The AI boom may contain a bubble. That does not mean the infrastructure boom is a mistake. On September 30, Forbes published a profile of Clayco founder Bob Clark under a headline that was hard to miss: A Billionaire Data Center Builder Insists There Is No AI Bubble.
Bob Clark has earned the right to have an opinion on the subject.
Clayco completed 16 data centers last year. Forbes reports that it has another 77 in the pipeline across 13 states, and that data centers are expected to account for roughly 75 percent of the company’s nearly $12 billion in 2026 revenue. Bob Clark has been building data centers since the 1980s. He has watched the industry grow from computer rooms tucked behind banks into multibillion-dollar campuses containing hundreds of thousands of processors.
Asked about the possibility of an AI bubble, Bob Clark told Forbes that he sees no warning signs on the horizon. His reasoning is straightforward: this is not 1999. The companies buying this capacity are not flimsy dot-com startups burning through venture capital with no revenue. They are Amazon , Microsoft , Oracle , Google , Meta and other enormously capitalized companies.
Bob Clark put it even more simply: He is not worried about getting paid.
He may be right.
But I think that answers only one of the questions hiding inside the word “bubble.”
And that distinction matters more than you know.
Because the argument over whether artificial intelligence is a bubble has become so binary that we are in danger of learning the wrong lesson no matter which side wins.
Either AI is the next industrial revolution and therefore every dollar being spent must be rational.
Or AI is a speculative frenzy and therefore the entire infrastructure buildout is doomed.
Neither conclusion follows from the premise.
The truth is much messier.
And we have seen this before.
Two Questions Hiding Inside One Word
Earlier this year, while writing The Death of a BANANA Republic, I devoted an entire chapter to the bubble argument.
I tried to make the strongest version of it I could.
There are legitimate questions about capital spending. Legitimate questions about valuations. Legitimate questions about circular financing between chip companies, AI labs, cloud providers and the investors funding all three. Legitimate questions about how quickly expensive GPUs depreciate. Legitimate questions about whether projected productivity gains will arrive quickly enough to justify what is being spent.
Those are not tinfoil-hat questions.
They are financial questions.
And that is where the debate starts going sideways.
There are really two questions hiding inside the word bubble:
Is the technology real, and is demand for it durable?
And:
Is every company, financing structure, valuation and capital expenditure being attached to that technology economically rational?
Those are completely different questions.
You can answer yes to the first and no to the second.
In fact, I do.
Artificial intelligence is already embedded in fraud detection, cybersecurity, logistics, manufacturing, medical systems, software development, defense, entertainment and thousands of applications that have nothing to do with asking a chatbot to write an email.
The demand is real.
That does not mean every company selling AI will survive.
It does not mean every startup deserves its valuation.
It does not mean every GPU cluster being financed today will produce the return modeled in somebody’s spreadsheet.
It does not mean every data center currently proposed should be built.
And it certainly does not mean communities should accept every promise made to them simply because somebody attached the letters “AI” to a project.
Transformative technologies can produce terrible investments.
History is full of them.
The Dot-Com Bubble Was Real
This is where people often oversimplify the comparison to the late 1990s.
The dot-com bubble was not imaginary.
It was not something invented afterward by people who failed to understand the internet.
The skeptics were right about an enormous amount.
Companies with laughable business models received ridiculous valuations. Telecommunications firms loaded themselves with debt. Capacity was financed ahead of demand. Accounting games hid weaknesses. Investors convinced themselves that the normal rules of economics had somehow been repealed because the internet was different.
Then the music stopped.
Trillions of dollars in market value disappeared almost overnight.
Companies failed.
People lost retirement savings.
Entire sectors became toxic to investors.
The fact that the internet later transformed the world did not magically refund the people who bought the wrong stock at the wrong price.
The lesson of the dot-com crash is not that bubbles are harmless.
They are not.
But there is another part of the story we tend to forget.
The fiber stayed.
What the Fiber Knew
During the telecommunications boom, companies laid enormous amounts of fiber-optic cable.
Some of it was built far ahead of demand.
