New
articleMigrating a 20-Year WordPress Blog to Be Agent-Readytechnology bytesAGI Has Arrived. Now Give It a Simpler Problem.podcastEpisode 100: The Next Era of AI — From Agents to Enterprise IntelligencearticleMigrating a 20-Year WordPress Blog to Be Agent-Readytechnology bytesAGI Has Arrived. Now Give It a Simpler Problem.podcastEpisode 100: The Next Era of AI — From Agents to Enterprise Intelligence
Navveen Balani Making systems intelligent, efficient & responsible

Taming AI's Cost and Environmental Footprint — APAC AI and Sustainability Council

View as markdown

I joined the APAC AI and Sustainability Council for a session on one of the most common patterns in enterprise AI today: adoption scales from pilots to organization-wide rollout, and the costs escalate far beyond what anyone budgeted.

The scaling surprise

This isn’t a problem of immature teams. Even organizations with highly evolved cost management practices — Microsoft, Amazon, Uber, and Starbucks were among the examples we discussed — have been caught off-guard and had to dial down their AI usage. And there is a less talked about, equally important consequence: energy, water, carbon, and hardware demand all scale in line with AI costs. Solve one and you solve the other — which is precisely why efficiency and sustainability are the same engineering problem.

Where the budget bust comes from

The root cause is a lack of discipline in how AI applications are built and used:

  • Using generative AI where it isn’t needed at all
  • Reaching for more powerful — and more expensive — models than the task requires
  • Poorly constructed prompts
  • Unclear goals for what the AI is meant to achieve
  • Excessive context shared with every call

None of these look like big decisions individually. Multiplied across an enterprise, they are the bill.

Lean Agentic AI

The remedies fall under what I call Lean Agentic AI: right-sizing models to tasks, routing by difficulty, disciplined prompts and context, and explicit budgets for what agents may consume. These are engineering practices, and they work.

Accountability above architecture

A perspective I shared with the Council, and one that resonated in the discussion: technical fixes alone will not hold unless a single person is made responsible for efficient AI, and unless it is treated as a board-level risk. The organizations that tame AI’s cost and footprint tend to be the ones that put accountability in place first, and let the architecture follow.

Thank you to the APAC AI and Sustainability Council for the invitation and the sharp discussion.

aigreen software