Start Here

Welcome to The Inference Lab.

The 10-Day H100 Burn Down is happening right now. Here’s what you need to know:

### The Setup

Over the next 10 days, $4,000 of Azure H100 time is getting deployed to intentionally break modern AI inference stacks—DeepSeek, vLLM, KEDA—so you don’t have to learn those lessons on your production infrastructure.

Every experiment ships with:

- The exact infra question being tested

- The configs that worked (or blew up)

- Raw performance traces and cost breakdowns

### Why this matters

Most teams don’t have an H100 problem. They have a scheduling, scaling, and cost problem. This lab log documents what actually happens when you push these stacks to their limits.

Once the experiments are done, the raw data gets archived. If you’re not subscribed before then, you miss the safest way to derisk your AI infra roadmap.

### What happens next

**Free subscribers get:**

- The Day 1 Config (baseline H100 wiring in your inbox today)

- Analysis of each experiment in the open

- Access to the exact YAML, Terraform, and Grafana boards (gated)

**When you’re ready to ship:**

- Use the ROI Simulator (top menu) to sanity-check whether any config pays off in your environment before you commit spend.

This isn’t a course. It’s a live lab notebook. Treat it like one.

[Grab the Day 1 Config and subscribe →]