Agentic AI Return on Investment

EY's Total Cost of Agents analysis finds that AI tokens represent only roughly one-third of the true enterprise cost of running agentic AI systems, with the remaining two-thirds accumulating across subscriptions, platform infrastructure, governance, organizational change, expected failure costs, and emerging regulatory compliance—making the all-in enterprise cost approximately three times the token invoice. Global AI spending is projected to rise from $1.76 trillion in 2025 toward $9.1 trillion annually by 2035, generating an annualized economy-wide AI recovery charge of approximately $9.8 trillion by 2035 under a 12% capital recovery model. To earn a market return on this build-out through cost reduction alone, white-collar industries in upper- and upper-middle-income economies would need to shed roughly 16% of employer labor costs by 2031 and nearly 22% by 2035; alternatively, white-collar GDP output would need to expand by 10.5% by 2031 and 14.5% by 2035. The analysis concludes that 'AI saves time' is no longer a sufficient business case, and that organizations must govern agent capacity like capital and direct investment toward value pools large enough to justify the full cost stack.