against the companies.

Burry sees a three-way squeeze

Burry’s concern is that several problems could hit AI companies at the same time. First, AI demand could grow more slowly than expected. Second, higher depreciation could put pressure on earnings as companies begin recognising the cost of their infrastructure. Third, companies with large debts or contractual commitments could face higher financing costs.

Burry has described this as a three-way “compression”. The basic idea is that weaker demand could reduce returns, higher expenses could hurt earnings, and higher financing costs could add another burden. This could become particularly important for companies that have committed large amounts of money to AI infrastructure before knowing exactly how much revenue that infrastructure will generate.

Burry has compared today’s AI cycle with the technology, media and telecom boom of the late 1990s. He believes heavy investment can continue even after the market reaches a peak. The same pattern, in his view, could emerge with AI. Companies may continue spending because they have already committed capital, even if the returns from that spending start to weaken.

Burry’s Monday post brought this argument into sharper focus. He wrote that he was “moving timelines up” because “the bubble in AI may burst sooner than later.” He also said he was “more confident than ever before” that his bearish view would “play out over the next year.”

Burry says the spending may keep rising

He expects companies to continue investing heavily in the near term. “I have little doubt the next few quarters will set still higher and higher net investment/GDP marks,” he wrote. This is central to his capital-cycle argument.

More spending can make the AI boom appear stronger in the short term. It can also create more infrastructure and capacity. The question Burry is raising is whether the eventual demand and cash flows will be large enough to support all that investment.

Burry wants AI to move towards reasoning

Burry believes the answer may lie in changing how AI systems are designed. He pointed to Google’s AlphaGeometry and Meta’s Coconut as examples of work moving towards what he calls a “reason-first” architecture. He argues that such systems are moving towards an idea that was already discussed in the 1880 Smithsonian presentation.

“Language is the output of understanding, not the engine of reason,” Burry wrote. He also claimed that the future could involve “compression”, using “System 2” reasoning to work with the relatively limited set of questions humans actually ask instead of continuously processing enormous amounts of information.

According to Burry, this could reduce the computing resources required by current AI systems. He contrasted this approach with the idea of AI models continuously talking to one another and generating more and more language. He called that approach, “a directionless waste of resources made impossible by lack of a basis in economic realities.”

The second 1880 story was about speculation

Burry’s post did not stop with the case study about thought and language. The same June 19, 1880 edition of the New York Times contained another story that caught his attention. It was titled “San Francisco’s Wealth, A Population of Bonanza Speculators.” Burry saw a connection between that story and today’s Silicon Valley. He quoted the French saying, “the more things change, the more they stay the same.” The 1880 article described San Francisco as a place where people were deeply interested in speculation.

It said the city had experienced a stock “boom” that attracted people with relatively small amounts of capital. The article described how people could buy small interests in mines and participate in the speculative boom. Burry believes the similarities with today’s technology investment cycle are worth considering. The old newspaper article described a population driven by the hope of getting rich from another big opportunity.

It said, “Almost the entire population is imbued with the passion for speculation.” The article warned that if another major opportunity emerged, stocks could once again rise to “absurd figures”. It also warned that San Francisco could again move from a period of prosperity into another period of losses. Burry connected that warning with his own concerns about Silicon Valley. He wrote, “Whether Silicon Valley will ‘again pass through the period of flush times to again suffer,’ as she has repeatedly, or improbably break the pattern, I hope you found this helpful.”

Burry’s larger warning

Burry’s latest argument therefore has two parts. One is technological. He believes the AI industry may have built its current generation of systems around language before solving the deeper problem of reasoning and understanding. The other is financial. He believes companies are committing enormous amounts of capital based on expectations that AI demand will continue growing rapidly.

The two arguments are connected. If current language models cannot develop the kind of reasoning and understanding Burry expects from true AI, he believes the huge infrastructure buildout could eventually prove difficult to justify.

Disclaimer: This article provides factual analysis only and is not, and should not be construed as, an offer, solicitation, or recommendation to buy or sell securities. Investors must conduct their own independent due diligence and seek advice from a registered financial advisor in the respective jurisdiction.