Chatham Financial estimates that AI infrastructure capital expenditure will exceed $830 billion in 2026. That figure, cited by managing partner Amol Dhargalkar in a semiannual market update, is roughly half the volume of the entire US investment-grade bond market and two-thirds of leveraged loans. It significantly exceeds the high-yield bond market. The implication: AI infrastructure is now competing for capital at a scale that could squeeze out every other borrower.
This is not a theoretical concern. Data center capex grew 57% in 2025 to $726 billion, according to Dell’Oro Group, the fastest growth the firm has recorded since it began tracking the statistic in 2014. Data center capex is expected to cross $1 trillion in 2026, a milestone that just a year ago was projected for 2029.
The Hyperscaler Cash Vacuum
The four largest hyperscalers, Amazon, Google, Meta, and Microsoft, increased data center capex by 76% in 2025, according to Dell’Oro. Their 2026 plans are even more aggressive. Amazon CEO Andy Jassy told investors the company expects to spend about $200 billion on capex this year, “predominantly in AWS, because we have very high demand.” Google CEO Sundar Pichai said Google plans to spend around $180 billion.
The demand signal is reinforced by backlogs. Amazon’s backlog stands at $244 billion, up 40% year over year. Google reported a $240 billion backlog. Pichai noted that “the number of deals in 2025 over a billion dollars surpassed the previous three years combined.”
Where the Squeeze Hits
Credit spreads remain relatively tight, which suggests the crowding-out effect hasn’t fully materialized yet. But Dhargalkar warned in Global Finance Magazine that the pressure could build as asset managers begin favoring bonds issued by giant hyperscalers (with top ratings and ample cash) over more obscure, high-yield industrial companies. “There’s so much money flowing into AI, there could be more confidence in what their risk-adjusted returns will be,” he said.
The downstream effects are already visible in hardware markets. “Hardware prices are going up,” Dell’Oro analyst Baron Fung told Network World. “The cost of memory has gone up by double digits.” Memory can account for half the total cost of a server. Enterprise customers and smaller companies can’t match hyperscaler purchasing power. “So, they’re buying fewer servers, or using existing servers for a longer period until prices settle down,” Fung said.
The Recursive Loop Problem
Man Group, the $175 billion hedge fund, published an analysis this week arguing that the AI capex cycle is structurally different from prior technology booms. The core issue: the demand signal is circular. Microsoft, NVIDIA, Amazon, Meta, Google, OpenAI, and Anthropic “act simultaneously as suppliers, customers, investors, and validators.” Revenue growth looks spectacular because each node in the loop pays another. Capex looks justified because demand from inside the loop appears endless.
“What are the real costs? Who are the real buyers?” Man Group asks. Their assessment: the demand signal has become “divorced from the market.”
The firm notes that individual firms are behaving rationally. The existential competitive pressure of AI means each company’s decision to invest heavily is defensible in isolation. “Yet collectively, the behaviour is deeply irrational,” Man Group wrote. “Multiple companies training similar models, building overlapping infrastructure and bidding up the same constrained resources.”
The $7 Trillion Question
Dhargalkar framed the total scale at $7 trillion in cumulative AI infrastructure spending over less than half a decade. Actual expenditures may come in lower than estimated, he told Global Finance, given limited resources to build data centers and inflationary effects from the Iran conflict. “But at some fundamental level,” he said, “there’s not enough money for everything, and you certainly don’t want to get caught out on the wrong side of that transition.”
For non-AI companies planning to raise capital, the math is simple. The same pool of institutional investors that buys your bonds also buys Microsoft’s. Microsoft has a AAA-equivalent rating, unlimited demand for its product, and government contracts. You have a B+ rating and a capital improvement plan. When $830 billion of AI infrastructure debt competes for the same capital pool, the question for every CFO outside tech becomes: can you still raise at a price that works?