AI can help you work faster today. That part isn’t in doubt.
What is in doubt is whether those gains will turn into the kind of profit boom the market is already pricing into AI stocks, funding rounds, and data-center spending. In a Bloomberg Television discussion, Mark Cudmore argued that once you add debt, cross-holdings, and energy costs, the AI story looks less like a clean productivity trade and more like a crowded bet.
The case for AI productivity is strong, but the market may be pricing too much
The optimistic view starts with a fair point. Social media and the Internet of Things created huge excitement, yet neither produced the broad productivity surge that many investors once hoped for. AI may be different because it can cut time on real work, including writing, coding, research, customer support, and routine analysis.
Cudmore did not dismiss that. He said there is a productivity boom from AI. His concern was the scale of the boom that markets now assume. Investors are pricing an extraordinary outcome, one where OpenAI’s ChatGPT, Gemini, Anthropic, and xAI all build giant businesses in the same overlapping space and all pull huge revenue from the rest of the economy.
That is where the story starts to wobble. The industries expected to buy these tools do not generate infinite profit. If every major AI company is supposed to dominate and earn massive margins at the same time, someone has to pay for that future. The math gets tight fast.
The hard question is not whether AI helps people work faster. It is how much value that speed creates, and how soon that value shows up in earnings.
A simple way to frame the debate is this:
| What the market may be assuming | Why that assumption looks fragile |
|---|---|
| Several AI leaders can all dominate | Their products overlap, so all cannot capture outsize profits from the same customer spending |
| Faster output means higher earnings | More output only matters if buyers pay for it and quality holds |
| AI infrastructure is a long-lived asset | Chips can age fast as better hardware arrives |
| Big tech can finance the race with ease | Debt and cross-holdings tie more of the trade together |
The productivity story may be real. The concern is that the valuation story has already raced far ahead of it.
Producing more is not the same as creating more value
One of the sharpest points in the conversation was this: markets may be mixing up “we can produce much more” with “we create more value.” Those are not the same thing.
A worker can draft twice as many emails in a day. A team can generate more code, more product copy, or more customer responses before lunch. Yet a business only wins if that extra output lifts sales, cuts labor without creating new errors, or improves the product enough that customers are willing to pay more.
For example, a law firm might summarize cases faster with AI, but a lawyer still has to check every line. A retailer can generate thousands of product descriptions, but if shoppers do not buy more, the extra output is mostly noise. Speed helps, but value is what gets paid.
This gap matters because AI lowers the cost of creating many things. When supply rises quickly, prices often come under pressure. Content becomes easier to make, but not every piece of content becomes useful. Code arrives faster, but teams still need testing, review, and security checks. Customer support can scale, but a bad answer can undo the whole gain.
That is why past tech manias still matter. Social media changed how people communicate. The Internet of Things spread sensors and connectivity into more places. Both were important. Neither created the smooth, economy-wide productivity lift that early excitement suggested.
AI could go further than either of those waves. Even so, businesses need time to change workflows, train employees, and figure out where AI creates paid value rather than more volume. Until that shows up clearly, investors may be cheering capacity before they can measure return.
The AI funding race is turning the sector into one linked trade
The other major worry is structural. The AI race is not a set of isolated companies competing in neat lanes. It is a web of firms investing in each other, financing each other, and feeding off the same market story.
The discussion pointed to giant funding events, the possibility of new IPOs such as Anthropic, and other large capital raises tied to AI. Cudmore called these “massive risk transfer exercises.” That phrase sounds harsh, but it fits. When money enters the same theme through private funding, public markets, corporate stakes, and headline valuations, risk does not vanish. It gets passed around.
That matters for earnings. Part of the profit improvement at some large tech firms can come from the rising paper value of their investments in other AI companies. Cudmore used Amazon and Anthropic as an example. If the value of that stake rises, reported results can look better. On the way up, that supports the whole narrative. On the way down, it can hurt just as quickly.
After the 2008 global financial crisis, mega-cap tech built a reputation as the market’s stronghold. These companies were cash machines. They were less burdened by debt than many older industries. Investors learned to see them as safe, especially during stress.
Cudmore’s point was not that these firms are suddenly weak. He was clear that they still look strong. The change is that they are pushing harder, spending more aggressively, and taking on more debt to keep pace in the AI buildout.
When the same companies are funders, customers, suppliers, partners, and valuation benchmarks, the whole trade becomes tightly linked. That can feel stable during a rally because each rising price helps validate the next one. It also means a stumble in one part of the system can travel quickly through the rest.
Why talk of an “earnings rug pull” is starting to sound more serious
This is where the skepticism turns from abstract to immediate. The risk is not that AI demand disappears overnight. The risk is that the market has stretched future earnings so far into the present that even good results may fail to satisfy.
For a while, the easy response was that the cycle could keep going for at least one more earnings season. Then maybe one more after that. Markets often extend longer than critics expect because rising prices make the story look stronger. As long as companies post real sales and the narrative stays intact, fresh money keeps arriving.
Cudmore said he had been relatively complacent about that pattern. He no longer sounded that way. What changed was not the existence of AI revenue, but the mood around it. He said the market now looks euphoric.
That matters because euphoria does not need fraud or collapse to end badly. It only needs expectations to outrun delivery. A company can report growth, sign customers, and still disappoint if investors expected something close to perfection.
The phrase “earnings rug pull” captures that risk. It suggests a moment when long-dated projections stop carrying the stock and investors start looking harder at what the company earns now, what it spends now, and how durable those profits really are. If capital spending stays huge while monetization moves slowly, the gap becomes hard to ignore.
This is why the debate feels more fragile than the headlines suggest. AI can be useful, profitable, and important, while parts of the trade are still overpriced. Those two ideas fit together without strain.
Energy costs may be the hardest limit on the AI boom
The final piece of the argument leaves software and walks straight into the power grid. Data centers are physical assets in physical places, and some draw power on the scale of a small city. That turns the AI buildout into an energy story whether investors want it or not.
The panel tied that pressure to oil prices. With Brent crude near $94 a barrel, the concern was not only the front-month contract. Cudmore said the more troubling signal was December Brent, still above $85 a barrel. He added that this was more than 35% above the one-year average price. If that persists, it amounts to a real inflation shock.
That matters because AI infrastructure needs more than chips. It needs electricity, cooling, land, construction, backup systems, and transport. In the US, growing demand is pushing attention back toward oil and gas, even if supply responses take time. The policy world is already catching up. A recent House hearing on AI data centers and power consumption shows how fast this issue has moved into the open. The same tension appears in Fortune’s report on U.S. data center energy use amid the AI boom.
Cudmore also made a useful comparison to the dot-com bubble. People still talk about the overbuild in fiber-optic cables from that era, yet those assets had a long use case. AI chips are different. Better chips can replace them within a few years. That shortens the payoff window and raises the risk of spending heavily on hardware that ages before it fully earns back its cost.
He also joked about how conveniently accounting has changed around this issue. The joke landed because the point is serious. An asset can still sit on the books while its economic edge fades much faster in the real world.
That is the less visible pressure point in the AI boom. Software stories can stay airborne for a long time. Power prices, debt costs, and hardware obsolescence do not stay out of the picture forever.
Final thoughts
The strongest takeaway is about timing, not technology. AI may deliver meaningful productivity gains over time, but markets are pricing a large payoff now.
If that gap keeps widening, the problem will not be that AI failed. The problem will be that investors expected too much, too soon, from a trade that also depends on debt, linked balance sheets, and rising energy costs.