Some of the companies financing it disappeared.
Some of the investors who paid for it lost everything.
But when the stock certificates became worthless, nobody climbed into the ground and pulled the glass back out.
The fiber remained, purchased for pennies on the dollar by companies like Cogent Communications that understood the infrastructure might be worth far more than the bankrupt businesses that built it.
It sat there.
And waited.
It did not know that YouTube was coming.
It did not know Netflix would become a streaming company.
It did not know that cloud computing would move enormous portions of the corporate world out of local server rooms.
It did not know smartphones were coming.
It did not know Zoom was coming.
It did not know we would eventually train artificial intelligence models across giant distributed computing systems.
It did not know which companies would survive the crash.
It did not care.
It was infrastructure.
And eventually we invented enough things to consume it.
That is the part of the dot-com story that deserves far more attention today.
The bubble destroyed enormous amounts of financial wealth.
It did not destroy the infrastructure the bubble had financed.
Much of that infrastructure ultimately became part of the backbone of the economy that followed.
That pattern is older than the internet.
Railroad booms repeatedly bankrupted investors while leaving railroads behind.
Canal investors lost fortunes while the canals continued moving freight.
History does not tell us that overbuilding is free.
It tells us that the financial value of an investment and the strategic value of the infrastructure it creates are not the same thing.
That distinction is everything.
Different Scorecards
An investor and a nation do not evaluate infrastructure using the same scorecard.
An investor asks:
Did I make money?
A builder asks:
Will I get paid?
A hyperscaler asks:
Will we need the compute?
A utility asks:
Will this load actually materialize, and who pays for the capacity required to serve it?
A community asks:
What do we gain, what do we risk, and who is holding the bag if the assumptions are wrong?
And a nation asks:
When the smoke clears, where does the infrastructure exist, who owns it, and under whose laws does it operate?
Those questions overlap, but each one deserves its own answer.
This is why I think Bob Clark can be completely sincere when he says he is not worried about getting paid and still not have answered the bubble question.
From Clayco ’s perspective, creditworthiness matters enormously.
If Microsoft signs a contract, Microsoft is not Pets.com.
Fair enough.
But the fact that a contractor will be paid does not tell us whether an investor will receive an adequate return.
It does not tell us whether the customer ultimately needed every megawatt it reserved.
It does not tell us whether a utility built too much capacity too early.
It does not tell us whether a municipal incentive agreement properly allocated risk.
It does not tell us whether the first owner of the asset will be the owner who ultimately captures its value.
Those are separate questions.
And, interestingly, the same Forbes article illustrates the point.
Project Jupiter Is the Nuance
Forbes also discusses Project Jupiter, the enormous Oracle-linked data center project in New Mexico.
Regulatory delays affecting air, water and pipeline permits have complicated the project. According to Forbes, Oracle issued a force majeure notice indicating that if delays pushed completion beyond 2028, rent payments could also be delayed. Investors subsequently marked roughly $18 billion in bank loans associated with the project down to about 89 cents on the dollar.
That does not mean Project Jupiter is doomed.
It does not mean Oracle cannot pay.
It does not mean AI demand is fake.
It does not mean the facility will never operate.
It means infrastructure finance is complicated.
That is the point.
The building can be necessary.
The customer can be legitimate.
The technology can be transformative.
The contractor can be paid.
And somebody in the capital stack can go bankrupt.
Those things are not mutually exclusive.
Calling the entire phenomenon either “a bubble” or “not a bubble” flattens all of that complexity into a slogan.
We should be doing better than slogans.
Some Things Will Become Stranded Assets
There is another mistake infrastructure advocates need to avoid.
We cannot simply wave at the fiber story and declare that everything being built today will eventually prove valuable.
It will not.
Some projects will fail.
Some sites are poorly located.
Some facilities may be overbuilt for a customer that never arrives.
Some electrical upgrades may be structured in ways that expose ordinary ratepayers to risks they should never have been asked to absorb.
Some municipal incentive agreements will prove embarrassingly generous.
Some developers will fail.
