Jefferies questions LLM profitability as token prices collapse

Backlog pressure

In a 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. The analyst argues that Chinese AI models, open source or open weight models, and small language models threaten to undercut the prices the two US leaders charge, casting doubt on their future revenue.

Token traffic shifts

Jefferies points to research from global aggregator platform OpenRouter, which found Chinese models processed 76.12 trillion tokens in the week ended 27 September, up from 4.37 trillion in late April. The top four US providers processed 14.2 trillion tokens. It also says the launch of Moonshot AI open source Kimi K3 is competing at the frontier with an estimated 95 percent of the performance of Anthropic Claude Fable 5.

Capex and profit questions

The largest hyperscalers are set to collectively spend between $600 and $800 billion this year on building datacenters, funding it increasingly with debt rather than cash. Jefferies questions where future revenue streams will come from, saying 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 note adds that this is extremely relevant for the hyperscalers because they invested in datacenter capacity on the assumption that compute will be purchased by the likes of Anthropic and OpenAI.

Jefferies says the long term bet on AI may be sound, but there is a lot of uncertainty in the medium term. The paper asks whether the four hyperscalers are really going to spend $993 billion on capex in 2027. It also asks how much reported profits depend on investments in AI model builders on which the cloud providers will also depend for revenue. Other income and expenses, which describes gains in investments, has spiked in importance and likely includes positions in unlisted AI related stocks, according to the analyst.

Over the four quarters to 2Q26, the four major hyperscalers annualized earnings increased by $222 billion to reach $581 billion, and Jefferies says other non operating income or loss accounted for all of that increase. Other non operating income or loss increased by $224 billion year on year to $237 billion over the same period.

Revenue gap

Bain and Company 2026 annual Global Technology Report says the AI industry must generate $6 trillion in annual revenue by 2031 to keep funding the infrastructure needed to meet anticipated demand, meaning it will need new uses beyond employee productivity. Bain suspects existing AI applications might make up $1.8 trillion in revenue, leaving a $4.2 trillion shortfall. Emerging AI applications such as search engines and ads, AI built autonomous vehicles, and AI simulations could help. Even if they work out, these new categories would only add up to $1.5 trillion in revenue, leaving a $2.7 trillion gap that might be filled by new products and uses that do not exist today.

Sources: Theregister