Steve Hanke has seen financial manias before. The Johns Hopkins University professor of applied economics, who served on President Reagan’s Council of Economic Advisers, is now issuing one of his starkest warnings yet: the artificial intelligence trade that has powered Wall Street’s rally over the past two years is a bubble, and it will pop — likely by 2026.
His argument gains unexpected reinforcement from an insider source: Yann LeCun, Meta’s chief AI scientist and a Turing Award winner, who has publicly stated that the current generation of large language models will not achieve the kind of artificial general intelligence that investors are pricing into the market. Together, these two voices — one from the world of macroeconomics, the other from the frontlines of AI research — are forming a potent counter-narrative to the prevailing euphoria on Wall Street.
A Bubble Built on Capital Expenditure and Faith
According to Business Insider, Hanke laid out his case in stark terms, pointing to the staggering capital expenditure commitments made by the so-called hyperscalers — Microsoft, Alphabet, Amazon, and Meta — which collectively plan to spend upwards of $300 billion on AI infrastructure in 2025 alone. Hanke argues that these investments are based on speculative assumptions about future AI revenues that have yet to materialize in any proportion remotely close to justifying the outlay.
“This is a classic bubble,” Hanke told Business Insider. He drew parallels to previous technology manias, including the dot-com bubble of the late 1990s, where massive capital was deployed on the promise of transformative technology long before sustainable business models emerged. The core of his thesis is straightforward: when investment spending dramatically outpaces actual revenue generation, a correction is inevitable. Hanke has suggested that the reckoning could arrive as early as 2026, when the gap between AI spending and AI-generated revenue becomes impossible for even the most bullish analysts to ignore.
Meta’s Own Chief Scientist Pours Cold Water on AGI Hopes
What makes Hanke’s warning particularly compelling is that it aligns with the technical assessment of Yann LeCun, who leads AI research at Meta — one of the very companies driving the spending boom. LeCun has been vocal in his criticism of the hype surrounding large language models (LLMs), the technology underpinning products like ChatGPT, Google’s Gemini, and Meta’s own Llama models. As reported by Business Insider, LeCun has argued that LLMs are fundamentally limited — they can generate plausible-sounding text but lack genuine understanding of the physical world, cannot reason reliably, and are nowhere close to achieving artificial general intelligence.
LeCun’s position is not that AI research is a dead end. Rather, he believes the current architectural approach — scaling up transformer-based models with ever-larger datasets and compute — will hit a wall. He has advocated for entirely new approaches to machine learning that incorporate world models and planning capabilities. But those breakthroughs, if they come, are years or even decades away. The implication for investors is sobering: the technology that hyperscalers are spending hundreds of billions of dollars to build out may represent an intermediate step, not the final destination that stock valuations currently assume.
The Revenue Gap That Haunts Silicon Valley
The numbers tell a troubling story. While companies like Nvidia have seen explosive revenue growth from selling the GPUs that power AI training and inference, the downstream economics remain murky. Microsoft has reported growing AI revenue through its Azure cloud platform and Copilot products, but the company has been vague about margins and the degree to which AI features are driving net new spending versus cannibalizing existing product lines. Alphabet has integrated AI across its search and cloud offerings, but CEO Sundar Pichai has acknowledged that monetization is still in early stages.
Amazon’s AWS division has seen AI-related bookings increase, yet the company’s overall operating margins have not expanded in a way that suggests AI is yet a major profit driver. Meta, for its part, has positioned its AI spending as essential to maintaining competitive relevance in advertising targeting and content recommendation — defensive investments as much as offensive ones. The question Hanke raises is whether the market has priced in a future where AI generates trillions in new economic value, when the present reality is far more modest.
Historical Echoes: From Railroads to Dot-Com to AI
Hanke’s bubble thesis draws on a well-established pattern in economic history. The railroad boom of the 1840s in Britain, the telegraph mania, the dot-com era — each involved a genuinely transformative technology that nonetheless produced a speculative bubble and painful bust before the technology’s real economic benefits materialized over a longer time horizon. The internet did indeed change everything, but not before the Nasdaq lost nearly 80% of its value between 2000 and 2002, wiping out trillions in paper wealth.
The parallel is instructive. Few serious observers doubt that AI will have significant long-term economic effects. The question is one of timing and valuation. If the market has priced stocks as though the AI transformation is largely complete or imminent, and the actual timeline is five to fifteen years, then a substantial repricing is likely. Hanke’s warning is not that AI is worthless — it is that the market has gotten ahead of the technology, and that the correction, when it comes, could be severe.
Wall Street’s Divided Camps
Not everyone agrees with the bearish assessment. Goldman Sachs and Morgan Stanley have both maintained bullish outlooks on AI-related equities, arguing that the productivity gains from AI adoption across industries will justify current valuations over a multi-year horizon. Nvidia CEO Jensen Huang has repeatedly described the current moment as the beginning of a new industrial revolution, with AI factories replacing traditional data centers as the foundational infrastructure of the global economy.
But even among bulls, there are signs of caution. Several Wall Street analysts have noted that the concentration of market gains in a handful of AI-related stocks — the so-called Magnificent Seven — creates fragility. If any single company misses earnings expectations or signals a slowdown in AI spending, the ripple effects could be outsized. The recent volatility in Nvidia’s stock price, which has swung by double-digit percentages on earnings days, underscores the market’s sensitivity to any shift in the AI narrative.
The Hanke-LeCun Convergence
What is remarkable about the current moment is the convergence of skepticism from two very different vantage points. Hanke approaches the question as an economist who studies money, credit, and asset prices. His concern is with the macroeconomic dynamics of overinvestment and the inevitable correction that follows. LeCun approaches it as a scientist who understands the technical limitations of the systems being deployed. Their conclusions, while arrived at through different methods, point in the same direction: the market’s expectations for AI are running well ahead of what the technology can deliver in the near term.
This convergence matters because it undermines the most common defense of AI valuations — the argument that “this time is different” because the technology is so powerful. When one of the world’s leading AI researchers says the current approach has fundamental limitations, and one of the world’s most experienced economists says the spending patterns look like a classic bubble, the burden of proof shifts to the bulls.
What Comes Next for Investors and the Industry
If Hanke’s timeline proves correct, the next 12 to 18 months could be pivotal. By mid-2026, the hyperscalers will have spent a cumulative sum approaching half a trillion dollars on AI infrastructure. Investors will demand to see commensurate returns. If those returns do not materialize — if AI revenues remain a fraction of AI capital expenditures — the market will be forced to recalibrate.
That recalibration need not be catastrophic. A measured repricing of AI stocks, combined with continued technological progress, could set the stage for a more sustainable growth trajectory. But history suggests that bubbles rarely deflate gently. The dot-com bust destroyed companies that had real technology and real users, simply because their valuations had become untethered from economic reality. The risk is that the same dynamic plays out in AI, punishing even well-run companies whose only sin was being overvalued by an exuberant market.
For now, the voices of Hanke and LeCun remain in the minority on Wall Street, where the dominant narrative is still one of unbounded AI potential. But as the economist himself might note, minority opinions in financial markets have a way of becoming majority opinions very quickly — usually at the worst possible moment for those who ignored them.