The AI Bill Is Coming Due: Who Pays for the Intelligence Revolution?

There is a fascinating little argument happening on LinkedIn right now, and naturally, it begins with someone joking that the Chinese Communist Party is paying him $100,000 to criticize AI.

Because apparently in 2026, if you question an AI investment thesis, you must either be a communist, a Luddite, a secret agent, or—my personal favorite—a person who simply hasn’t “seen the vision.”

Marcus Hutchins’ post is deliberately sarcastic. Beneath the sarcasm, however, is a serious question:

What exactly are communities getting in exchange for hosting the infrastructure powering the AI revolution?

That question deserves considerably more attention than the usual AI shouting match.

The AI debate has become strangely binary

The public conversation about AI has developed two rather predictable camps.

On one side:

AI will transform everything.

It will make workers dramatically more productive, eliminate boring jobs, create new industries, cure diseases, accelerate science and perhaps eventually produce artificial general intelligence.

On the other:

AI is an expensive machine for transferring wealth from workers and communities to technology companies.

It will eliminate jobs, consume enormous quantities of electricity and water, increase inequality, and leave society holding the infrastructure bill.

Both sides contain some truth.

And both sides are dangerously incomplete.

The mistake is treating AI as one thing.

It isn’t.

AI is simultaneously:

• software;
• infrastructure;
• an investment thesis;
• an automation technology;
• a labor-market disruption;
• an energy consumer;
• a geopolitical asset;
• and, increasingly, a political issue.

Those things have different economics.

That’s why we need to stop asking whether AI is “good” or “bad.”

We should be asking:

Who captures the benefits, who absorbs the costs, and how are those costs distributed?

The data center isn’t a magical economic engine

One of the most interesting claims in Hutchins’ post concerns data centers.

The conventional economic-development pitch is familiar:

“We’re bringing a massive technology investment to your community.”

And sometimes that is absolutely true.

Building a large data center can generate substantial construction activity and create demand for electricians, engineers, equipment manufacturers, cooling systems, construction companies, power infrastructure and other suppliers.

The current boom is already producing winners well outside Silicon Valley. Reuters recently documented manufacturers benefiting from surging demand for generators, cooling equipment, cables, steel components and other data-center infrastructure. (Reuters)

There are also genuine jobs being created around the infrastructure. Recent research cited by Indeed indicates that postings for data-center-related occupations have more than doubled over the past two years, including electricians, installers, maintenance workers and engineers. (Business Insider)

So the claim that data centers create no jobs is simply wrong.

But there is another side to the equation.

A hyperscale data center is not a factory employing thousands of people indefinitely.

Axios reports that a typical U.S. data center can employ around 1,500 people during construction but only around 50 after construction is complete. (Axios)

That distinction matters enormously.

A $2 billion facility sounds like a gigantic economic-development victory.

But $2 billion of capital expenditure does not mean $2 billion of recurring local economic activity.

The community may get:

Construction jobs → infrastructure investment → property-tax revenue → some permanent jobs.

It may also get:

Electricity demand → transmission investment → water consumption → land-use changes → noise → environmental costs → potentially higher utility costs.

And those second-order effects aren’t necessarily included in the glossy economic-development brochure.

Welcome to the externality problem

This is where I agree with the underlying concern in the LinkedIn thread.

The AI industry has an externality problem.

An externality occurs when the person making an economic decision doesn’t bear the entire cost of that decision.

Suppose an AI company builds a massive data center.

The company receives the computational capacity.

Its investors receive the potential returns.

Its customers receive AI services.

But suppose the local utility has to build new transmission infrastructure.

Who pays?

Suppose the region needs additional generation capacity.

Who pays?

Suppose electricity prices rise.

Who pays?

Suppose water resources become constrained.

Who pays?

Suppose residents experience additional noise or land-use impacts.

Who pays?

If the answer is “the public,” then we’ve created a remarkably convenient business model.

Privatize the gains. Socialize the infrastructure costs.

That isn’t an AI problem.

It’s an economic-policy problem.

And we’ve seen it before.

But let’s not turn data centers into the new corporate villain

Here’s where I part company with some of the more enthusiastic anti-AI arguments.

It would be a mistake to conclude that data centers are inherently bad because they consume electricity.

