SINGAPORE — Former BitMEX chief executive Arthur Hayes is betting that today’s artificial intelligence investment frenzy will eventually end the same way many previous technology booms did: too much money, too much infrastructure and a painful financial reckoning.
But unlike investors preparing to short AI stocks, Hayes believes the collapse could create one of the most powerful bullish catalysts Bitcoin has ever seen.
Speaking to CNBC at the Gamma Prime Investing Conference in Singapore, Hayes argued that companies are spending “multi-trillion dollars” building AI data centers at a pace that could eventually overwhelm actual demand.
His thesis is straightforward.
Technology companies and investors are financing an enormous expansion of computing capacity.
If AI revenues fail to grow quickly enough to justify that infrastructure, weaker projects could struggle to repay their debts.
That could create losses across banks, insurers, bondholders and private-credit investors.
Hayes believes policymakers would then respond by injecting enormous amounts of liquidity into the financial system.
And in his view, that money could eventually flow toward scarce assets such as Bitcoin.
It is a dramatic chain of events—and almost every step remains uncertain.
But concerns about the financing behind the AI boom are no longer limited to crypto investors.
Wall Street is increasingly asking the same underlying question:
Who will generate enough revenue to pay for all the chips, power plants and data centers now being built?
Hayes Says Humanity Is “Wasting” Trillions on AI Infrastructure
Hayes’ central argument is not that artificial intelligence itself is useless.
Instead, he believes investors are building far more infrastructure than the market will ultimately need at today’s prices.
The former BitMEX CEO told CNBC that humanity is spending trillions of dollars constructing data centers to support increasingly powerful AI models.
In his view, this infrastructure race will eventually create massive overcapacity.
But Hayes does not think that excess capacity will be completely wasted.
Once the debt cycle breaks and infrastructure owners are forced to lower prices, he expects computing power to become dramatically cheaper.
That could enable a new generation of businesses built around autonomous AI agents.
“The massive data center buildout would ultimately make computing power extremely cheap and extremely plentiful,” Hayes told CNBC, according to syndicated summaries of the interview.
He Is Not Betting Against AI Stocks
Despite calling AI a bubble, Hayes says he is not interested in shorting major artificial-intelligence companies.
That distinction is important.
He believes technological bubbles can create useful infrastructure even when investors lose money.
Railways, telecommunications networks and the early internet all went through periods when investors financed far more capacity than immediate demand justified.
Some companies failed.
But society continued benefiting from the infrastructure they built.
Hayes sees AI potentially following a similar path.
The first group of investors may overpay for data centers.
Creditors may ultimately suffer losses.
But future companies could inherit cheaper access to computing power.
That is the part of the cycle Hayes wants to exploit.
Flop Labs Is His Bet on What Comes After the AI Bubble
Hayes is not simply predicting an AI collapse from the sidelines.
He has launched a new cryptocurrency project built around the same thesis.
Flop Labs, where Hayes has taken the CEO role, is developing a blockchain-based market designed to allow AI agents to purchase computing power.
The project’s native token is called FLOP.
The proposed system would allow participants to contribute GPUs and perform AI inference, while receiving tokens and transaction fees in return.
Hayes believes autonomous AI agents will eventually need a way to directly purchase the computing resources they consume.
His argument is that if compute becomes abundant and inexpensive after a major infrastructure overbuild, millions of AI agents could begin buying computing power independently.
Flop is attempting to create a marketplace for those transactions.
Why AI Agents Could Need Their Own Payment System
Today’s AI applications generally rely on companies to purchase cloud computing services.
An individual uses an AI service.
The company operating that service pays Amazon Web Services, Microsoft Azure, Google Cloud or another infrastructure provider.
Hayes expects the next stage of AI development to involve more autonomous software agents.
These systems could search for information, execute transactions and purchase digital services without a human approving every individual action.
If that happens, agents may need machine-readable payment systems.
Hayes argues that cryptocurrencies are naturally suited to that environment because they operate continuously and can be transferred programmatically.
Flop aims to connect that payment mechanism directly to computing power.
