A financial analyst is warning that the sudden arrival of cheap Chinese and open source AI models could force a massive "reset" of the cloud computing industry's trillion-dollar infrastructure spending plans, as collapsing costs threaten to undermine the business case for artificial intelligence. In a research note published this month, investment firm Jefferies estimates that roughly half of the approximately $2.3 trillion in contracted revenue backlog held by Microsoft, Alphabet, Amazon and Oracle belongs to just two customers: OpenAI and Anthropic. But the analyst questions whether those AI companies will ever turn a profit as Chinese models, open source alternatives and smaller language models drive token prices sharply lower.
The numbers paint a dramatic picture of shifting market power. Chinese AI models handled 76.12 trillion tokens during the week ending September 27, up from only 4.37 trillion in late April, according to data from global aggregator platform OpenRouter that Jefferies cited. By comparison, the top four US providers processed 14.2 trillion tokens. Meanwhile, Moonshot AI's open source Kimi K3 model is delivering an estimated 95 percent of the performance of Anthropic's Claude Fable 5 while "competing at the frontier." The four largest hyperscalers are on track to collectively spend between $600 billion and $800 billion this year constructing datacenters, funding those projects increasingly with borrowed money rather than cash reserves.
The Jefferies paper questions whether the market has fully accounted for these developments. "It can no longer be assumed, most particularly with the arrival of the likes of Kimi K3, that large language models will ever be profitable given the related ongoing collapse in token pricing," the report states. This matters enormously for cloud providers, the analyst writes, since they've built datacenter capacity expecting companies like Anthropic and OpenAI to purchase that computing power. The analyst also flags a striking accounting pattern: while the four major hyperscalers' annualized earnings climbed by $222 billion to reach $581 billion in the four quarters through the second quarter of 2026, "other non-operating income/loss" — which captures gains from investments, likely including stakes in unlisted AI companies — jumped by $224 billion year-on-year to $237 billion, meaning it accounted for all of the earnings growth.
The analyst argues that while artificial intelligence may prove sound as a long-term wager, medium-term uncertainty is mounting. "The AI story will be due for a big reset at some point in the not-so-distant future," Jefferies writes, questioning whether the four hyperscalers will truly follow through on projected capital expenditures of $993 billion in 2027. Separate research from management consultants Bain & Company underscores the revenue challenge: the AI industry must generate $6 trillion in yearly revenue by 2031 to justify continued infrastructure investment and meet anticipated demand. Bain estimates existing AI applications might produce $1.8 trillion in revenue, leaving a $4.2 trillion gap. Emerging categories like AI-powered search advertising and autonomous vehicles could add another $1.5 trillion, still leaving a $2.7 trillion shortfall that would need to come from "new products and uses that don't exist today."
The core concern is a fundamental mismatch between massive infrastructure commitments and rapidly eroding unit economics. Cloud providers have borrowed heavily to build capacity for AI workloads, counting on premium-priced inference from leading US model builders. But if Chinese and open source competitors deliver comparable performance at a fraction of the cost, those revenue assumptions collapse — and with them, the rationale for hundreds of billions in datacenter construction. The industry now faces a critical question about whether anticipated demand will materialize at price points that justify the investment already underway.