Governing AI Spend Without Slowing Engineers Down

Corin Bishop built FinOps at Cloudera from scratch — no team, no prior function — and scaled it to cover multi-cloud spend across a billion-dollar enterprise. In this episode, he walks through how he is now applying that same discipline to AI: a centralized LLM gateway that provides secure, auditable, cross-cloud model access for engineers at scale; a FinOps champion program designed to distribute cost ownership across hundreds of teams; and a blunt observation that the field has not yet solved for AI value measurement, regardless of what vendors are claiming. For FinOps and engineering leaders trying to get ahead of AI spend governance, this is a concrete technical and operational reference point.

This episode covers how Corin is approaching the new layer of complexity that AI introduces: governing access without blocking engineering velocity, building a centralized LLM API gateway (LiteLLM) to provide secure, auditable, and cost-attributed AI access at scale, and keeping an eye on the cloud spend that teams are quietly leaving behind while chasing the next model release. He also shares how carbon footprint data has become an unexpected but effective lever in convincing engineering teams to make smarter infrastructure decisions.

Key Takeaways

Governance does not have to be a blocker. The right model is: pre-approve categories of use, build secure access pathways, and then get out of the way. The goal is for the governance layer to be invisible to engineers once it is working correctly.
Visibility at the team and user level is a prerequisite for AI cost accountability — but the next frontier is efficiency, not just spend. Sending 500 input tokens and receiving 28 in return is a signal worth investigating. That kind of conversation requires gateway-level data, not just billing reports.
The FinOps champion model — embedding nominated cost owners within each engineering team and connecting them into a central virtual FinOps team — is how Corin is approaching a $5M incremental savings target in 2026. Low-hanging fruit is gone; distributed ownership is what scales from here.
Carbon footprint metrics are a surprisingly effective engineering engagement tool — not because engineers care about sustainability by default, but because region-level emissions data can justify infrastructure moves that also reduce cost. The conversation changes when you show the data rather than making a policy argument.
The single most honest observation in this episode: no one has actually figured out how to measure the business value of AI spend yet. Any practitioner who says otherwise is ahead of the evidence. Building the attribution infrastructure now is what puts you in position to answer that question when it becomes unavoidable.

About the guest

Corin Bishop manages public cloud operations at Cloudera, a $1B+ enterprise data company, where he is responsible for both keeping multi-cloud infrastructure running and ensuring that every dollar spent on AWS, GCP, and Azure is owned, allocated, and accounted for. He introduced FinOps as a formal practice at Cloudera from scratch and has driven it largely as a one-person function, using a combination of tooling, cross-functional partnership, and governance discipline. His current work spans AI access governance, LLM cost attribution, and cloud carbon footprint reporting.