LinkedIn·Tuesday, 25 August 2026·1d ago
We went to look at what it would take to run our own models rather than renting capacity, and treated it as a financial decision rather…
madewithlove
3,118 followers
We went to look at what it would take to run our own models rather than renting capacity, and treated it as a financial decision rather than an engineering one.
The number quoted for a single server with eight of the latest-generation GPUs was around half a million euros. Set that against what the same compute costs as rented credits and owning looks absurd, yet it remains a good business for the companies buying them. The gap between those two figures is worth understanding before you commit either way.
💬 1
View on LinkedIn Cross-referenced
Related on the wire
Data centres used to be sold on how many racks were free. That conversation now runs in megawatts, and sometimes gigawatts. If you are…
We run a private leadership dinner in Belgium once a quarter. No stage, no slides, just a table of CTOs and founders from the SaaS world…
Most people who commission a technical due diligence report have never actually seen one. You sign a term sheet, the clock starts running,…
By 2027, writing code by hand might be a weekend hobby. Like woodworking, or growing your own veg. Pleasant, and quietly beside the point…
Every year we go through the hardware in our own office, and this year it turned into a real question: should we run our own models here…
Flat-rate AI was always a promotional price. Budget for it the way you budget cloud compute, not the way you budget software seats. A meter…