Mercator

Mercator started because I kept drawing the same diagram on whiteboards. The AI stack has layers — silicon, infrastructure, model, orchestration, application — and every conversation about strategy required redrawing them, arguing about where the boundaries fall, and disagreeing about which layer is commoditizing fastest.

So I built the diagram once, made it interactive, and started updating it monthly with real pricing data, market share estimates, and my own annotations about where the interesting margin is moving.

The map is not the territory, but a good map changes which territory you decide to explore.

How It Works

Each layer is a horizontal band. Within each band, companies are positioned by openness (left) vs. proprietary (right) and by maturity (bottom) vs. emerging (top). You can drag and drop to rearrange, click to see pricing and positioning notes, and export the whole thing as a Mermaid diagram for your own presentations.

The data is hand-curated. I tried automating it with scrapers and LLM-generated summaries, but the editorial judgment — what to include, what to omit, what to flag as changing fast — turned out to be the value. Automation would have produced a comprehensive, useless map. Curation produces an opinionated, useful one.

Stack

• React for the UI — drag-and-drop with a custom physics layer for node positioning
• D3 for the visualization — force-directed layout with manual pin overrides
• Cloudflare Workers for the API — monthly data updates ship as JSON blobs to the edge

Mercator is live and free. About 6,200 people use it monthly, mostly founders and investors trying to figure out where to place bets. The most-clicked layer, consistently, is orchestration — which tells you something about where the uncertainty lives.