AI Bubble: Repeat of the DotCom Mania?
Bappa Sinha
MICHAEL BURRY, the investor made famous by the movie The Big Short after correctly predicting the 2008 housing collapse, is in the news again. Nearly twenty years ago, he warned that the US mortgage boom, built on weak loans, creative accounting and financial engineering, was bound to collapse. The American establishment dismissed him until the crash made his bets against the housing boom both immensely profitable and impossible to ignore. Now he has turned his attention to the artificial intelligence sector. His public comments and market positions indicate that he sees the same signs of speculative excess. But one does not need to rely on Burry alone. Across the technology industry, financial press, and independent analysts, concerns are surfacing about whether the American AI boom rests on sustainable foundations.
It is worth stating at the outset that AI is a transformational technology. It has already begun to reshape services, production, science, logistics and knowledge work. Even as we write, Google has released its latest Gemini 3.0 model, which demonstrates clear progress in reasoning, multimodal processing and efficiency. No serious observer denies that AI will become a central part of future economies. The issue is not the technology. The issue is the way the current US-led AI build-out is being financed and organised, and whether the promises being made by American corporations are grounded in economic reality.
First, there is the question of returns. The United States is witnessing an unprecedented surge in AI-related capital expenditure. Hyperscalers like Microsoft, Amazon, Google and Oracle are spending hundreds of billions of dollars each year on data centres, power infrastructure and GPUs. Investment banks project that total AI capex will cross three trillion dollars by 2028. Yet at the same time, the core cloud business, which funds most of this spending, is slowing sharply. Growth rates for Amazon Web Services, Google Cloud and Microsoft Azure have fallen significantly from their peak. AI companies themselves have limited revenue, high burn rates and uncertain business models. Much of their demand for compute is simply training ever-larger models in the hope that scale alone will deliver the breakthroughs required to justify these investments.
Second, there is growing discussion about accounting practices in the technology sector. For years, server hardware was depreciated over three or four years. As spending shifted toward expensive AI accelerators and specialised chips, many companies extended that period to five or even seven years. This reduces current expenses on paper and raises reported earnings at a time when actual cash outflows are rising. The result is a picture of improving profitability even when real returns have not materialised. Independent analysts estimate that this practice has inflated profits or cut losses on paper by vast sums. Whether this is manipulation or simply optimistic projection, the outcome is the same. Capital is being allocated on the basis of exaggerated numbers.
Third, the financial structure of the AI ecosystem is unusually circular. Nvidia sells GPUs based on demand from AI labs like OpenAI and Anthropic. Those labs raise money on the basis of their access to GPUs. Hyperscalers invest in these labs because they generate demand for their cloud platforms. Venture capital pours money into startups whose first major spending item is also compute. Nvidia, in turn, invests in these very firms, ensuring future demand for its chips. This is not a healthy industrial chain. It is a closed loop where valuation, spending and hype reinforce one another. The system functions only as long as money continues to flow in at increasing rates.
Fourth, the entire build-out is being financed through vast quantities of debt. US firms are using opaque special-purpose entities, private credit deals and off-balance-sheet structures to raise funds for data centre construction. Even cash-rich firms prefer high-cost borrowing to preserve their liquidity for share buybacks. The Wall Street Journal reports that interest rates on many of these loans are far above standard corporate levels. This resembles the leverage-on-leverage structure seen before during the 2008 housing crisis. These debts will be serviced only if AI revenues rise rapidly. If it does not, the consequences will extend well beyond the technology sector to banks, bond markets and pension funds.
A further pressure point is energy. Training and deploying large AI models require enormous amounts of electricity. The United States is already facing grid shortages in many regions. Data centre expansion is running ahead of transmission capacity. Energy sector experts warn of rising power costs and localised shortages. In contrast, China’s massive growth in renewable energy capacity and rapid grid modernisation place it in a far better position to meet the increasing energy demands of AI. The Chinese state has coordinated investments in solar, wind, hydropower and ultra-high-voltage transmission in a way that the fragmented US system simply cannot match.
The contrast between the American and Chinese approaches to AI is becoming clearer by the day. The dominant US narrative over the past few years has been centred on the pursuit of AGI, human-like intelligence - the so-called “holy grail” of artificial intelligence. The belief was that simply scaling up model size would unlock new emergent capabilities and deliver a decisive global advantage. On this basis, American AI labs raised billions of dollars. Their focus was less on cost, efficiency or deployment, and more on the rush to reach AGI. This approach aligned with the US venture capital model, which aims to create monopolies through early and massive funding, as seen in the cases of Google, Microsoft, Amazon, Apple and Meta.
China has taken a different path. Its firms have focused on efficient engineering, open source development and cost discipline. Projects like DeepSeek have shown that high-performance models can be built at a fraction of the cost of their American counterparts. Chinese labs have emphasised distribution and real-world integration rather than speculative leaps toward AGI. As a result, they have produced multiple models that are already near competitive with the US offerings at roughly one-tenth the cost. This has made them highly attractive across the global South and even among AI startups in the West.
The situation has reached a point where a partner at the leading technology Venture Capital firm, Andreessen Horowitz, admitted publicly that nearly eighty percent of AI startups pitching to them for funding are running on Chinese open source models. This is a remarkable indicator of where innovation and cost efficiency are now located. If this trend continues, China will become the natural supplier of AI infrastructure to much of the world, while the United States is left with heavily indebted firms whose returns depend on capturing a global market they may no longer dominate. Even within the US market, only existing tech monopolies like Google may corner the market, leaving many AI labs and hyperscalers with investments they cannot recover.
Like earlier bubbles, the AI boom has generated real physical investment. Data centres, power plants, transmission lines and semiconductor fabrication systems are being built at high speed. These investments currently sustain the US economy. Economists note that a significant share of recent US GDP growth is driven by AI-related construction, capital goods and market valuation effects. This creates a one-sided economic structure. When the bubble bursts, the firms that built these facilities and the banks that financed them will face enormous losses. The impact will spread across local economies, energy utilities, credit markets and international investors who have poured money into the US, leading to global economic shocks similar to those witnessed during the bursting of past bubbles.
The parallels with the dot-com era are clear. The internet was a real and transformative technology. But the financial claims made during the late 1990s far exceeded what the technology could deliver at the time. When the crash came, trillions of dollars in paper wealth evaporated. The same pattern appeared during the railway mania of the nineteenth century. The technology survived. The speculative structures collapsed. Under capitalism, this cycle repeats because investment is directed toward short-term profit rather than long-term social need.
AI will reshape our economies and societies. But left to the logic of speculative finance, it risks deepening inequality, destabilising economies and concentrating control in a handful of private corporations. The emerging AI bubble is not a verdict on the technology. It is a verdict on the economic model under which it is being developed. The question is not if the bubble bursts, but when it will burst. When it does, the impact on the global economy and the working people across the capitalist world will be significant.
