Microsoft Leans on Neoclouds to Ease AI Crunch

Bloomberg

Watch Microsoft Leans on Neoclouds to Ease AI Crunch - Bloomberg on Dailymotion

Transcript

00:00There's been $33 billion spent by Microsoft on NeoClouds.

00:04You're talking European players like N-Scale, Neveus.

00:07There's also local player CoreWeave.

00:09Why would the AI data center, like, renter-outer

00:14that we all know for Azure be turning to others?

00:17Yeah, it's kind of funky, right?

00:18It's like if I was paying somebody to write stories

00:21and I was still writing stories.

00:22I mean, it's like, essentially,

00:24because we have huge capacity constraints, right?

00:27Microsoft needs to get as many chips online as it can,

00:31both for its customers, itself, and OpenAI,

00:34and it's needing to kind of pull every single lever it can.

00:37And so it's emerged as a major customer for all of these names

00:41like the CoreWeaves and the Neveuses,

00:43which have become, you know, very newsy in recent months.

00:47It's like managing assets, right?

00:49And, you know, the point of a NeoCloud is it's dedicated to AI,

00:53either training or influence and, you know,

00:55storage, running other software.

00:57You know, that's a different thing.

00:59Talk to me about the figure of $33 billion, though.

01:01Like, is that news?

01:02Is it something that we were able to work out

01:04because Microsoft doesn't disclose it

01:06in, like, quarterly earnings or something like that?

01:08So Microsoft has just kind of disclosed it in piecemeal ways,

01:11and likely it is going to be much higher than that 33.

01:15What we have been able to discover

01:17is what this capacity is actually being used for.

01:20And in many cases, it's for Microsoft to build their own AI models.

01:24And that's surprising because it's a larger amount of investment

01:28in their own internal AI teams than many had realized.

01:31And it points to them saying,

01:33man, we better catch up with the open AIs and Anthropics

01:36and have our own models on hand,

01:37and we're going to use the NeoClouds to do it.

01:39And Mustafa Suleiman, who came from DeepMind, went to Microsoft

01:43with an, obviously, inflection being bought in this rather odd way.

01:46He's the man behind the consumer AI offering.

01:49He's, I mean, you found out that basically

01:51the first large language model they're building internally

01:53under him has been using CoreWeave's assets

01:56over in Oregon, I think it is.

01:58Which is another way of saying they're using NVIDIA gear to do it.

02:01Yeah, using NVIDIA via CoreWeave.

02:04Right.

02:04But also what's interesting is the numbers

02:07that you have to orientate yourself around,

02:09because basically it allows Amy Hood

02:11not to have to write this all as capex.

02:13That's a really important point, right?

02:15If you buy a bunch of servers,

02:17now you have to depreciate them.

02:19Now it's on your capital expenditures,

02:21not your operating,

02:22and investors want to see a good balance there.

02:24And, you know, Microsoft's then able to,

02:26when it's renting from Neoclouds,

02:28to say, if in five years,

02:30we actually don't really need that many GB300s,

02:32we'd rather use some more Vera Rubens, bye-bye.

02:35We don't need to deal with all these servers

02:37that we don't have necessarily a use for.

02:39That's what it allows them to do, potentially.

Category

Transcript

Be the first to comment

Recommended