The AI Boom: Revolution or Capital Destruction?

The Trillion-Dollar Question

The AI revolution has undoubtedly transformed the way businesses and investors think about the future. But the biggest question today is no longer whether AI will change the world - it almost certainly will. The real question is whether the companies investing trillions of dollars into AI infrastructure will generate returns that justify this unprecedented spending.

Jefferies’ Global Head of Equity Strategy, Chris Wood, has warned that the hundreds of billions of dollars being poured into AI infrastructure could culminate in “massive capital destruction” as cheaper Chinese open-source models erode the economics underpinning America’s investment frenzy.

Wood’s longer-term base case is that market share will gradually shift towards Chinese large language models, while investors increasingly question whether US technology companies can generate adequate returns from their unprecedented capital expenditure. Microsoft, Alphabet, Amazon and Meta are expected to spend a combined US$695 billion on capital expenditure in 2026, rising to US$870 billion in 2027. Together, that amounts to nearly US$1.57 trillion over just two years.

Alphabet alone recently raised its 2026 capital expenditure guidance by another US$15 billion, taking expected spending to between US$195 billion and US$205 billion.

The scale of spending has transformed businesses once celebrated for their asset-light business models. After revising their guidance earlier this year, the four hyperscalers are now expected to spend an astonishing 92% of their operating cash flow on capital expenditure in 2026.

Initially, investors welcomed this spending because demand for AI appeared to validate these investments. Anthropic’s annualised revenue run rate, for example, reportedly increased from US$9 billion to US$47 billion within a matter of months, reinforcing optimism around enterprise adoption and agentic AI.

However, the question investors had postponed is now becoming impossible to ignore: Where will the returns on all this capital actually come from?

China Is No Longer Just Catching Up

The threat from China is no longer limited to offering cheaper AI models. Increasingly, China is being viewed as a genuine technological peer rather than merely a fast follower.

According to the report, Chinese AI models processed 36.39 trillion tokens on OpenRouter during one week in July, compared to just 7.39 trillion tokens for leading US models. Competition intensified further with the launch of Moonshot AI’s open-source Kimi K3 model, which reportedly delivers around 95% of the performance of Anthropic’s Claude Fable 5.

This follows what many investors now refer to as the “DeepSeek moment” in early 2025, when Chinese open-source models first demonstrated that high-quality AI could be delivered at dramatically lower costs. Why does this matter? Because if high-quality AI becomes freely available through open-source alternatives, pricing power shifts away from proprietary model developers. AI risks becoming a commodity rather than a premium product, making it much harder to earn attractive margins.

Adding to this concern, the Silicon Data LLM Token Expenditure Index shows that the average price paid for one million AI tokens has fallen 25% since late May to around US$1.55. Lower prices will almost certainly increase AI adoption over time.

But from an investor’s perspective, higher usage does not automatically translate into higher profits.

The Story Is No Longer Just About Equity Markets

Perhaps the most underappreciated aspect of this AI boom is that it has now become a credit-market story as well.

Rather than funding all this investment purely through internally generated cash, the hyperscalers have collectively issued approximately US$194 billion of investment-grade debt during 2026. Credit markets are beginning to respond. Bond spreads for Amazon, Alphabet and Meta have all widened, suggesting investors are demanding greater compensation for lending to these companies.

Oracle presents an even starker example. With ~US$120 billion of debt, its credit rating was downgraded to BBB-, just one notch above junk status, causing its bond spreads to widen significantly.

The Customer Concentration Risk

Another concern lies in the US$2.1 trillion of Remaining Performance Obligations (RPOs) reported by Microsoft, Alphabet, Amazon and Oracle. Simply put, RPOs represent future revenue that companies have already contracted but have not yet recognised. They provide visibility into future sales but are only as strong as the customers backing those commitments. The issue is that roughly half of these commitments are reportedly linked to OpenAI and Anthropic. Neither company is currently profitable.

This effectively means that hyperscalers are building enormous data centres today based on demand from customers who themselves continue to consume cash rather than generate it.

Risks Beyond the Balance Sheet

Chris Wood also highlights another issue that many investors may overlook.

Future data-centre lease commitments are estimated at approximately US$662 billion, compared to just US$152 billion at the end of 2023. Another study estimates that the five largest hyperscalers now have approximately US$1.65 trillion of off-balance-sheet obligations - greater than the debt reported on their balance sheets.

Meanwhile, accounting has yet to fully reflect today’s spending. The four companies invested approximately US$130 billion in capital expenditure during the first quarter while recognising only US$41.6 billion of depreciation and amortisation expense. Since depreciation is recognised gradually over many years, current earnings may still understate the true economic cost of today’s investment cycle.

Wood also notes that a substantial portion of recent earnings growth has come from non-operating income rather than core business operations, potentially flattering headline profitability.

A Counterpoint Worth Considering

While Chris Wood’s concerns deserve serious attention, investors should also remember that the hyperscalers are not merely selling AI models. Microsoft can monetise AI through Microsoft 365 and Azure, Alphabet through Search and Cloud, Amazon through AWS, and Meta through improvements in advertising efficiency.

In other words, even if AI models themselves become commoditised, these companies may still earn attractive returns by embedding AI into products and ecosystems that already generate substantial cash flows.

Final Thoughts

Chris Wood is not arguing that AI is a passing trend. If anything, he believes falling computing costs will make AI even more pervasive across industries. His concern is about the economics of today’s investment cycle.

History offers several examples where transformative technologies created enormous value for society but not necessarily for investors. The internet boom of the late 1990s is perhaps the best example - while the internet transformed the global economy, many companies that built the underlying infrastructure failed to generate commensurate shareholder returns.

Whether AI follows a similar path remains to be seen.

The question is no longer whether AI will change the world. The question investors must answer is far more important: Who will ultimately capture the economics of that change?