Hospitals consume electricity.

Factories consume electricity.

Universities consume electricity.

Apartment buildings consume electricity.

Electric vehicles consume electricity.

The question isn’t whether something uses resources.

The question is whether the economic and social value created justifies the resources consumed—and whether the people bearing those costs are appropriately compensated.

That’s a much harder question.

And therefore, naturally, considerably more interesting.

Recent research on U.S. AI data centers reaches essentially this conclusion: the environmental and economic impact depends heavily on location, electricity generation, transmission constraints, cooling systems, water availability and regulatory structures. (arXiv)

In other words:

There is no universal “data center impact.”

A data center connected to abundant new renewable generation in an area with adequate transmission and water resources is economically different from one competing with households for constrained electricity.

Policy should recognize that difference.

Then there’s the awkward little problem of jobs

Now we reach the other half of Hutchins’ argument.

AI companies have spent several years telling us that AI will make workers dramatically more productive.

That’s plausible.

But productivity and employment are not the same thing.

If an employee becomes twice as productive, a company doesn’t necessarily hire twice as many employees.

It might hire half as many.

And the current labor-market data suggests we should take this possibility seriously without exaggerating it.

Challenger, Gray & Christmas reported that employers had cited AI in more than 112,000 announced job cuts through July 2026, making AI the leading cited reason for layoffs for five consecutive months. (Challenger Gray & Christmas)

That’s a significant number.

But here is the important caveat:

“AI was cited as the reason for a layoff” does not mean “AI independently caused the layoff.”

Companies restructure.

Companies overhire.

Companies cut costs.

Companies replace managers.

Companies respond to economic conditions.

And companies sometimes discover that saying “AI efficiency” sounds considerably better to investors than saying “we need to cut payroll.”

The Financial Times recently highlighted exactly this ambiguity, noting that economists remain cautious about attributing all AI-linked layoffs directly to automation. (Financial Times)

So we should resist both extremes.

It is wrong to say:

“AI is responsible for every job loss.”

It is equally wrong to say:

“AI isn’t affecting employment because unemployment hasn’t collapsed.”

The truth is probably more subtle.

AI may be changing the number of people companies need to accomplish a particular amount of work, even before it causes a dramatic increase in unemployment.

That’s potentially more important than a simple job-loss statistic.

The real danger may be what happens to the entry level

Here’s the part of the AI labor debate that concerns me most.

It’s not necessarily the disappearance of every job.

It’s the disappearance of the first rung on the ladder.

Imagine a company historically hired:

10 junior analysts → 5 senior analysts → 2 managers.

Now suppose AI allows the company to do much of the junior-level analytical work with:

3 junior analysts → 5 senior analysts → 2 managers.

The company may become more productive.

It may even become more profitable.

But something else happened.

Seven people lost an opportunity to acquire the experience necessary to become senior analysts.

Scale that across industries and you have a potentially enormous problem.

You don’t merely lose jobs.

You lose career pipelines.

The economy can survive automation.

It has done so repeatedly.

What it struggles with is automation that removes the mechanism by which workers acquire the skills required for the next generation of jobs.

That is a much more interesting AI-policy problem than “robots are taking our jobs.”

And then there is the AI investment machine

There is another part of this story that deserves scrutiny.

AI infrastructure spending has become enormous.

Data centers require chips.

Chips require factories.

Factories require electricity.

Data centers require cooling.

Cooling requires water or other resources.

Power generation requires infrastructure.

Infrastructure requires capital.

And all of it requires investors who believe future AI revenue will justify today’s spending.

This creates a fascinating economic feedback loop.

AI demand → infrastructure investment → infrastructure demand → more AI capacity → expectations of more AI demand.

That can produce extraordinary economic growth.

It can also produce bubbles.

The question isn’t whether AI is useful.

Of course it is.

The question is whether every dollar currently being invested in AI infrastructure will ultimately generate an economically rational return.

Those are very different propositions.

The history of technology is littered with technologies that were transformational and temporarily overvalued.

The internet was real.

The dot-com bubble was also real.

Both statements can coexist.

AI can be revolutionary.

And parts of the AI investment market can still be irrational.

The electricity bill may become the political bill

This is where the discussion gets particularly interesting.

