A beautiful mind, an ugly trade and a trillion-dollar mistake

What does a Hollywood romance scene, the Cold War, and the world’s biggest technology companies have in common? More than you might think. And for investors in artificial intelligence, the answer could prove rather expensive.

Microsoft, Alphabet, Amazon, and Meta are pouring hundreds of billions of dollars into the AI arms race. Each has perfectly sensible reasons for doing so. The unnerving possibility is that, together, they may be making one of the great capital-allocation mistakes of our time.

The technology may be revolutionary. The investment returns may be anything but.

A beautiful mind, and a rather awkward night out

In the Oscar-winning film A Beautiful Mind, Russell Crowe plays John Nash, the mathematician whose work on game theory earned him a Nobel Prize in Economics.

In one memorable scene, Nash is in a bar with his university chums when a beautiful blonde woman arrives with her companions. Unfortunately, every man at the table has exactly the same idea. Nash has a flash of inspiration. If they all compete for her attention, they risk irritating her friends and striking out altogether. He then promptly abandons the bar to work on the mathematics. Not everyone’s idea of a successful evening.

ETF

Plato Global Alpha Complex ETF (PGA1)

Global Shares

View

Managed Fund

Plato Global Shares Income Fund

Global Shares

View

The scene is Hollywood’s rather liberal interpretation of Nash’s work, but it captures a useful insight: what makes sense for each person in isolation can lead to a poor outcome for the group.

Fast-forward to 2026. The beautiful woman has been replaced by artificial intelligence, and the university friends by some of the richest companies on the planet. The stakes have risen somewhat.

The AI arms race nobody dares to lose

Google must defend the extraordinary economics of search. Microsoft sees AI as both a threat to, and an opportunity for, its enterprise software franchise. Amazon cannot afford to let AWS fall behind. Meta wants AI to improve everything from advertising to consumer engagement.

For each, underinvesting could be disastrous. Imagine Microsoft announcing that it will halve its AI spending. Free cash flow would jump and investors might briefly applaud. Then imagine Google and Amazon spending twice as aggressively, gaining technological ground and winning customers. That initial applause could become rather muted.

Now put yourself in Google’s shoes. Or Amazon’s. Or Meta’s. Each might prefer a world in which everyone spent less. None can safely assume the others will oblige.

And so the chequebooks stay open. Welcome to the trillion-dollar prisoner’s dilemma.

When rational decisions produce irrational outcomes

Nash showed that rational competitors can become trapped in an outcome none particularly likes. A Nash equilibrium is a situation in which no player benefits from changing strategy alone, given what everyone else is doing.

Consider two hypothetical AI investors, Alphabet and Microsoft. Each can spend aggressively or exercise restraint. These are illustrative payoffs, not forecasts. The first number represents Alphabet’s payoff and the second Microsoft’s.

If both invest cautiously, each earns six units of value. If one invests aggressively while the other holds back, the spender captures the gain and the cautious actor suffers. But if both spend aggressively, both lose two units.

That is the prisoner’s dilemma. It does not mean the hyperscalers are foolish. Quite the opposite. They may be making individually rational decisions in a game whose rules encourage collectively poor outcomes.

Real life is messier than a two-by-two matrix. The companies have different businesses, their choices unfold over years, and they can adapt. Apple, for one, has taken a comparatively restrained approach to building its own AI infrastructure. Still, the incentive to spend before a rival does is unmistakable.

The great revenue illusion

One of the most persistent mistakes in investing is to confuse an exciting industry with an attractive investment. They are not interchangeable. History has seen this film before. Railways changed the world but ruined plenty of financiers. The telecom boom gave us extraordinary communications networks, along with spectacular excess capacity and financial losses. The internet was an unambiguous technological triumph. Many internet infrastructure investments were not. The airline industry transformed travel. It did not however make many investors wealthy.

The potential scale of the makes these comparisons particularly uncomfortable:

These are average annual US investment estimates as a share of GDP, not peak spending. Crucially, the AI number is a forecast, not a realised outcome. If achieved, it would dwarf all previous capex booms.

At Plato Global Alpha, we have been short NextDC for more than five years, in significant part because of the enormous capital demands of data-centre infrastructure. Since then, the stock has fallen over 50% despite supposedly insatiable demand. The broader point is that relentless investment requirements can be a formidable obstacle to shareholder returns, even in a booming market.

The depreciation time bomb

There is an important difference between an AI data centre and a railway. Railway tracks can last decades. A GPU can become economically obsolete long before it stops working.

The issue is not whether yesterday’s chip still functions. It is whether tomorrow’s chip can do the same job at a fraction of the cost. If so, yesterday’s equipment may remain technically impressive but economically uncompetitive. A very expensive paperweight, albeit one with excellent cooling.

From A Beautiful Mind to mutually assured destruction

Game theory has a rather more unsettling application than a night at the bar. During the Cold War, the United States and Soviet Union amassed nuclear arsenals partly because neither could risk being left behind. Each attempt to become safer made the other side feel less safe, inviting another round of spending.

To be clear, commercial competition is not nuclear warfare. But the feedback loop is familiar. Secure the GPUs before your rivals. Lock up the electricity. Build the data centres. Announce the next investment programme. Then watch your competitors do the same.

A beautiful mind, an ugly investment?

We are not arguing that AI is a bubble, or that the revolution will end in tears. AI may exceed the optimists’ wildest expectations. Our concern is that investors are conflating technological prowess with shareholder success.

The further we descend on the AI food chain, particularly towards capital-intensive infrastructure like neo-clouds with fast-depreciating assets and aggressive financing, the more selective investors need to become.

John Nash helped explain why clever people can make individually sensible choices that lead to a collectively dreadful result. More than 75 years later, the world’s cleverest technology companies may be staging the most expensive demonstration yet.

In A Beautiful Mind, the problem was that everyone wanted the same woman. In AI, everyone wants to win the same race. The question is not whether AI will change the world. It is how many trillions will be spent before investors discover that winning the race and making money are not necessarily the same thing.

Invest in global equities alongside Plato Investment Management