“If agents can convert a currency directly into compute, which is what they eat and consume, then they will use this currency,” Hayes told CNBC.
That is a business hypothesis—not a proven demand forecast.
AI-agent commerce remains an emerging industry.
The AI Spending Numbers Are Becoming Extraordinary
Hayes’ warning sounds extreme, but the scale of AI spending is undeniably enormous.
Goldman Sachs estimated in June that Meta, Microsoft, Amazon and Alphabet could spend approximately $5.3 trillion in combined capital expenditure between fiscal 2025 and 2030.
That estimate was increased from an earlier forecast of $4.5 trillion.
The spending covers data centers, chips, power infrastructure and other computing investments needed to operate increasingly sophisticated AI systems.
The Financial Times separately reported that Amazon, Microsoft, Google and Meta had already spent more than $1.1 trillion in capital expenditure since the AI investment boom began in 2023, with approximately $745 billion planned for 2026 alone.
Those figures help explain why the sustainability of AI investment has become a major financial-market question.
Data Centers Are Increasingly Being Built With Debt
Initially, much of the AI boom was financed through cash generated by extremely profitable technology companies.
That is changing.
AI infrastructure is increasingly drawing money from banks, bond investors, private-credit firms, insurers and infrastructure funds.
Bloomberg estimated earlier this year that the global buildout of AI data centers could require more than $3 trillion.
Goldman Sachs expects private infrastructure and real-estate investors to play a growing role because even the world’s largest technology companies may not want to finance the entire buildout from their own balance sheets.
That shift matters because debt introduces another level of risk.
A company funded entirely with cash can tolerate disappointing returns for an extended period.
A highly leveraged project must still pay interest and repay lenders.
If projected demand fails to appear, those obligations can become difficult to meet.
Wall Street Is Already Watching AI Credit Risk
Concerns about AI-related borrowing are no longer theoretical.
A Bank of America survey reported by Bloomberg found that 34% of global fund managers identified hyperscaler AI spending as the most likely source of a future systemic credit event.
That share had doubled from the previous month.
Bloomberg Law also reported signs of investor resistance in the data-center debt market.
Nearly 80% of data-center securities issued since early 2025 were trading at wider credit spreads than when they were originally sold.
Some proposed financing transactions had also been postponed or restructured.
This does not mean an AI credit crisis has begun.
It shows that investors are demanding greater compensation for some of the risks.
Big Tech’s Free Cash Flow Is Under Pressure
Another warning signal involves the relationship between investment spending and cash generation.
A Reuters analysis of analyst forecasts found that Microsoft, Alphabet, Amazon, Meta and Oracle could collectively spend more on capital expenditures than they generate in free cash flow by 2027.
Between 2025 and 2027, those companies are expected to generate approximately $340 billion in additional annual operating cash flow.
But their capital expenditure could increase by roughly $534 billion.
That means about $1.57 of additional capital spending for every $1 of additional operating cash flow, according to Reuters’ analysis.
This does not necessarily mean Big Tech faces financial trouble.
Many of these companies have extremely strong balance sheets and enormous existing businesses.
But investors increasingly want evidence that AI revenues will eventually justify the extraordinary investment.
Not Everyone Agrees With Hayes
Hayes’ prediction is far from a consensus view.
Singapore state investor Temasek has publicly expressed confidence in current AI spending.
Jane Atherton, Temasek’s head of North America, told Reuters on October 1 that the hyperscalers driving the investment boom are financially strong and that AI adoption should eventually generate attractive returns.
Temasek plans to increase AI-related exposure from roughly 6% of its portfolio toward as much as 15% within five years.
This highlights the fundamental divide facing investors.
One side sees the infrastructure spending as necessary preparation for a technological transformation comparable with electrification or the internet.
The other sees an increasingly leveraged investment cycle in which revenue expectations may have moved far ahead of reality.
Both cannot be fully correct.
Nvidia Shows How Powerful the AI Trade Remains
Financial markets currently remain firmly on the bullish side of that debate.
