AI's Fatal Flaw
AI bots can replace a worker but not a shopper
I’ve said for at least 20 years now that technology would “fix” the demographic-based labor shortage caused by the retirement of the Baby Boomers.
Centuries before the letters “AI” started popping up everywhere, technology was automating jobs away, just as it has been since the dawn of the Industrial Revolution.
Handloom weavers, spinning wheel operators, millstone operators, threshers, teamsters… All of them found their jobs redundant in the early stages of the Industrial Revolution.
Later, elevator operators, switchboard operators, bowling alley pinsetters and lamp lighters saw their jobs disappear. And I could write tomes on the number of jobs eliminated over the past quarter century thanks to the Internet.
AI promises to accelerate the evolution of the job market, of course. But the productivity enhancements of technology are nothing new.
An AI bot can replace a worker.
But you know what it can’t replace?
A shopper.
An AI bot isn’t going to buy a house, a car or even a cappuccino. And that’s a problem… because our entire economy ultimately revolves around spending.
Say’s Law — that supply creates its own demand — is a core tenet of classical economics. It also completely falls apart when you don’t have a constantly growing supply of new shoppers.
My friend Jeff Remsburg brought this up in his latest issue of the InvestorPlace Digest. I republished it here, edited for brevity.
Enjoy!
AI has a problem no one is discussing
By Jeff Remsburg, Editor of InvestorPlace Digest
I’m waiting for someone on Wall Street to highlight this, but so far, no one is touching it…
AI may be fantastic at replacing workers… but AI is terrible at replacing consumers.
It’s the paradox hiding beneath this groundbreaking technology. And when you follow the money far enough through the AI economy, it raises uncomfortable questions about the AI narrative…
Beyond profits for infrastructure builders, does AI math pencil out?
What do you pay for AI today?
For ChatGPT Plus, Claude Pro, Gemini Advanced, and so on, maybe $20 a month? Or maybe you don’t even pay.
Now, hyperscalers do not expect $20 chatbots to cover the cost of their trillion-dollar infrastructure investments. They expect the real payoff will come from enterprise AI adoption – corporate customers paying tens of thousands of dollars a month (or more) to integrate AI across their entire workflows.
But here’s issue #1…
So far, even when you include the early enterprise revenue, the revenues coming in are tiny compared to the capital going out.
From a different Axios article:
MIT researchers studied 300 public AI initiatives to try and suss out the “no hype reality” of AI’s impact on business…
95% of organizations found zero return despite enterprise investment of $30 billion to $40 billion into GenAI, the study says.
The “real” AI value proposition – the one after this infrastructure buildout phase – is murky at best. Revenue remains tame compared to the firehose of spending. So, unless AI monetization grows far beyond today’s numbers, the math becomes increasingly strained.
And remember, because a meaningful portion of the buildout is financed through private credit, any revenue disappointment doesn’t just affect hyperscaler earnings. It could feed back into the credit system, creating a domino effect of pain:
If AI spending slows, liquidity tightens…
If liquidity tightens, refinancing becomes harder…
If refinancing becomes harder, capex slows…
Lower capex leads to slower earnings growth…
Slower earnings growth drags down the market’s valuation multiples…
Lower market valuation multiples ding stock prices…
Lower stock prices create panic in Upper-K investors who sell, accelerating the unwinding of the wealth effect
I’m not predicting a collapse. I’m recognizing that a highly leveraged boom assumes highly reliable future cash flows – but those cash flows are not a certainty.
Now, let’s pivot to issue #2 – the deeper, more complex question that almost no one wants to confront…
The risk of excluding Lower-K Americans from the economy
As noted above, the hyperscalers don’t plan to make trillions of dollars from chatbots. Rather, they’re banking on million-dollar contracts with Fortune 500 companies.
But that thesis rests on some very large assumptions. Here are three:
The enterprise adoption will be enormous
The AI tools that companies buy will generate measurable, solid ROI
AI won’t be commoditized into a low-margin utility like cloud storage or broadband. [Note from Charles: This is exactly the scenario I expect.]
