Building Your Own AI Model Is Usually the Wrong First Step
Core idea: Most teams should use existing models before training their own.
The Temptation to Build
When AI became a strategic priority, many organizations' first instinct was: "We need to build our own model." The thinking was that a proprietary model trained on proprietary data would become a new competitive moat.
The economics tell a different story.
The Pricing Scissors
Steven Brovich highlighted a critical economic dynamic in his AWS leadership talk:
- Training costs are rising 2.4x per year
- Inference costs are falling 10x per year
- The gap between these two curves — the pricing scissors — is opening by 12-24x per year
What does this mean practically? The cost to create a frontier model is now approaching the billions. Only a handful of companies can afford it. But the cost to use one is collapsing toward zero.
The Three Worlds
Brovich outlined three worlds for AI economics:
| World | What You Do | Leverage | Differentiation | Cost |
|---|---|---|---|---|
| Use | Consume managed AI services | Highest | Lowest | Lowest |
| Compose | Stitch frontier APIs into your workflow | Medium | Medium | Medium |
| Build | Train or fine-tune your own models | Lowest | Highest | Highest |
The key insight: don't try to live in one world. Let economics and your actual differentiation drive which world each part of your workflow lives in.
The Healthy Path
- Day one: The frontier model does everything. You learn what works and what doesn't.
- Month six: You understand the economics. You move some tasks to "use" and some to "compose."
- Year two: The high-volume, high-differentiation stuff moves to "build."
The unhealthy path is the leader who says "We're a build shop on day one" and burns the company trying to train models before they understand their own workflow.
When Building Makes Sense
Building your own model is still a valid option — but only when:
- You have a clear differentiation that a custom model enables
- You've already exhausted what managed services and composition can do
- You have the talent and budget to do it properly
- You understand your workflow well enough to know what the model needs to do
For most organizations, most of the time, the answer is "use" or "compose" — not "build."
What This Means for You
- Start with managed services. They're the highest-leverage, lowest-cost way to get started.
- Customize through composition. Your proprietary data and workflow are where your differentiation lives — not in the model itself.
- Build only when you must. If you can't articulate why a custom model creates defensible value that APIs can't provide, you're not ready to build.
- Let economics guide your strategy. The pricing scissors mean that using frontier models will only get cheaper over time. Building your own will only get more expensive.
Portions of this article are based on insights from Steven Brovich's talk "A leader's guide to advanced team structures in an agentic world" at AWS Events. You can watch the full talk here: A leader's guide to advanced team structures in an agentic world.