Trillions "Wasted" on the AI Boom? Arthur Hayes Says Yes, and He's Betting on What Comes Next

Every big technology wave comes with a bill. Right now, the world is writing an enormous one for artificial intelligence. Data centers, chips, power plants, talent, and endless software projects are soaking up money at a pace that has never been seen before.

Investor Arthur Hayes thinks a huge share of that money is being thrown away. His claim is blunt: trillions are being "wasted" on the AI boom. But here is the part that matters more than the complaint. Hayes is not sitting on the sidelines. He says he is betting on what comes next, the phase of the AI story that arrives after the spending spree.

That single idea is worth unpacking carefully, because it shapes how the next several years of AI will actually play out. It also tells businesses, workers, and investors what they should be doing right now.

What "Waste" Really Means in an AI Boom

When people hear the word "wasted," they usually picture money set on fire. That is not quite what is happening here. In a gold rush, money is not destroyed. It is misdirected.

Some of the classic forms of waste in a technology boom look like this:

• Overbuilding: Too much capacity gets built at once, faster than real demand can absorb it.
• Duplication: Dozens of teams race to build the same thing, and most of them lose.
• Speculation: Capital chases the story rather than the revenue.
• Unproven use cases: Companies buy first and figure out the problem later.
• Talent inflation: Payrolls and prices climb faster than the value created.

Hayes's argument fits neatly into this pattern. The AI buildout is real, but the speed and scale of the spending have raced ahead of the evidence that every dollar will earn a return. When capital moves that fast, a lot of it lands in the wrong place.

The important nuance is that waste and progress can happen at the same time. The companies that overspent may still leave behind something valuable: faster chips, cheaper computing, better models, and a trained workforce. The money is lost, but the infrastructure and knowledge often survive. That is why booms are so painful for investors and so useful for everyone who comes later.

Why Smart Money Is Splitting in Two

There is no longer one AI trade. There are two, and they pull in opposite directions.

Camp one believes we are still early. In this view, the spending is not waste at all, it is the foundation of a new economy. Every dollar spent on compute and energy will be repaid many times over as AI spreads into every industry. For this camp, the risk is not spending too much. The risk is spending too little and falling behind.

Camp two believes the opposite. Spending has become a contest of ambition rather than a calculation of return. When everyone builds at the same time, capacity arrives all at once and prices collapse. This camp looks for the moment when the music stops. That is where Hayes sits.

Both camps can be right about different things. The first can be right about the long term. The second can be right about the next few years. A boom can be real and still be overpriced at the same time, and that gap is where fortunes are made and lost.

What makes Hayes's position interesting is that he is not simply calling the boom a bubble and walking away. He is calling out the waste and placing a bet on what comes after the clean-up. That is a very different strategy from being bearish. It is a bet on sequencing, not on failure.

What Comes Next After the Spending Wave

If you accept the idea that a large chunk of current AI spending will not pay off, the natural question is: what does the next phase look like?

History offers a consistent answer. After the first wave of infrastructure spending, three things tend to happen.

1. Prices fall, and access spreads

Overbuilding leads to excess capacity. Excess capacity leads to price cuts. Price cuts open the door to a much larger group of users who could never afford the technology in its early, expensive form. The winners in this phase are rarely the firms that built the most. They are the firms that took advantage of cheap supply.

2. The focus shifts from building to using

In the first phase, attention goes to the foundation: hardware, data centers, and models that keep getting bigger. In the second phase, attention moves to application. The interesting question stops being "how powerful is the model?" and starts being "what problem does it solve, and who pays for it?"

This is usually when real, durable businesses get built. The infrastructure boom is a means to an end. The end is a tool that people actually use every day.

3. Weak players consolidate or disappear

When money is cheap, everyone survives. When money tightens, only the efficient remain. This is painful, but it makes the remaining players stronger and the market healthier.

Hayes's bet on "what comes next" only makes sense if you believe this cycle plays out. That is the real thesis: the money being spent today is not a verdict on AI's future. It is a cost of getting there.

