🤚 The Open-Palm Illumination

The four horsemen of the cloud apocalypse — Alphabet, Amazon, Microsoft, and Meta — are on track to spend a combined $725 billion on AI infrastructure in 2026. Their reward? A combined free cash flow that, by Q3 2026, is projected to hit $4 billion. Not each. Total.

For context, these companies generated a combined $372 billion in net operating profit across their entire businesses in 2025. They are now spending faster than they earn, their free cash flow is approaching levels not seen since 2014, and the question Peter Diamandis poses in his latest MOONSHOTS segment has migrated from contrarian provocation to legitimate financial inquiry: Is the AI boom a bubble?

The answer, naturally, is more complicated than anyone with a position wants it to be.

👐 The Two-Handed Reality Check

Let’s start with the numbers that make investors reach for the antacids.

Alphabet’s free cash flow is projected to plummet almost 90% this year — from $73.3 billion in 2025 to $8.2 billion. The company just raised $80 billion in a June 2026 equity offering ($30B public, $40B at-the-market, $10B private placement from Berkshire), signaling that capex now exceeds what the company can self-fund. When Google needs to pass the hat, the magnitude of the spending becomes tangible.

Amazon’s $200 billion AI budget has pushed its free cash flow negative — approximately -$2.5 billion. Microsoft has seen free cash flow decline from an $89 billion peak to $71.6 billion. Meta compressed from $54.1 billion to $48.3 billion. Across the group, capital expenditure is in what analysts describe as “vertical ascent,” with a compound annual growth rate of 65% over four consecutive years.

And what are they getting for it? According to a MIT NANDA study, 95% of enterprise generative AI deployments delivered zero measurable impact on profit and loss. UBS estimates total third-party AI software revenue across all listed companies at just $2.5 billion, with Microsoft representing over 80% of that figure. OpenAI reports 900 million weekly active users but only 50 million paying subscribers — a 5.5% conversion rate that would make a freemium mobile game blush.

J.P. Morgan estimates the industry needs $650 billion in annual new AI revenue just to achieve a modest 10% return on deployed capital. Current AI revenue sits somewhere between $50 billion and $150 billion. The gap is somewhere between “ambitious” and “hallucinatory.”

🌿 The Gentle Awakening

Before you short everything and move to a cabin, the bull case deserves its moment in the velvet spotlight.

Cloud computing took 15 years to generate its first $100 billion in annual revenue. Mobile advertising took a decade. These companies are not spending blindly — they’re spending on the thesis that AI infrastructure, like electricity grids and fiber optic networks before it, will generate returns over years, not quarters. Jefferies analysts point to a $2 trillion backlog and accelerating cloud growth as evidence that demand is real, even if revenue recognition lags.

The historical parallel that cuts both ways is instructive. The 1990s fiber optic boom laid infrastructure that powered the next two decades of internet growth. The technology was transformative. The capital cycle was still a bubble. Corning dropped 80% after 2001. The fiber was still there, still carrying data, still changing the world. The shareholders were still broke.

Token consumption has increased 19-fold in 18 months. But per-million-token pricing has collapsed 65% in the same period, and API pricing has fallen 99.7% since GPT-4’s March 2023 launch. You can sell 19 times more product and still make less money if prices fall fast enough. This is Jevons Paradox doing what Jevons Paradox does, and it is delighting users while quietly terrifying CFOs.

Perhaps most concerning: Michael Burry — yes, that Michael Burry — estimates the industry understates depreciation by approximately $176 billion between 2026–2028, inflating combined earnings by roughly 20%. Moody’s has identified $662 billion in off-balance-sheet data center lease commitments, larger than the companies’ combined on-balance-sheet debt. The accounting, it turns out, is doing as much heavy lifting as the GPUs.

👑 The Crown Verdict

The honest answer to “Is the AI boom a bubble?” is the one nobody wants to hear: it can be both transformative and a bubble simultaneously. The technology can be the most important development since the internet while the capital deployed to build it still exceeds what the market will return to investors on any reasonable timeline.

The math is currently unforgiving. At a 20% return hurdle, these companies need $320 billion in annual AI profits — approximately 86% of their combined profits from all existing business operations. The AI segment would need to be nearly as profitable as everything Google, Amazon, Microsoft, and Meta already do. By next year.

Bain projects the industry needs $2 trillion in annual AI revenue by 2030 to justify the current scaling trend. We are at somewhere around $100 billion. A 20x increase in four years is not impossible — it is merely the kind of growth rate that has historically been achieved by exactly zero infrastructure categories in the history of technology.

The AI boom is real. The question is whether the companies building it will be the ones who profit from it, or whether they’ll be remembered as the railroad barons who went bankrupt laying track that someone else’s trains eventually used.

Inspired by “Is the AI Boom a Bubble?” | MOONSHOTS by Peter Diamandis.

Your portfolio allocation is showing. Diversify wisely.