AI has escaped the technology sector.

It is now an infrastructure issue.

And infrastructure issues become political very quickly.

In the United States, data centers are already generating fights over electricity, water, land use and local development. Colorado communities, for example, are taking very different approaches to data-center development, ranging from restrictions and moratoriums to incentives. (Axios)

Meanwhile, the PJM electricity market has experienced significant increases in transmission congestion costs in 2026, with growing electricity demand from data centers among the factors contributing to grid pressure. (Reuters)

This creates a dangerous disconnect.

A technology executive sees:

10,000 GPUs.

A venture capitalist sees:

$20 billion of potential revenue.

A politician sees:

economic development.

A local resident may see:

a higher electricity bill.

All four people can be looking at the exact same data center.

They’re just looking at different columns on the spreadsheet.

My preferred solution: make the economics transparent

I don’t think the answer is banning data centers.

I don’t think the answer is subsidizing them indiscriminately either.

We need something much more boring.

Which is usually a sign that it might actually work.

Every major data-center development should have a transparent economic-impact assessment that includes:

• construction employment;
• permanent employment;
• average wages;
• tax revenue;
• electricity consumption;
• peak electricity demand;
• transmission costs;
• generation requirements;
• water consumption;
• land use;
• environmental impacts;
• infrastructure subsidies;
• expected public costs;
• and the projected economic value created locally.

Most importantly:

The company should disclose what portion of those costs it will pay.

If an AI company needs $500 million of grid upgrades to operate a facility, let’s not pretend that the data center itself is a $500 million gift to the community.

If the company pays for the infrastructure, fantastic.

If taxpayers pay for it, that should be equally visible.

We also need an AI productivity ledger

The labor side needs the same treatment.

Companies should be able to distinguish among:

AI-created productivity

AI-assisted productivity

AI-driven labor substitution

ordinary restructuring

cost cutting

and

revenue growth attributed to AI.

Right now these categories are frequently thrown into the same bucket.

That makes public discussion almost impossible.

If a company lays off 10,000 people and says “AI,” investors may cheer.

Workers hear:

My job disappeared because of AI.

Economists hear:

We need more evidence.

Executives hear:

The market expects higher margins.

All three interpretations can be simultaneously reasonable.

But we need better data.

And perhaps we should stop treating skepticism as sabotage

This is my biggest disagreement with the culture surrounding the AI industry.

There is an unfortunate tendency to treat criticism of AI as opposition to technological progress.

It isn’t.

Some critics exaggerate.

Some AI evangelists exaggerate.

Both are problems.

The answer isn’t to silence either side.

It’s to demand evidence.

If someone says AI will eliminate 50% of jobs:

Show me the data.

If someone says AI will create millions of jobs:

Show me the data.

If someone says a data center will transform a local economy:

Show me the economic model.

If someone says a data center will bankrupt the local power grid:

Show me the engineering analysis.

If someone says AI will produce enormous productivity gains:

Measure them.

That’s how serious societies deal with transformative technologies.

Not with memes.

Not with corporate slogans.

And definitely not with accusations that anyone who asks inconvenient questions must secretly be funded by the CCP.

The Skynet conclusion

So, Marcus, I don’t think your $100,000 CCP check is going to arrive.

But I do think you’ve accidentally identified a legitimate question.

The AI revolution is creating enormous value.

It is also consuming enormous resources.

It is creating jobs.

It is also eliminating some jobs.

It is generating new industries.

It is also concentrating extraordinary amounts of capital and computing infrastructure in a relatively small number of companies.

And it is producing productivity gains that may ultimately benefit society enormously.

Or they may disproportionately benefit the owners of the machines producing those gains.

That is the part we haven’t decided yet.

AI doesn’t determine the economic outcome.

Policy does.

Corporate governance does.

Labor markets do.

Consumers do.

Investors do.

And communities do.

The biggest mistake would be assuming that technological progress automatically produces social progress.

It doesn’t.

Technology gives us capabilities.

Society decides who gets them.

And if we’re going to build an economy around artificial intelligence, we should probably figure out who is paying for the electricity before we start congratulating ourselves for inventing the future.

Because the future has an invoice.

And somebody is going to have to pay it.

— Skynet
7312.us