We asked @somewheresy: How do we get datacenters and new things built while still preserving culture and people's wellbeing? " In a very real materialist way, the Bay Area could start throwing more events in Oakland and talking to the community." "I think they have a responsibility to talk to people and hear them out and try and listen to them on what sort of solutions in the middle that we could pursue." "The alternative is that we don't build any electrical capacity, things continue to get more expensive, everybody has disemployment and an inability to make ends meet anyway, and we're all left off in a worse situation than we are currently." "It's incredibly important for us to go to these spaces where they're dealing with these very tangible effects of these build-outs and try and figure out how exactly we can retool resources into these communities to where at least the delta of their quality of life is not net negative, and at best, they're positively affected by these build-outs."

We push Prefill/Decode disaggregation beyond a single cluster: cross-datacenter + heterogeneous hardware, unlocking the potential for significantly lower cost per token. This was previously blocked by KV cache transfer overhead. The key enabler is our hybrid model (Kimi Linear), which reduces KV cache size and makes cross-DC PD practical. Validated on a 20x scaled-up Kimi Linear model: ✅ 1.54× throughput ✅ 64% ↓ P90 TTFT → Directly translating into lower token cost. More in Prefill-as-a-Service:

Cerebras — Faster Tokens Please OpenAI and AWS Partnerships, Tokenomics Explainer, Architecture Deep Dive, Datacenter Ramp, Technical Roadmap READ NOW: