Arthur Hayes Bets AI Infrastructure Boom Will End in Crash, Bailout and Crypto Surge

Former BitMEX CEO Arthur Hayes is betting that the artificial intelligence boom will follow a familiar pattern in financial history: massive investment, excessive capacity, a sharp correction, and eventually a policy response that floods markets with liquidity, creating a powerful tailwind for bitcoin and other cryptocurrencies.

Hayes, co-founder and chief investment officer of crypto investment firm Maelstrom, said at the Gamma Prime Investing Conference in Singapore that the world is spending too much money on AI infrastructure and is ultimately heading toward an enormous surplus of computing capacity.

Humanity is "wasting multi-trillion dollars" on AI data centres, Hayes said in response to questions from CNBC.

His argument is not that AI will fail as a technology. Instead, he expects the infrastructure supporting it to become dramatically overbuilt as technology companies race to secure the computing power required to train and operate increasingly sophisticated models.

The result, in his view, would be a glut of computing power that eventually makes compute extraordinarily cheap.

The prediction puts Hayes against the prevailing investment narrative surrounding AI infrastructure, where companies are committing hundreds of billions of dollars to data centers, GPUs, networking equipment, power generation and related infrastructure in anticipation of rapidly growing demand.

Hayes believes that spending will eventually overshoot actual economic demand.

"The massive data center buildout would ultimately make computing power 'extremely cheap and extremely plentiful,'" he said.

His investment thesis then moves beyond the AI sector itself. Hayes expects the eventual correction to trigger a policy response similar to those seen after previous financial and technology booms, with governments and central banks providing liquidity to stabilize markets.

"If you study financial history and you study every single major technological rollout, it always is overbuilt. There always is a crash, and there always is a bailout," Hayes said.

For crypto investors, he believes that final stage could be more important than the initial AI boom.

"Thankfully, we have bitcoin and other crypto to soak up that excess liquidity, and so we know the asset that's going to perform the best when the bailout comes," Hayes said, adding that "you just have to be patient."

The Problem Hayes Sees in AI Economics

Hayes' argument hinges on a potential mismatch between the enormous capital being committed to AI infrastructure and the revenues ultimately generated by the companies using that capacity.

SpaceX, OpenAI and Anthropic are among the major end users driving demand for computing power, Hayes said, but he argued that none of those companies currently makes money.

That line of argument is considered valid because data-center operators can secure enormous commitments from AI companies without necessarily knowing whether those customers will generate enough cash flow to support the infrastructure being built on their behalf. Once the facilities under construction are completed, infrastructure providers will expect payment for the computing capacity that AI companies have committed to use.

Hayes expects the pressure to become more apparent around late 2027 or 2028, when much of the current wave of data-center capacity is expected to come online.

The risk is therefore one of timing as much as demand. AI companies may have an enormous appetite for compute today, but infrastructure projects require years of capital expenditure and long-term commitments. If model efficiency improves faster than expected, AI companies require fewer chips for the same amount of useful output, or revenue growth fails to keep pace with infrastructure spending, the market could find itself with more capacity than it needs.

That could put pressure on data-center operators, chip suppliers and investors that have priced in continued exponential growth.

Hayes' argument also contains an important caveat. He said the bullish scenario remains possible if AI becomes sufficiently useful over the next year to generate enough demand for AI companies to become profitable.

The bull case is that AI becomes "so useful" over the next 12 months that demand expands rapidly enough to change the economics of the industry, he said.

That would postpone or potentially weaken the overcapacity scenario.

Nvidia and Chipmakers Face A Valuation Question

Hayes is not arguing that every company benefiting from AI is unprofitable. He pointed to Nvidia and memory-chip manufacturers as examples of companies already generating substantial earnings from the AI investment boom. The question, he said, is whether investors are paying an appropriate valuation multiple for those companies based on their expected future earnings.

A company can be profitable and still be vulnerable to a sharp share-price correction if investors have already priced in years of extraordinary growth. The same logic applies to the broader AI infrastructure complex. Demand can remain strong while investment returns deteriorate if companies collectively build capacity faster than end-user revenues grow.

Hayes said he does not favor shorting AI companies, describing bets on falling prices as "not really a great investment opportunity." Instead, his strategy is to wait for the excesses of the cycle to work themselves out and position for what he expects to happen after the eventual correction.

His thesis is rooted in the belief that technological revolutions tend to attract more capital than the underlying economics can initially absorb. Railroads, telecommunications, the internet, and other major technology waves all produced periods of aggressive infrastructure investment followed by corrections.

The key difference this time is the potential scale of the AI buildout and the financial system surrounding it.

Why The AI Bust Could Become A Crypto Catalyst

Hayes' most unconventional prediction is that an AI infrastructure crash would ultimately be bullish for crypto. His reasoning follows a familiar liquidity cycle. If AI investment produces excess capacity and financial stress, policymakers could respond with monetary and fiscal support. If that response creates abundant liquidity, Hayes expects investors to search for assets capable of absorbing that liquidity.

He believes bitcoin and other cryptocurrencies could become major beneficiaries. That is effectively a bet on the aftermath rather than the boom itself.

If AI companies become highly profitable and demand for compute continues to surge, the infrastructure boom could persist for longer than Hayes expects. If investment eventually overshoots demand, however, Hayes believes the resulting downturn could create the conditions for another round of monetary stimulus.

The timing is therefore central to his thesis. He expects much of the current infrastructure expansion to become operational around 2027 and 2028, potentially creating the moment when investors can assess whether the enormous capital expenditure was justified by actual AI revenues.

Flop Is Hayes' Bet on Cheap Compute

Hayes is also building a crypto project around the second part of his thesis: that computing power will eventually become abundant and inexpensive. His new venture, Flop, is an AI-agent payments project expected to launch in the first quarter of 2027. The project aims to create a spot market for computing power in which participants provide GPUs and perform AI inference in exchange for Flop tokens.

Hayes sees the potential abundance of compute as a prerequisite for a much larger AI-agent economy. If AI agents become widespread, they will require continuous access to computing resources, creating demand for a payment mechanism that can operate between autonomous software systems and compute providers.

"There is currently no payments network for AI agents," Hayes said.

Flop is intended to create that market by connecting the currency used by AI agents directly to the resource they consume.

"If agents can convert a currency directly into compute, which is what they eat and consume, then they will use this currency," Hayes said. "That's our bet."

The project effectively links Hayes' two investment ideas. He expects the current AI infrastructure boom to produce an eventual surplus of computing capacity, while simultaneously betting that cheap compute will enable an expansion of autonomous AI agents that need to transact independently.

The contradiction is central to the thesis. Hayes expects today's infrastructure spending to be excessive from the perspective of near-term economics, but potentially transformative for the longer-term cost of computing.

That could mean the AI boom fails as an investment cycle without failing as a technology.