For a decade, cloud cost management meant watching compute, storage and network. Now there's a new line item growing fast and hiding in plain sight: GenAI and LLM tokens. Every prompt, every agent, every RAG call adds up — and most organisations have no idea which model, team or feature is driving the bill.
Why AI spend is the new blind spot
- Token pricing is per-call and opaque — cost scales with usage, not headcount.
- Spend is spread across providers — Claude, OpenAI, Gemini, Azure AI — each with its own billing.
- Engineering ships fast; finance sees the bill weeks later.
What good looks like
From our FinOps practice (built on Aquila Clouds' Andromeda™ platform), the fundamentals are:
- Token-level observability — spend and usage by provider, model, user, project and feature, in real time.
- Model-level analysis — know whether premium-model calls or a cheaper model are driving cost, and where you can safely downshift.
- Forecasting & budgets — project AI spend and set guardrails before overruns happen.
- Governance — token-level policy enforcement so AI adoption scales without runaway cost.
From our experience
Helping enterprises optimise large-scale cloud and AI spend (₹600 Cr+ under management, up to 30% saved), the pattern is consistent: the teams that win treat AI spend like any other cost centre — tagged, owned, forecast and reviewed. AI copilots like Agent Sherlock even let non-technical stakeholders simply "ask their data" — "what's driving this month's AI cost?" — and get an answer in seconds.
Three things to do this quarter
- Tag every AI workload to a team and a use-case.
- Set per-team budgets and anomaly alerts on token spend.
- Review model choice — often a large share of tokens can move to a cheaper model with no quality loss.
Takeaway: AI is now a first-class cloud cost. Get visibility before the bill gets your attention.