Goldman Sachs: AI Sector to Hit Profit Margin Inflection Point by 2026

Goldman Sachs' technology report released on May 5th suggests that the key change in AI for intelligent agents is not just bringing more AI usage, but potentially transforming the return logic of AI capital expenditure: when Token costs continue to decline and prices stabilize, the growth in usage has the opportunity to translate into profit margin expansion.

Over the past two years, the core question for the market regarding AI capital expenditure has been whether cloud vendors and model companies can achieve sufficient returns after investing heavily in chips, electricity, and data centers. Goldman Sachs' answer is that if Token costs decrease faster than Token prices, the growth in AI inference demand is not just a revenue story, but may also improve the unit economic model.

The cost reduction mainly comes from improvements in semiconductor performance and computational efficiency. Goldman Sachs states that leading semiconductor companies are driving an annual cost reduction of 60% to 70% for Tokens, while the Token prices of leading large models have shifted from a previous annual decrease of about 40% to stabilizing or partially increasing.

This makes more complex intelligent agent tasks commercially viable - compared to traditional chatbots, intelligent agents require longer context windows, multi-round reasoning, result verification, tool invocation, and continuous backend monitoring, with each task consuming significantly more Tokens than ordinary Q&A.

Goldman Sachs believes that this will form a positive cycle: a decrease in Token costs makes complex intelligent agents easier to implement, complex intelligent agents bring higher Token consumption, higher usage improves infrastructure utilization rates, thereby improving the unit economics of cloud vendors and model providers.

Therefore, the bull case for AI capital expenditure is not just about "more people will use AI in the future," but more critically, whether the new usage can accumulate with higher incremental profit margins, and Goldman Sachs believes that this profit margin inflection point may become apparent starting in 2026.

In terms of demand estimates, Goldman Sachs forecasts that by 2030, intelligent agent AI could drive global Token consumption to grow 24 times compared to 2026, reaching 120 trillion Tokens per month. Of this, consumer intelligent agents may contribute about 60 trillion Tokens, and enterprise intelligent agents may contribute about 56 trillion Tokens.