5 stories · “inference”.
Meta will deploy its MTIA 450 inference chip, designed with Broadcom and built by TSMC, in data centers in the first half of 2027, as it scales custom silicon.
Z.ai says an AI agent built the production inference stack for GLM-5.3-Flash on more than 100,000 Chinese accelerators, tripling throughput in under two weeks.
TypeSafe AI left stealth with $40 million and a model called Jev that returns typed probabilities instead of prose. It is faster and cheaper than an LLM, but it is not trying to be one.
Bigger isn't always better. Small language models are getting shockingly capable, and they run on the hardware you already own.
Sparse mixture-of-experts models activate a small fraction of their parameters per token. How routing, capacity and serving shape the largest AI models.