Some AI companies will disappear.
And computing equipment is not fiber.
A GPU is a rapidly depreciating piece of hardware. It will become economically obsolete long before the building around it does.
Even the building itself may not be infinitely adaptable.
A highly specialized data center constructed in the wrong location, with limited access to future power and poor connectivity, can absolutely become a stranded asset.
So no, “the fiber stayed” is not an excuse for reckless development.
It is an argument for understanding what part of the project retains value under the downside scenario.
That is a very different proposition.
The Question Communities Should Be Asking
This matters enormously at the local level.
Over the past year I have spent a great deal of time talking with communities, developers, infrastructure professionals and public officials about data centers.
One thing has become increasingly clear to me:
We are asking local governments to make judgments they are not equipped to make and should never have to make.
A city council should not be expected to determine whether NVIDIA is overvalued.
A county commissioner should not have to forecast OpenAI’s revenue five years from now.
A planning commission should not be forced to decide whether artificial general intelligence arrives in 2029, 2035 or never.
Goldman Sachs does not know the answer with certainty.
Silicon Valley does not know.
Neither do I.
Fortunately, a community does not need to answer those questions.
It needs to answer different ones.
If the project fails, who carries the downside?
If the projected electrical load does not materialize, who pays for the transmission and generation built to serve it?
If the developer leaves, what financial guarantees remain?
If the customer changes, can the facility serve another one?
If the water infrastructure is expanded, does that improvement benefit the broader community?
Are tax incentives tied to actual investment, actual employment and actual performance?
Are ratepayers protected from costs created by a single large customer?
Who owns the substation?
Who owns the transmission upgrades?
Who owns the land?
Can the site be repurposed?
What happens if construction stops halfway through?
What happens if the tenant disappears ten years into a thirty-year agreement?
And perhaps most importantly:
If every optimistic projection is wrong, what is the community, utility, and developer left with?
That is a much better question than:
Is AI a bubble?
Show the Math
This is where the bubble debate connects directly to the argument I have been making since The Cloud Has Hit the Ground and The Death of a BANANA Republic.
I am not arguing that every data center deserves approval.
I am not arguing that every transmission line should be built wherever a developer wants it.
I am not arguing that communities should simply trust enormous corporations because they have large balance sheets.
I am arguing for something much less exciting:
Show the math.
Show the water.
Show the power.
Show who pays.
Show who benefits.
Show the infrastructure agreement.
Show the guarantees.
Show the downside case.
Show what happens if the projections are wrong.
Demand the same standard of evidence from the developer, the opposition, the utility and the government.
Then make a decision.
That is what responsible infrastructure development looks like.
Certainty does not mean guaranteeing that every prediction will be correct.
Nobody can do that.
Certainty means everybody understands the rules, the obligations and the consequences before the concrete starts pouring.
Good developers should want that.
Good communities should demand it.
The Bubble May Be in the Financing, Not the Future
There is real financial engineering layered on top of real technological demand.
That distinction is uncomfortable because people prefer clean narratives.
Either everything is transformational or everything is fraud.
Reality rarely cooperates.
We may discover that certain AI companies were wildly overvalued.
We may discover that some financing structures only worked as long as capital was cheap and everybody involved assumed the next funding round would arrive.
We may discover that depreciation assumptions on computing hardware were too generous.
We may discover that projected revenue arrives slower than projected capital expenditure.
We may discover that some data center announcements were little more than land speculation with a power request attached.
If so, those corrections will hurt.
But none of them automatically answer the larger infrastructure question.
The technology can be real while the valuation is wrong.
The demand can be durable while the financing is fragile.
The data center can be useful while its original owner goes bankrupt.
The transmission line can be valuable long after the first business plan attached to it has disappeared.
The substation does not read earnings reports.
The fiber does not check the Nasdaq before carrying a packet.
Concrete does not care what price somebody paid for the stock.
The Efficiency Argument Is Not as Simple as It Sounds
There is another version of the bubble argument that deserves serious treatment.
AI is becoming more efficient.
Models are getting smaller.