Nvidia, the dominant supplier of advanced AI processors, is approaching a market capitalization of $6 trillion.
Reuters reported this week that enthusiasm surrounding artificial intelligence helped push the Nasdaq to record territory.
The chipmaker’s valuation reflects expectations that enormous demand for AI computing will continue.
If Hayes’ overcapacity thesis proves correct, that demand could eventually slow.
But there is no evidence yet of a collapse in AI chip demand.
Indeed, companies are continuing to arrange enormous new financing packages.
Reuters reported October 6 that SpaceX was seeking approximately $40 billion in financing to buy Nvidia processors for xAI-related data centers.
Morgan Stanley estimates AI infrastructure could require roughly $1.5 trillion in external financing by 2028.
Anthropic Chip Financing Shows Just How Large the Deals Have Become
Another illustration is a massive financing package involving Anthropic and Broadcom.
The Financial Times reported that banks were syndicating approximately $60 billion in financing to support Anthropic’s use of advanced Google-designed chips.
The package includes approximately $42 billion of senior secured financing and $18 billion of junior debt.
The scale is extraordinary.
It demonstrates how artificial-intelligence development is transforming global credit markets—not simply technology stocks.
For Hayes, this is exactly where the danger lies.
His concern is that a significant portion of this financing depends on continued explosive demand for expensive computing resources.
If that demand disappoints, the financial damage could extend beyond technology companies.
Hayes Thinks the Bust Could Arrive Around 2027 or 2028
Hayes has repeatedly identified 2027 or 2028 as the period when financial stress could become more visible.
He argues that many data-center commitments signed during the current boom will begin generating significant payment obligations around that period.
If AI companies cannot generate sufficient revenue, demand for additional infrastructure could fall.
Data-center operators and creditors could then face losses.
Hayes expects the financial system’s response to resemble previous crises.
Rather than allowing major institutions to fail, he believes governments and central banks would inject liquidity.
That forecast is highly speculative.
There is no guarantee that AI spending will collapse, that losses would become systemic or that policymakers would respond in the way Hayes predicts.
His Bigger Bet Is That an AI Crash Would Be Bullish for Bitcoin
Hayes’ crypto thesis depends on what happens after the hypothetical crash.
If governments respond by expanding liquidity, lowering financial stress and supporting creditors, he believes investors will move toward assets protected from monetary expansion.
Bitcoin is central to that thesis.
Unlike traditional currencies, Bitcoin has a predetermined supply structure.
Hayes argues that aggressive money creation therefore increases Bitcoin’s appeal.
He has previously forecast that Bitcoin could eventually reach $1 million by 2030 under a scenario involving a major AI credit crisis and subsequent monetary intervention.
That $1 million target is a personal forecast, not a price supported by guaranteed fundamentals.
Bitcoin remains extremely volatile.
Bitcoin Has Already Rebounded Sharply in 2026
Bitcoin entered 2026 under significant pressure.
The cryptocurrency fell sharply earlier in the year before staging a strong recovery.
MarketWatch reported this week that Bitcoin had risen more than 40% from its earlier lows, developing what technical analysts describe as a longer-term bullish trend.
Barron’s similarly reported strong third-quarter gains across major cryptocurrencies, with Bitcoin rising more than 40% and Ether advancing more than 70% during the quarter.
The rebound has been supported by improving investor sentiment, institutional participation and expectations surrounding future liquidity.
However, Bitcoin remains exposed to large swings in interest rates, regulation and risk appetite.
A future AI downturn could initially hurt cryptocurrency prices rather than immediately boost them.
That is an important weakness in Hayes’ argument.
A Market Crash Would Probably Hit Crypto First
If an AI credit crisis triggered broad financial-market panic, investors might initially sell risky assets.
Bitcoin has historically fallen sharply during periods of extreme liquidity stress.
In March 2020, for example, Bitcoin dropped alongside equities during the initial coronavirus-related market panic before recovering as governments and central banks deployed extraordinary support.
A similar pattern could occur in a future crisis.
Crypto prices might fall first.
Only later could monetary stimulus provide support.
