The artificial intelligence boom has moved far beyond software demos and chatbot excitement. It has become one of the largest infrastructure buildouts of the modern technology era, reshaping stock markets, corporate capital spending, energy demand, semiconductor supply chains and investor psychology.
Across the United States and global markets, investors have treated AI as the next great productivity revolution. Chipmakers, cloud giants, data-centre operators, memory suppliers, power infrastructure firms and networking companies have all become part of the “AI trade.” Reuters recently described the rally as one that has helped push equity markets to record highs, with Nvidia alone rising more than 1,300% since the end of 2022.
But the Bank for International Settlements, often called the central bank for central banks, is now asking a harder question: what happens if the promised returns from AI infrastructure do not arrive fast enough?
In its 2026 Annual Economic Report, the BIS warned that the sustainability of the AI boom has become one of the pressure points facing the global economy. The institution said AI optimism has supported economic activity and kept financial conditions favourable, but it also cautioned that intense competition, supply bottlenecks and increasingly leveraged financing could make the boom vulnerable.
“Optimism surrounding AI may not last, despite its promise of future productivity gains.”
The warning does not mean the BIS believes AI is fake or useless. In fact, the central bank forum acknowledges the long-term productivity potential of artificial intelligence. The concern is different: real technologies can still attract too much capital too quickly. Railways, canals, telecom networks and the internet all changed the world — but many investors who funded the early infrastructure waves still lost money when spending ran ahead of near-term cash flows.
That is the tension now gripping Wall Street.
The new AI infrastructure race
The current boom is driven by a simple belief: the future of AI will belong to companies that control the most computing power, the most advanced chips, the largest data centres and the strongest cloud platforms.
That belief has triggered a massive capital expenditure race among hyperscalers — the largest cloud and technology companies building the physical backbone of AI. These firms are spending on graphics processing units, high-bandwidth memory, servers, networking equipment, cooling systems, power supply, land, data-centre campuses and long-term energy contracts.
Reuters reported that chipmaker stocks staged a record 75% rally in the second quarter of 2026, helped by another wave of rising capital expenditure forecasts from hyperscalers racing to build AI infrastructure. The same report said forecast AI capital spending by the five biggest hyperscalers is approaching $1 trillion this year, while Goldman Sachs estimates cumulative spending could reach $7.6 trillion by 2031.
For investors, the logic is compelling. If AI becomes the next general-purpose technology — like electricity, the internet or cloud computing — then today’s spending could create decades of future profits. The companies selling the infrastructure may enjoy strong demand. The companies deploying AI may automate work, improve margins and create new revenue streams. The entire economy could benefit if productivity rises.
This is why investors have continued to buy the AI story even after sharp market swings. The most immediate winners are the so-called “picks and shovels” companies: semiconductor manufacturers, chip equipment makers, memory suppliers, cloud infrastructure providers and data-centre players. Their revenues are not imaginary; many are already benefiting from real orders, real shortages and real pricing power.
Reuters noted that some chipmaker valuations still look supported by earnings growth. Micron, for example, has seen its stock more than triple since March, but revenue estimates have also risen sharply, keeping its forward valuation relatively contained.
That is why the AI bubble debate is not straightforward. This is not a market built only on dreams. It is a market built on both genuine technological change and extremely aggressive expectations.
Why the BIS is worried
The BIS warning focuses on a classic financial cycle problem: when companies believe only a few winners will dominate a new technology, they may all spend heavily at the same time to avoid being left behind.
In AI, that competition is especially intense. No major cloud company wants to be seen as underinvesting. No chipmaker wants to miss the demand wave. No data-centre developer wants to be short of capacity. No investor wants to miss the next Nvidia.
The result is a self-reinforcing boom. Rising AI demand pushes up chip orders. Higher chip orders push up supplier revenues. Supplier revenues lift stock prices. Rising stock prices make investors more confident. Investor confidence lowers financing pressure. Easier financing encourages even more expansion.
But the BIS warns that this loop can reverse if returns disappoint.
“Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust.”
That line captures the central risk. The danger is not that AI disappears. The danger is that the investment cycle becomes too large relative to the monetisation cycle.
In other words, companies may build data centres faster than customers generate profitable AI use cases. They may sign long-term supply contracts for chips, memory and electricity before demand is fully proven. They may borrow heavily to expand capacity, only to find that pricing, utilisation or enterprise adoption does not justify the spending.
The BIS also highlighted supply bottlenecks in power generation, electricity grids and memory chips. These shortages can force companies to lock in long-term contracts to secure scarce resources. That protects them in the short run but increases overcommitment risk if demand later softens.
Debt-funded AI spending is the new pressure point
The AI boom is increasingly not just an equity market story. It is becoming a credit market story.
Reuters reported that the BIS sees AI financing as increasingly reliant on debt and complex funding structures across the supply chain. That matters because debt changes the nature of risk. A stock can fall without immediately threatening the financial system. But if heavily financed projects fail to generate cash flows, stress can spread through lenders, private credit funds, suppliers and contractors.