Inference costs are falling.
New architectures are doing more with less compute.
Therefore, the argument goes, we will not need all of these data centers.
Maybe.
But history gives us reason to be cautious with that conclusion.
When computing became cheaper, we did not buy fewer computers.
We put one on every desk.
Then we put one in every pocket.
Then we put dozens inside every modern automobile.
When bandwidth became cheaper, society did not consume less bandwidth.
We invented streaming video.
Then HD.
Then 4K.
Then cloud gaming, video calls, social media feeds and applications that would have been economically absurd when bandwidth was expensive.
Economists have understood this phenomenon since William Stanley Jevons observed in the nineteenth century that improvements in the efficiency of coal use did not reduce coal consumption. They expanded the number of economically viable uses for it.
Cheaper intelligence may do the same thing.
If AI inference becomes ten times cheaper, companies may not spend one-tenth as much.
They may deploy AI in one hundred times as many places.
That does not guarantee infinite demand.
Nothing does.
But efficiency alone does not prove overcapacity.
Sometimes efficiency is what creates the next wave of demand.
And Then There Is the Rest of the World
There is one more reason I am reluctant to frame this discussion purely as an American stock-market question.
Other countries are making infrastructure decisions on different time horizons, with different financial structures and different definitions of strategic value.
They are not all waiting for Wall Street to determine the correct terminal multiple for an AI company before deciding whether power generation, transmission, semiconductor capacity and compute infrastructure matter.
That does not mean the United States should copy authoritarian industrial policy.
It does mean we should recognize that our competitors do not necessarily share our ability, or our willingness, to stop building when markets become frightened.
A financial correction can slow American capital expenditure dramatically.
It does not necessarily slow everybody equally.
That matters.
If we respond to every period of market uncertainty by refusing to construct strategic infrastructure until every financial question has been resolved, we may eventually discover that somebody else was perfectly willing to build while we waited.
The infrastructure of the AI age will exist.
The legitimate debate is over how much, where, when, under what terms, at whose risk, and under whose rules.
Bob Clark May Be Right
Which brings me back to Bob Clark.
He may be right.
The hyperscalers are not the dot-com startups of 1999.
There is enormous underlying demand.
Clayco may have more work than it can physically execute for years.
The checks may clear.
But “I will get paid” is not a complete economic theory of the AI infrastructure boom.
It is one piece of a much larger picture.
The better question is not:
Is AI a bubble?
The better questions are:
Where is the speculation?
Where is the durable demand?
Who owns the risk?
What retains value if the optimistic case fails?
And what infrastructure do we want standing when the financial cycle eventually turns?
Those are harder questions.
They are also the ones that matter.
The Fiber Already Taught Us This
Twenty-five years ago, investors built infrastructure for a future that arrived later, and differently, than they expected.
Many of them were financially destroyed.
The future arrived anyway.
The fiber never knew what Amazon's stock price would be.
It never knew which telecom carrier would survive.
It never knew what application would eventually justify the capacity.
It just knew how to carry light.
That is why I keep coming back to it.
The lesson of the dot-com crash was not that financial discipline does not matter.
It matters enormously.
The lesson was not that every infrastructure investment eventually works.
It does not.
The lesson was that we made a category error when we assumed the collapse of the companies financing the infrastructure meant the infrastructure itself had no future.
A financial bubble and an infrastructure transformation can exist at exactly the same time.
One can destroy fortunes.
The other can reshape civilization.
And sometimes they are built with the same money.
So, yes, we should question the valuations.
Question the financing.
Question the projections.
Question the incentives.
Question the power agreements.
Question the water numbers.
Question the promises being made to communities.
Question all of it.
But do not confuse skepticism with paralysis.
And do not assume that if the bubble pops, the future goes away with it.
The companies may come and go.
The investors may change.
The business models may be rewritten.
The equipment will be replaced.
But the power plants, transmission lines, substations, fiber routes, industrial sites and data centers we choose to build, or refuse to build, will shape what comes afterward.
The fiber already knows that story.
We should, too.
