• AI profitability may overstate true economic earnings
• Massive capex risks compressing future shareholder returns
• Transformational technology rarely guarantees transformational investment outcomes

There is a classic joke often told in the halls of investment banks.

A business owner is interviewing candidates for a senior role and asks each a simple question:

"What is two plus two?"

The mathematician answers instantly: "Four. Exactly four."

The statistician hedges: "On average, four—with a standard deviation of plus or minus 2%."

The accountant pauses, locks the door, pulls down the blinds, and whispers: "What do you want it to be?"

Wall Street loves the accountant"s answer.

Today, headlines suggest we have entered a golden age of profitability. Earnings are strong, margins are expanding, and growth appears effortless.

However, whether you should buy into this narrative depends entirely on which lens you apply: the rigid truth of the mathematician, the probability of the statistician, or the creative "flexibility" of modern accounting.

Buffett Frames Earnings Differently

Many decades ago, Warren Buffett introduced the concept of owner earnings.

It was designed to cut through accounting narratives and focus on economic reality.

Buffett has long argued that reported earnings — the numbers companies present to shareholders every quarter — are often a poor proxy for the actual cash economics of a business.

Or more bluntly: accounting earnings can often obscure underlying reality.

His framework was deceptively simple:

• Start with reported profits
• Add back non-cash charges like depreciation
• Subtract the capital expenditures required to maintain the business (e.g. ~75–80% of total capex)

For example:

The final step is the one markets often overlook.

Because if a company must continually reinvest enormous amounts of cash simply to preserve its competitive position, those earnings were never truly "yours" to begin with.

And it"s those businesses which Buffett selectively avoids.

The Red Queen Problem

The current AI arms race resembles the "Red Queen Race" from Through the Looking-Glass:

"Now, here, you see, it takes all the running you can do, to keep in the same place."

If every major technology firm spends $100 billion on AI infrastructure to become 20% more efficient, the competitive landscape remains static.

No one truly gains a definitive edge; they all simply end up with $100 billion less in the bank.

Managers are not necessarily investing entirely because the returns are immediately attractive; they are investing because the cost of not playing is existential.

This leads to a critical question:

If a company reports record profits while simultaneously spending the majority of its operating cash flow on CPUs, GPUs and data centers just to preserve its market share, how "distributable" are those profits?

Buffett would argue they aren"t profits at all—they are survival costs.

And those costs are staggering (and accelerating):

• Alphabet Inc. has guided to $190B in annual AI capex
• Microsoft has also guided to $190B – to scale data centre investment aggressively
• Amazon is reinvesting heavily through AWS – with $200B in capex this year; and
• Meta is guiding to around $145B in capex for AI

Morgan Stanley offers this 2 year forecast:

Under current accounting rules, this spending is capitalized and depreciated over several years—sometimes as many as seven.

This creates a "smooth" earnings profile on the income statement, but the cash is gone the moment the check is signed.

History shows that markets are often right about the direction of change, but poor at judging the economics.

We saw this with canals, railroads, and the fiber-optic boom of the late 1990s.

Each wave brought genuine transformation, yet transformational technology does not automatically translate into transformational shareholder returns.

Consider Cisco in the late "90s. It was the "picks and shovels" play of the era, much like Nvidia is today.

It had a bulletproof balance sheet and was enormously profitable.

Yet, as the infrastructure build-out hit a point of "digestion," Cisco lost ~90% of its value.

The technology (the internet) succeeded beyond anyone"s wildest dreams, but the assumptions embedded in the stock price failed.

Circular Capital

Disturbingly, we are seeing the emergence of circular capital flows that mirror the excesses of the dot-com era.

As further context – this article from Morningstar talks to the risks:

In short, what we are seeing is infrastructure providers invest in AI startups — which then commit that same capital back to the provider"s cloud services.

Take Google"s relationship with Anthropic.

Recently, Google reported a cloud compute backlog of $462 billion—a figure the market cheered. However, further analysis revealed that $200 billion of that backlog came from Anthropic alone, a company in which Google has invested roughly $40 billion.

These feedback loops reinforce growth narratives during expansion but create significant risk if the startup"s "demand" is merely recycled investment capital.

As Buffett once joked, "does management think the tooth fairy pays for capital expenditures?"

The economic burden of this spending remains, regardless of how the accounting treatment masks it.

A Margin of Safety

Many of today"s AI leaders trade at elevated multiples (e.g. 30x+ P/FCF; and more than 30x EV/EBITDA) because the market expects:

• Structurally high margins in perpetuity
• Declining capital intensity over time
• Sustained excess returns on capital.

If the AI race remains a Red Queen cycle, these expectations could prove fragile.

Why?

When infrastructure becomes abundant, pricing power weakens.

In the 1990s, the companies that built the fiber (the "picks and shovels") often struggled, while the application-layer businesses (like Google and Amazon) captured the long-term value.

Putting it All Together

The explosion we are seeing in capex echoes similar transformational cycles in the past:

• Railroads transformed economies, yet many railroad investors were destroyed.
• Telecommunications changed communication forever, yet many telecom operators collapsed.
• The internet created trillions in value, but much of it ultimately accrued to application-layer businesses rather than the companies laying fibre in the ground.

Each wave brought genuine technological transformation. And importantly, the market was not wrong about the direction of change.

That is what makes these periods so seductive: the underlying thesis is usually correct.

Now if it turns out the AI race increasingly resembles a Red Queen cycle — where enormous spending becomes necessary simply to maintain parity — then current profitability may prove less durable than markets expect.

Econ 101 tells us that if you make more of something – it is typically worth less.

For those chasing these valuations higher – it"s a high-risk endeavor.

But for those who have established positions at reasonable prices – there"s every chance momentum carries these further. I don"t pretend to guess how high.

In any case, one should be mindful that the gap between "accounting earnings" and "owner earnings" is widening.

The internet changed the world, but it didn"t save the investors who overpaid for the dream in the late 1990s.

In closing, I would always rather be aggressive when investors are fearful and pessimistic. Currently, they are neither.

In the end, cash flow always closes the gap between the story and reality.

History may not repeat, but as the massive capex bills come due, there is every chance it will rhyme.