🧰 AI Factory as a Service · Overview

AI Factory as a Service at a Glance: Who Sells, Who Buys, What It Costs

How everyone outside Big Tech gets an AI factory, on one page: the four routes (build, install, buy, rent), the providers behind each, the customers who have announced what they bought, a quick route finder and a break-even calculator. Every figure comes from the six-part series and is sourced there.

Explore the routes ↓ Start at Part 1 →
⚠️ About the numbers. Everything here comes from the six articles (as of October 2026), which cite their sources. Vendor claims, such as customer counts, build times and performance, are labelled as claims. The calculator is our own simplified model.

🧭 The four routes

🏢 Provider explorer

Click a column heading to sort, or filter by route. Click a provider to read the part that covers it.

👥 Customer map

Named customers from Part 5 and earlier parts, on a schematic world map, coloured by sector. Tap a dot.

Tap a customer.

❓ Which route fits?

Three quick questions. An illustration, not advice; Part 1 has the full five-question version.

🧮 Break-even in three sliders

At what utilisation does owning GPUs beat renting on demand? Fixed assumptions match Part 6's defaults: $40k hardware per GPU, 1.8 kW per GPU, $0.12/kWh, $250/kW-month colocation, one engineer per 400 GPUs, $3,000 software per GPU-year.

Full model with ten editable inputs: Part 6 →

⏱️ Market timeline

Tap a milestone.
🚀 Bonus: starting your own frontier lab? Compute, cost and capital, with a Bangalore playbook →

📚 Reading tracks

🔤 Glossary

The other "Build the Machine" hubs: 🖥️ AI Compute Stack · 🏭 Producing Intelligence · 🧑‍🔧 Who Runs the AI Factory · 🏗️ AI Factories at a Glance