Notes on compute economics.
Plain-English guides to GPU pricing and AI infrastructure.
This blog answers the questions AI teams ask before they sign a GPU contract. We compare reserved and on-demand GPU pricing, explain NVIDIA B200 vs B300 specs, and show when a 36-month commitment pays off. Each guide uses real numbers. Want your own figures? Try the GPU savings calculator.
- 01GPU economics · 6 min read
The 30% rule: when leasing GPUs beats buying tokens from open-model inference providers
Paying per token for an open-weight model is the right way to start. But once your own GPUs would be busy about 30% of the time, a financed lease costs the same — and every hour of use above that line is close to free.
OCT 17, 2026 - 02AI, deployed · 6 min read
Kimi K3 and an agent harness: an AI chief of staff for your leadership team
Moonshot’s open-weight Kimi K3 is the brain. An agent harness is what turns it into an assistant that drafts memos, triages a calendar, and knows which actions need a human’s yes first — all on hardware your company controls.
OCT 16, 2026 - 03AI, deployed · 5 min read
GLM-Legal on private GPUs: a 100-person law firm plan
Z.ai's new legal model is a real step forward. Here's how a 100-person firm can run frontier-class legal AI on two private B200 nodes, at a fixed monthly cost.
OCT 14, 2026 - 04Hardware, explained · 4 min read
B200 vs B300: which GPU fits your workload
Same generation, different memory ceiling. The spec gap between B200 and B300 — and which workloads actually need the upgrade.
OCT 13, 2026 - 05Pricing, explained · 5 min read
Reserved vs on-demand GPU pricing, sourced
We pulled the public numbers on CoreWeave, AWS, and Lambda — then show why utilization, not sticker price, decides whether committing saves you money.
OCT 13, 2026 - 06Unit economics · 4 min read
Your GPU bill is part of your product economics
Inference compute isn’t overhead — it’s cost of goods sold. Once you see it that way, everything about how you buy GPUs changes.
SEP 30, 2026 - 07Infrastructure strategy · 4 min read
Baseload vs burst: the 80/20 of GPU infrastructure
Most of your GPU fleet is doing something predictable. Price it accordingly — and keep the cloud for everything else.
SEP 23, 2026 - 08Pricing, explained · 5 min read
What ‘qualifying deployments’ actually means
A plain-English explainer of the ~$5/GPU-hour target rate — and the five things that move your real price.
SEP 16, 2026 - 09Infrastructure strategy · 3 min read
If you run GPUs 24/7, you’re operating infrastructure
On-demand pricing is built for elastic workloads. If nothing about your fleet is elastic, you’re paying the wrong rate.
SEP 09, 2026 - 10Honest answers · 5 min read
The 3-year commitment question, answered honestly
When long-term GPU leasing makes sense, when it doesn’t — and the risks we’d talk you out of signing.
SEP 02, 2026