LLMBOTTLENECK.COM
For GPU clouds and hardware makers

Be seen by people choosing what to run a model on

Every reader here has a model in mind and a question about the hardware for it. A sponsorship puts your name in front of that question, in a labelled box — and buys nothing else.

Placements

Cloud GPU page

/cloud-gpus: live rental prices for every card and what each runs. The reader is choosing a provider.

Model pages, rental section

Under the table of machines that hold a model, on every model page. The reader knows the model and is pricing it.

Local vs cloud

/local-vs-cloud: buy, rent or pay per token. The reader is deciding how to spend.

Terms

A flat monthly fee agreed in writing, one sponsor per placement at a time, a headline of up to 80 characters and one line of text. Traffic and click figures for the placement are shared before you sign.

What a sponsorship never buys

  • A position in any table. Rental tables are ordered by the provider’s own live price, and that does not change.
  • A recommendation. The recommended machine is the lowest cost per token at 20 tokens a second or more, for everyone.
  • A figure. Memory, speed and evidence labels come from the engine and the published sources, and are the same with or without you.
  • An unlabelled mention. Every sponsored box says Sponsored, names who paid, and links here.

The code enforces the first three: the ranking and the engine never read the sponsorship file, and a test fails the build if they start to. Referral commissions are separate and published on the cloud GPU page.

Ask about a placement →

llmbottleneck
catalogue 2026-10-03models 327devices 135