Jefferies는 저가의 중국산 및 오픈소스 LLM이 수조 달러 규모의 AI 인프라 '리셋'을 촉발할 수 있다고 주장하며, 토큰 가격의 지속적 폭락 속에서 LLM이 과연 수익을 낼 수 있을지 의문을 제기했다.
A financial analyst is questioning whether the collapse in AI costs, driven by the arrival of low-cost Chinese and open source models, will cause a "reset" in the AI story as cloud providers struggle to recoup trillions of dollars of capital spending. In a recent research note focused on Asian markets, Jefferies estimates that about half of the roughly $2.3 trillion contracted backlog at Microsoft, Alphabet, Amazon and Oracle is owned by OpenAI and Anthropic. Chinese AI models, including open source (open-weight) and small language models, threaten to undercut the prices the two US leaders charge, casting doubt on their future revenue.
Jefferies points to research from global aggregator platform OpenRouter, which found that Chinese models processed 76.12 trillion tokens in the week ended 27 September, up from just 4.37 trillion in late April. The top four US providers processed 14.2 trillion tokens. Meanwhile, the launch of Moonshot AI's open source Kimi K3 is "competing at the frontier," with an estimated 95 percent of the performance of Anthropic's Claude Fable 5. "The question now is whether all of the above has been priced fully into markets. This writer does not think so," the Jefferies paper states.
The largest hyperscalers are set to collectively spend between $600 billion and $800 billion this year on building datacenters, increasingly funding it with debt rather than cash. Jefferies, however, questions where the hyperscalers' future revenue streams will come from. "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. This is extremely relevant for the hyperscalers since they have invested in datacenter capacity on the assumption that the compute will be purchased by the likes of Anthropic and OpenAI," the paper says.
While the long-term bet on AI may be sound, the paper argues, there is considerable uncertainty in the medium term. "The AI story will be due for a big reset at some point in the not-so-distant future. Are the four hyperscalers really going to spend $993 billion on capex in 2027?" Jefferies asks.
There are also important questions about how much of the hyperscalers' reported profits come from their investments in AI model builders, on which they will also depend for revenue. "Other income and expenses," which describes gains in investments, has spiked in importance. Jefferies says these investments are "likely to include positions in unlisted AI-related stocks." While the four major hyperscalers' annualized earnings increased by $222 billion to reach $581 billion in the four quarters to 2Q26, "other non-operating income/loss" accounted for all of that increase, according to the analyst. Jefferies says "other non-operating income/loss" increased by $224 billion year on year to $237 billion over the same period, meaning it accounted "for all of the increase in earnings."
Other research has underlined how much revenue and borrowing cloud providers may need to fund their datacenter plans. The 2026 annual Global Technology Report from management consultants Bain & Company says the AI industry must generate $6 trillion in annual revenue by 2031 to keep funding the infrastructure needed to meet anticipated demand. That means the industry will need new uses beyond employee productivity. Bain suspects existing AI applications might account for $1.8 trillion in revenue, which still leaves a $4.2 trillion shortfall. That gap could be partly filled by emerging applications such as search engines and ads, AI-built autonomous vehicles, and AI simulations. Even if these work out, the new categories would add only $1.5 trillion in revenue, leaving a $2.7 trillion gap that might be filled by "new products and uses that don't exist today."
The Register does not offer investment advice.®