$METAMeta

Anthropic: Customer Concentration in $META @PatrickToulme points out the real risk to Anthropic is Meta switching onto its own coding models. I agree that this is a real risk for Anthropic but not AI compute buildout because instead of Anthropic buying bare metal, it will be Meta buying bare metal. Furthermore, I don’t think Meta can make this switching within a year because it doesn’t have enough compute secured. If Meta’s Watermellon model is that good, then it’s illustration where it’s about the compute capacity and not about the software. However what’s holding back the quality of Anthropic’s model quality is compute capacity. In real world usage, outside the benchmarks, it will be a race for compute capacity. Without compute capacity, having the best model doesn’t convert to re