I think they quickly reached the limits of CPU inference and that's why they decided to scale up with GPU's. Because of how they operate, they can't just relea

I think they quickly reached the limits of CPU inference and that's why they decided to scale up with GPU's. Because of how they operate, they can't just release in a single DC like other providers. It's a partial "all or nothing" scenario.