Cheap models put AI infrastructure spending under scrutiny

Jefferies is questioning whether cheaper Chinese, open source/weight and small language models could force a reset in AI infrastructure economics, especially for hyperscalers counting on OpenAI and Anthropic to absorb new datacenter capacity. The firm 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 pressure is rising as alternatives gain usage and capability. OpenRouter data cited by Jefferies showed Chinese models processed 76.12 trillion tokens in the week ended 27 September, up from 4.37 trillion in late April, while the top four US providers processed 14.2 trillion tokens. Moonshot AI’s open source Kimi K3 is described as “competing at the frontier” with an estimated 95 percent of the performance of Anthropic’s Claude Fable 5.

Jefferies says the largest hyperscalers are set to spend between $600 and $800 billion this year on datacenters, increasingly funded with debt, while token prices keep falling. It asks whether the four hyperscalers are really going to spend $993 billion on capex in 2027, and points to non-operating investment gains as a major driver of recent earnings growth.

Bain & Company’s 2026 annual Global Technology Report adds to the concern, estimating the AI industry must generate $6 trillion in annual revenue by 2031 to fund expected infrastructure demand. Existing applications may produce $1.8 trillion, leaving a $4.2 trillion shortfall before speculative new uses are considered.