The big threat imo is smaller open source models running at the edge, not on expensive GPU clouds. Their lead rn is GIANT MODEL + AZURE GPU CLOUD. Anything that

The big threat imo is smaller open source models running at the edge, not on expensive GPU clouds. Their lead rn is GIANT MODEL + AZURE GPU CLOUD. Anything that undermines that is trouble, because it is so expensive. For a long time, everyone focused on training costs, but GPU inference at scale is mind-boggling. Current valuations do not seem to take this risk into account.