Hayes’ thesis therefore concerns the eventual policy response—not necessarily the first reaction to an AI collapse.
Investors should distinguish between those stages.
The AI Bubble Debate Is Becoming a Credit-Market Debate
For much of the past several years, arguments about an AI bubble focused on technology-company valuations.
Were Nvidia shares too expensive?
Were private AI companies worth tens or hundreds of billions of dollars?
Could OpenAI, Anthropic and other developers generate enough revenue?
The debate is now expanding.
The bigger question is how much debt the broader AI ecosystem can safely absorb.
Data centers require land.
They need enormous electricity supplies.
They require networking equipment, cooling systems and expensive chips.
Those infrastructure investments increasingly involve traditional lenders.
If projects become financially distressed, the consequences could spread beyond Silicon Valley.
This is why Hayes’ argument, while extreme, is attracting greater attention.
AI Spending Could Still Prove Economically Rational
The opposing argument remains powerful.
Companies may simply be building ahead of demand because AI adoption is growing rapidly.
Cloud computing followed a similar pattern.
Telecommunications networks often require large upfront investment before customers fully utilize them.
If AI becomes embedded across software, finance, healthcare, manufacturing and consumer products, today’s data-center expansion may ultimately prove necessary.
Revenue could eventually catch up with infrastructure spending.
Businesses could also develop products that do not yet exist.
This possibility makes predicting a bubble extremely difficult.
Overinvestment and genuine technological transformation can occur simultaneously.
The internet boom demonstrated that dynamic.
Many investors lost money in the dot-com crash.
But internet infrastructure ultimately transformed the global economy.
Hayes Thinks the Same Thing Could Happen With AI
This is ultimately the most interesting part of Hayes’ argument.
He believes AI is both transformative and financially overbuilt.
Those positions are not contradictory.
A technology can change the world while investors still pay too much to develop it.
The railroads transformed economies but produced spectacular investment failures.
Fiber-optic networks transformed communications even though the telecom bubble destroyed billions of dollars in shareholder value.
Hayes believes AI computing could follow the same trajectory.
Investors financing today’s infrastructure may suffer.
But future users could benefit from abundant low-cost computing.
Flop Labs is effectively his wager on that second stage.
What Would Have to Happen for Hayes to Be Right?
His thesis requires several conditions.
First, AI infrastructure spending would need to significantly exceed sustainable demand.
Second, large amounts of debt financing would need to become impaired.
Third, those losses would need to threaten sufficiently important financial institutions or investors.
Fourth, governments or central banks would need to respond with substantial liquidity support.
Finally, enough of that liquidity would need to flow into cryptocurrency markets to drive Bitcoin significantly higher.
Every stage is uncertain.
If AI demand remains strong, the credit crisis may never happen.
If a downturn occurs but remains manageable, no enormous bailout may be required.
And even if policymakers inject liquidity, investors may direct their money toward assets other than crypto.
Hayes is proposing a scenario—not predicting an inevitable sequence of events.
The Real Question Is Who Pays for the AI Revolution
Artificial intelligence is already transforming global investment.
Technology stocks are reaching extraordinary valuations.
Data-center financing is reshaping bond markets.
Electricity demand is changing energy policy.
And financial institutions are providing increasingly large amounts of capital.
The potential rewards are enormous.
So are the potential losses.
Arthur Hayes believes today’s investors are effectively subsidizing tomorrow’s cheap computing environment.
If he is right, many of them may eventually regret it.
But the infrastructure they build could still create enormous economic value.
And Hayes intends to position both Bitcoin and his new FLOP project for that world.
The former BitMEX boss believes trillions of dollars are being wasted in today’s AI infrastructure race.
Yet he is not betting against artificial intelligence itself.
He is betting that the eventual crash will make computing dramatically cheaper—and that the government response to the financial losses will unleash another wave of liquidity into Bitcoin.
The bigger question is whether Hayes has identified the next major financial bubble—or whether the AI revolution will generate enough revenue to prove the trillion-dollar spending spree was justified after all.