All possible – but not guaranteed.
As of today, outside of the companies on the receiving end of the hyperscaler firehose of money, most businesses experimenting with AI still can’t show a clean bottom-line return.
Yet even if those assumptions do play out exactly as the bulls expect, we run straight into an even bigger, more foundational issue – the one that surfaces when you follow the money one layer deeper:
Where do the enterprise customers get the revenue to pay the hyperscalers for all this AI?
I’m talking big picture – across the next decade.
When you trace that revenue pathway back to its origin, you land in the same place every time…
The U.S. consumer.
And that’s where today’s closed-loop economy may come back to haunt us.
What AI can’t do well – consume
The bullish narrative around AI’s enterprise value focuses on cost savings – and those savings are real. AI doesn’t demand raises. It doesn’t need health care. It doesn’t go on strike or take vacation…
From a corporate cost structure standpoint, AI is an efficiency dream.
But there’s a problematic flip side…
AI doesn’t buy anything.
In other words, AI doesn’t drive consumption – and consumption is still nearly 70% of U.S. GDP.
Source: Fed data
So, let’s follow a new potential doom loop:
If AI reduces the need for human workers in large enough numbers, we are implicitly reducing wage income for a significant portion of consumers…
If consumer income weakens, consumer spending weakens…
If consumer spending weakens, corporate revenues weaken…
If corporate revenues weaken, enterprise software budgets weaken…
And if enterprise budgets weaken, cloud and AI spending weakens – the very spending hyperscalers depend on to justify their infrastructure.
This ties into the “closed-loop economy” we’ve been discussing: the cycle in which AI boosts productivity, companies need fewer workers, profits rise, more money flows to AI, and even fewer workers are needed – all while the consumer base becomes less central to economic growth.
“But Jeff, I remember a recent Digest when you wrote that the top 10% of American earners were spending as much as the rest combined! You’re being inconsistent!”
That stat is true – but it doesn’t apply evenly across all categories of the economy.
Yes, the Upper-K consumer can support high-margin discretionary spending – travel, restaurants, even big-ticket items like homes and cars. In those areas, one wealthy household can spend as much as several middle-income households.
But there are enormous parts of the real economy where demand cannot be concentrated among the top 10%.
The wealthiest Americans can’t eat ten meals a day… or buy forty times the amount of deodorant, detergent, or toothpaste… or pay for thirty streaming subscriptions…
Bottom line: Wealth can concentrate – but consumption can’t.
Stepping back from the doom-and-gloom
Let’s be measured…
History gives us plenty of reasons for optimism.
Technological revolutions have a way of creating entirely new job categories no one could have predicted…
The top tier of U.S. consumers – the “Upper K,” as I often refer to them – has enormous spending power and continues to anchor demand…
Governments can soften transitions through transfers, taxes, and safety nets…
And the global middle class remains a massive and growing consumer base for U.S. companies.
But…
AI is different than past technological breakthroughs. Its ability to replace – well, just about all of us – creates a fundamentally different type of technology adoption curve. It’s one that shifts the balance between labor, capital, and consumption in ways we need to monitor carefully…
Wrapping up
The explosion of AI capex – much of it financed through private credit – is real, spectacular, and enormously profitable for the companies supplying the infrastructure.
And despite our recent jittery market, the next 12–18 months could continue to deliver outsized gains as the hyperscalers race to build the digital backbone of the next era.
But long-term profitability depends on something we seem to be forgetting…
A healthy, well-funded consumer.
Bottom line: AI can automate, optimize, and replace – but it cannot spend.
So, as our economy leans harder into automation and robotics, funded by massive spending and heavy private credit borrowing, perhaps it’s time we ask an uncomfortable question…
With AI, are we unwittingly sawing off the economic tree branch that we’re sitting on?
Have a good evening,
Jeff Remsburg