The core logic of Goldman Sachs' recent AI report

Agentic AI has the potential to turn AI from a capital-intensive, constantly burning money cost burden into a business where increased usage leads to expanding profit margins
As token costs continue to decline, more complex agents become economically feasible. These agents will consume far more tokens than ordinary applications because they require longer context windows, multiple reasoning cycles, result verification, tool calls, and 24/7 backend monitoring
The rise in token usage will improve infrastructure utilization, strengthen unit economics, and leave more room for cloud providers and model vendors to reinvest in model quality, distribution capabilities, and computing capacity
In other words, the multi-faceted logic of AI capital expenditure is not just about increasing usage. More importantly, this growth in usage is increasingly likely to flow into attractive incremental profit margins, reflecting in financial performance. Goldman Sachs believes that starting from 2026, this profit margin inflection point will gradually become apparent