AI budgets expand faster than many companies can explain their unit costs

Menlo Ventures’ November 3 study estimates enterprise deployment spending at a $176.8 billion annualized run rate by September, versus a reclassified $33 billion comparison for the prior year. This is modeled market sizing, not an audited sum of cash spent during the first nine months. [1][3][4]

Deloitte’s October 29 survey shows why a bigger market is not the same as a measured return for each buyer. Among 3,500 leaders involved in AI initiatives across 25 countries, 71% report tracking total inference and API spending, but only 44% track costs per inference unit. Twenty-eight percent report revenue gains from AI, compared with 20% in Deloitte’s January report. Those are respondents’ assessments, not causal estimates of AI’s contribution to revenue. [2][5]

Menlo draws on nearly 400 US enterprise decision-makers surveyed by Morning Consult in September, plus filings and other sources. It changed its prior-year taxonomy and invests in several companies it measures. Deloitte’s survey was fielded in May–June, so its October publication is a new report of an earlier operating picture. The two samples and methods should not be blended into a single adoption rate. [1][2]

The immediate founder decision is what to measure before expanding an agent workflow: cost per accepted result, human correction time, completion quality and the cost of failures. A falling token price or rising token volume alone cannot answer whether the workflow creates more value. [1][2]