Cerebras CS-4

Post by Mattwmaster58 » Wed, Aug 19, 2026, 1:04 PM UTC

aurareturn wrote: post #157

doctorpangloss wrote: post #42

Earlier quoted context omitted.

it only takes ~445 GB300 NVL72 (about $22b) to run ALL of openrouter demand for a year. Microsoft rolled out $32b of DC 2026Q1. imo the issue is that most openrouter demand is inauthentic activity (things that anthropic and openai models will refuse to do like pretend to not be bots when interacting with humans)

I thought your numbers must be wrong. So I plugged 288 trillion tokens/month (OpenRouter's current rate), 500 billion MoE model average, and the math comes out to be around 620 B200 GPUs minimum. So basically, OpenRouter's volume must be absolutely tiny compared to the volume hyperscalers are getting.

Post by alightsoul » Wed, Aug 19, 2026, 1:29 PM UTC

zarzavat wrote: post #126

alightsoul wrote: post #106

Earlier quoted context omitted.

if you are developing your own hardware, you provide your own stack to avoid lawsuits with nvidia. i don't think it's a technical problem at all, but a legal one. this is probably why zluda was scrapped by AMD and Intel. Nvidia technically bans the creation of CUDA reimplementations in their TOS if i remember correctly

Doesn't Google v Oracle provide protection here? Copying APIs is fair use. If it's patents that are the problem then presumably all these large semiconductor companies have defensive parent portfolios.

Post by ceejayoz » Wed, Aug 19, 2026, 1:34 PM UTC

VladVladikoff wrote: post #76

ceejayoz wrote: post #31

Earlier quoted context omitted.

There’s been a spate of Reddit AI bots using all lower case in hopes of evading detection. It’s still incredibly obvious.

I don’t really visit Reddit much these days but would love to see an example.

Post by bearjaws » Wed, Aug 19, 2026, 1:42 PM UTC

Post by alightsoul » Wed, Aug 19, 2026, 2:03 PM UTC

zozbot234 wrote: post #131

riknos314 wrote: post #124

Earlier quoted context omitted.

Kimi K3 is a 2.8T model that's available at about 1/4-1/3 the cost of Fable from multiple providers on openrouter. The math doesn't seem wildly off.

The raw margins on proprietary model inference are rumored to be quite high though (they have to successfully defray the entire investment into model training and datacenter capacity for inference, which is massive enough). The API cost you're paying for the model includes that raw margin.