What This Means for the Future of AI

If the waste thesis is even partly right, the future of AI will look different from the version that dominates headlines today.

The era of bigger-at-any-cost ends. Efficiency becomes the top priority. Smaller, cheaper, faster models that do specific jobs well will matter more than giant systems that try to do everything. Cost per task, not raw capability, becomes the scoreboard.

Energy and physical limits move to center stage. The constraint on AI is no longer ideas. It is power, land, water, hardware supply, and skilled people. That means AI's future is as much an infrastructure and policy story as a software story.

Trust and safety become commercial issues, not just ethical ones. When spending is easy, buyers overlook reliability. When budgets are tight, every deployment has to justify itself. Tools that are accurate, auditable, and easy to monitor win the budget line.

The gap between leaders and followers narrows. Open models and falling prices make it harder to stay ahead on raw capability alone. Advantage shifts to whoever owns the data, the workflow, and the customer relationship.

Practical Implications for Businesses

What should a business leader actually do with this argument? A few concrete moves stand out.

• Demand a return, not a demo. Every AI project should have a measurable outcome attached to it. If you cannot say what improves and by how much, you are not deploying AI, you are buying a story.
• Do not build what you can rent. Unless AI is your core product, buying access to models will usually beat building your own. Save your capital for the parts of the problem only you can solve.
• Lock in flexibility, not lock-in. Design your systems so you can swap models and providers as prices fall. In a quickly commoditizing market, portability is a real asset.
• Own your data and your workflow. Models are becoming similar. Your proprietary data and your process knowledge are not. That is where durable advantage lives.
• Watch your energy and cost exposure. AI workloads can run up bills quietly. Track cost per user, per task, and per outcome.
• Prepare for volatility. If the correction Hayes describes arrives, cheap capital disappears fast. Companies with clear returns will keep their budgets. Companies with vague pilots will not.

The businesses that come out ahead will not be the ones that spent the most. They will be the ones that waited for prices to fall, then moved quickly while others were still recovering.

The Bigger Picture for Society

Waste at this scale is not only a market story. It has a human dimension.

Trillions of dollars of misdirected capital could have gone into healthcare, education, housing, or clean energy. That is a real cost, and it deserves honest accounting. At the same time, the tools that survive the boom may still deliver enormous public value, cheaper medical analysis, faster scientific research, wider access to expert knowledge.

There is also a jobs dimension. Booms pull people into fields that may not survive the shakeout. Workers who build and operate AI systems will be best protected if they focus on the skills that transfer: data work, systems thinking, problem definition, and the ability to judge when a model is wrong. Tool-specific skills age fast. Judgment does not.

Finally, there is the question of concentration. If AI's next phase is driven by whoever can afford the remaining infrastructure, the benefits may pool in a few hands. Public policy, open standards, and competitive pressure all matter here. Waste today does not automatically become fairness tomorrow.

What to Watch Next

Rather than betting on a headline, watch a few clear signals:

• Pricing trends. Falling prices for AI capability signal that the next phase has begun.
• The shift to applications. When the biggest AI stories are about products people use, not hardware being built, the cycle has turned.
• Capital discipline. Watch whether spending is justified by revenue or by ambition.
• Consolidation. Mergers and shutdowns are a sign the correction is working through the system.

Conclusion

Arthur Hayes's claim that trillions are being wasted on the AI boom is uncomfortable, but it is not a prediction of failure. It is a description of how technology waves have always worked. Money rushes in, some of it is lost, and the useful parts survive. What matters is where you stand when the dust settles.

The real insight is in the second half of his position: he is not betting against AI. He is betting on what comes next. That is the smarter way to think about this moment. The AI buildout will produce both enormous waste and enormous progress. The winners will be the ones who treat the current spending wave as a cost of learning rather than a promise of returns, and who are ready to move when the technology gets cheap, reliable, and ordinary.