One interesting thing that Anthropic did was putting their stack on the various cloud providers, I wonder if they'll put it on GCP and Azure next since they've put it into AWS first at a level we have not seen a major AI provider do to date. Your company can have their own Claude stack just like an ELK stack on your cloud, if they can do this for both Azure and GCP then OpenAI has to really catch up.

In my eyes I would rather use the AI I can run on my own paid infrastructure, so if there's an outage its isolated, or I could potentially have a different region / DC to fallback on.

I'm still surprised that neither Microsoft nor Amazon have made their own models available on their cloud offerings. I guess Microsoft probably does have Phi on there, but it's not front and center, especially with something like Copilot for Devs (seriously Microsoft rebrand that damn thing to be clear what you mean by Copilot!) where they could use the cheaper compute by using something like Phi.

▲

SubiculumCode an hour ago | parent | next [-]

The recent deal with SpaceX AI to use their severely underutilized GPU compute is pretty telling to me. Being able to roll out compute is a hardware problem, rolling out good models needs more than compute, it needs good AI engineers. SpaceX, Amazon et al can do hardware very well. AI engineering, maybe not so much.

▲

NitpickLawyer an hour ago | parent | prev [-]

Claude is already on Vertex - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/p...

▲

giancarlostoro an hour ago | parent [-]

Is Vertex hosting it themselves on their own hardware? Because GCP, Azure and AWS all host it on their own hardware, but AWS lets you do everything that Claude lets you do if you use their APIs directly, and then some. This is what I want to see on Azure and GCP, and heck, maybe even DigitalOcean, if they ever stop expanding into so many spaces and focus on improving their current infrastructure, before I fully migrate off of them.

	▲	NitpickLawyer an hour ago \| parent [-]
		> Is Vertex hosting it themselves on their own hardware? Yes, you can even choose regions, for EU they serve it from Belgium. With all the encrypted at rest stuff and other guarantees that vertex provides. > Important: Accessing Claude models through Vertex AI meets the FedRAMP High requirements, and operates within the Google Cloud FedRAMP High authorization boundary.