I write when I have something worth saying.

That's the whole editorial policy.

Frontier AI labs are racing to build the biggest training clusters ever assembled. The race is not really about GPUs. It is about whether the rack is observable as one machine. Today it is not, and that is the bottleneck inside the bottleneck.

Over a decade I have shipped multiple product lines and made zero sales calls. The credibility came from the company I was working for. Now I am building an AI tool for a problem I was facing, and the marketing problem looks completely different.