Bridgewater Warns: AI Boom Enters More Dangerous Phase

Bridgewater Associates, the world's largest hedge fund, is warning that the investment boom around artificial intelligence is entering a more dangerous phase. According to co-chief investment officer Greg Jensen, the buildout of AI infrastructure is still accelerating and responding less and less to traditional warning signals such as interest rates or stock valuations.

What Bridgewater is saying

In an analysis titled 'The AI Boom Has Reached a More Dangerous Phase,' Jensen points to four reasons why the current stage of AI buildout differs from the previous one. First, exponential growth in demand for computing power means the boom, which had mostly played out in the digital realm, now requires a massive physical component: factories, data centers, power grids.

Second, big tech companies financed the early phase of AI infrastructure buildout mostly from their own free cash flow. With capital expenditure now growing exponentially, ever larger sums are needed from outside sources, from investors, banks and the bond market. Third, company valuations are rising and now assume continued, if no longer exponential, profit growth. Fourth, the US economy is becoming increasingly dependent on the AI sector itself as a growth engine.

Capital for the early buildout phase came mainly from the big tech companies themselves, from their free cash flow - Greg Jensen, co-chief investment officer, Bridgewater Associates

Alphabet halts buybacks

The clearest sign of this shift is Alphabet's own behavior. In June 2026, Google's parent company carried out a stock and debt issuance worth roughly $85 billion to finance the buildout of data centers and Google Cloud infrastructure for 2026-2027. It was the largest operation of its kind in more than two decades and caught the market by surprise.

Even more telling is the fact that Alphabet skipped share buybacks this quarter for the first time in nearly ten years. By comparison, as recently as the first quarter of 2025 the company spent more than $15 billion on buybacks. Among the four largest AI investors, Alphabet, Microsoft, Meta and Amazon, only Microsoft carried out buybacks in the first quarter of 2026, and only $3.4 billion worth, the lowest figure in almost a decade.

History has seen this before

Analysts point to three earlier episodes in which a wave of share issuance by companies in a then-hot sector preceded a market crash. The first was the power and electrical grid buildout of nearly a century ago, which ended in the 1929 crash, panic and the Great Depression. The second was the 'Nifty Fifty' bubble of the 1970s, a group of beloved growth stocks such as IBM, Xerox and Polaroid considered infallible at the time.

The third was the dot-com bubble at the turn of the millennium, when companies such as Cisco, WorldCom and Global Crossing built network infrastructure financed by successive share issuances. The analysis shows that in four of the five historical scenarios described, S&P 500 returns turned out deeply negative, reaching nearly minus 43 percent within 24 months of the issuance wave in 1972, and dropping by close to 20 percent a year after the issuance peak in 2000.

New bottlenecks

According to Bridgewater, by 2026 the problem is no longer just capital, hard physical constraints have also emerged. The bottlenecks have shifted toward power grids, which are running up against aging infrastructure, as well as the availability of copper, cooling systems and memory chips. Leading manufacturers such as SK Hynix are reportedly already sold out of HBM memory for AI needs through 2027.

For Polish investors and companies, this means the cost of capital for further AI buildout will keep rising, and tech companies will increasingly turn to debt instead of cash from ongoing operations. AI-focused ETFs, which have grown sevenfold over the past year, could in this scenario be more exposed to swings than investors have so far assumed.

Bridgewater does not argue that the bubble must burst in the coming months, it stresses instead that the mechanisms that have so far shielded the boom from typical market shocks are weakening. As long as demand for computing power grows faster than the supply of energy, copper or memory, pressure on the biggest tech companies to keep taking on debt will persist.