Data centres are capital-intensive. They require huge upfront investment before revenue arrives. They also depend on access to power, grid capacity, chips, cooling infrastructure and long-term customer demand. If AI demand keeps rising, these assets may become extremely valuable. If demand slows, some projects could become stranded or underutilised.
This is why the BIS is not only looking at technology stocks. It is looking at the financial plumbing behind the boom.
The concern is that hyperscalers, chip suppliers, data-centre developers, energy providers and private lenders are becoming more interconnected. If one part of the chain slows, stress may spread to others. Axios, summarising the BIS warning, noted that today’s AI buildout is unfolding through a concentrated ecosystem of hyperscalers, suppliers and private lenders linked by debt and opaque financing arrangements.
That does not guarantee a crisis. But it increases the number of channels through which disappointment can travel.
What this means for tech stocks
For technology investors, the BIS warning lands at a sensitive moment. AI-related stocks have delivered extraordinary gains, but market indicators are flashing caution.
Reuters reported that BofA Global Research’s Bubble Risk Indicator stood at 0.91 for the PHLX Semiconductor Sector and 0.82 for the Technology Select Sector, where 1 indicates extreme bubble-like price action. The same report said the Buffett Indicator, comparing total U.S. stock market value to GDP, reached 218% in the first quarter, just below the previous quarter’s record of 219%. The S&P 500 price-to-sales ratio stood at 3.22, far above its long-term average of 1.84.
These numbers do not prove a bubble. They do show that the market has become expensive and sensitive to disappointment.
The market has already shown signs of fragility. On June 23, Reuters reported that the Nasdaq and S&P 500 closed at more than one-week lows after a semiconductor selloff, with the Philadelphia Semiconductor Index falling 7.9%. The selloff was driven partly by concerns over growing debt-funded AI spending and expectations of a more hawkish Federal Reserve.
That is the key issue for investors: AI stocks do not need bad news to fall. They only need good news to become less good than expected.
If hyperscalers reduce capex guidance, semiconductor orders could weaken. If AI revenue grows slower than infrastructure spending, valuations may compress. If interest rates remain high, debt-funded data-centre expansion becomes more expensive. If power constraints delay projects, expected returns may be pushed further into the future.
The “AI trade” therefore depends on more than belief in the technology. It depends on timing, margins, financing costs, utilisation rates and customer willingness to pay.
Is this another dotcom bubble?
The comparison with the dotcom era is tempting but imperfect.
The dotcom bubble was filled with companies that had little revenue, weak business models and speculative valuations. Today’s AI leaders include some of the most profitable companies in the world. The infrastructure demand is real. Cloud revenue is real. Chip shortages are real. Enterprise adoption is real, even if uneven.
However, history shows that real technology revolutions can still produce investment bubbles. The internet changed the world, but many internet stocks collapsed. Railways transformed economies, but railway manias still destroyed capital. Telecom networks became essential infrastructure, but the early-2000s fibre buildout still produced bankruptcies.
The BIS is pointing to this historical pattern. Transformative technologies attract capital before their full economic value is visible. Investors extrapolate future productivity gains into present valuations. Companies overbuild to win market share. Financing becomes easier. Then, if returns take longer than expected, the boom can turn into a bust.
The most balanced view is that AI may be both a real revolution and a partial bubble. The technology can be genuine while some valuations, projects and financing structures become excessive.
The uncomfortable question: who earns the return?
The biggest unresolved question is not whether AI will be useful. It is who will capture the profits.
Chipmakers have already captured a large share of the early value because the infrastructure race requires their products immediately. Cloud companies may capture value if enterprises pay enough for AI services. Software firms may benefit if AI features improve productivity and pricing power. But many companies spending heavily on AI still need to prove that the returns will exceed the enormous cost of compute.
This creates a divide between suppliers and spenders.
As JJ Kinahan of Cboe Global Markets put it in a Reuters-cited note, the companies selling the picks and shovels are in strong shape, while those buying them still have to prove that the billions they are spending are worth it.
That is the heart of the BIS warning. A boom becomes dangerous when the entire system assumes that future returns will justify today’s spending, but those returns remain uncertain.
The bottom line
The AI boom is not simply hype. It is backed by real spending, real infrastructure and real technological progress. But the size and speed of the investment wave have created financial risks that investors can no longer ignore.
The BIS is not calling the end of AI. It is warning that even powerful technologies can produce overinvestment, leverage and market instability when capital rushes in faster than returns can be proven.
For tech stocks, the message is clear: the next phase of the AI trade will be judged less by excitement and more by evidence. Investors will want to see whether AI spending converts into revenue, whether revenue converts into profit, and whether profit justifies the trillions being committed to the buildout.
Until then, the AI boom remains both a growth story and a stress test.
