Goldman Sachs released a research report in early May 2026 titled “Decoding the Agentic Economy” that gives enterprise leaders one of the clearest financial frameworks yet for understanding where AI spending is actually heading — and why their budgets are about to feel very different.

The headline number: global token consumption is expected to multiply 24 times by 2030, reaching 120 quadrillion tokens per month. By 2040, if enterprise agents reach full-scale adoption, that figure could hit 55 times current levels.

Why Agents Consume So Much More Than Chatbots

This isn’t just bigger usage of the same thing. The underlying dynamic is fundamentally different.

A standard chatbot interaction is transactional — a user types a prompt, the model responds, the session ends. An AI agent works nothing like that. It continuously monitors its environment, reads and re-reads context as conditions change, automatically calls external tools, verifies its own outputs across multiple rounds, and often runs 24 hours a day without human prompting.

Goldman Sachs estimates that enterprise-grade agents will account for over 70% of all token usage by 2040. The report points to this “always-on” behaviour as the key driver — not more users asking more questions, but agents that never stop working.

This is already showing up in real numbers. Companies deploying agentic workflows are reporting that AI billing looks nothing like what they budgeted. One major technology company burned through its entire 2026 AI coding budget in just four months. The gap between “running a few AI experiments” and “deploying production agents across the business” is wider than most finance teams anticipated.

An Inflection Point for the AI Industry

Goldman Sachs identifies the first half of 2026 as a likely profit inflection point for the AI industry itself. Token prices have fallen substantially as compute costs dropped, but the volume surge driven by agents is expected to offset that decline and push AI infrastructure companies toward positive margins.

The bank named nine companies as primary beneficiaries of this shift: Nvidia, Broadcom, AMD, Amazon, Alphabet, Meta Platforms, Microsoft, Cloudflare, and Accenture. What’s notable about this list is how much of it is infrastructure and services rather than pure model providers — the real money, Goldman Sachs argues, flows to whoever handles the compute and orchestration layer.

What This Means for Business

For business owners and operations leaders, the Goldman Sachs report has three practical implications.

Budget for volume, not just access. Many companies have structured AI costs around seat licences or flat-rate API plans. As they move from chatbots to agents, that model breaks. Token consumption scales with how hard agents work, not how many people approved the tool. Finance teams need forecasting models that account for always-on agent activity.

Your data foundation determines your ROI ceiling. Goldman Sachs flagged a warning that often gets buried in the headline numbers: bad data could undercut the entire payoff. Agents that pull from inconsistent, unstructured, or siloed data will amplify those problems at scale. Companies with strong data literacy pull ahead on measurable outcomes — and those that invested in clean data infrastructure before deploying agents are already seeing that advantage compound.

The gap between pilots and production is real. Moving from a successful AI pilot to a production deployment that actually drives business value is not a linear step up — it requires rethinking workflows, governance, and cost models simultaneously. Three prerequisites determine whether that transition actually works: process documentation, data accessibility, and a named owner for the outcome. The companies seeing the strongest returns are treating AI deployment as an operational transformation, not a technology purchase.

Enterprise DNA works with businesses at exactly this transition point — from data readiness through to agent deployment and workflow redesign. If your team is preparing to scale AI beyond the pilot stage, getting the data foundation and operational model right before the volume hits is the difference between a cost that justifies itself and one that doesn’t.

---

Ready to build the data foundation that makes AI agents pay off? Talk to the Enterprise DNA team about getting your business AI-ready.