WSS: NVIDIA and Meta developments point to current state of long-term hyperscale strategies
The sector is back in full swing and the week saw developments on the hyperscale infrastructure side of the game that may have interesting implications for what happens down the road running into the next decade. The sector has been driven by a small number of hyperscale cloud infrastructure platforms. We are tracking eight of them on a regular basis and they include four US-based and four China-based hyperscalers. An important question to consider is how many more hyperscale platforms might there be in the future. Are NVIDIA and Meta possible additions to the hyperscale segment? Certainly NVIDIA is a possiblity and we have discussed this at length in our commentary. But recent developments suggest the road to hyperscale status is not moving quickly. There is reporting pointing to an internal reorganization that is folding the cloud-oriented groups into the larger engineering teams. It looks like NVIDIA is going to stick to its service provider channel strategy and direct service delivery is not currently a strategic priority.
Meta has not played in the compute infrastructure game, but builds data centres and infrastructure in order to enable its various social media platforms and businesses. That might look to be changing. In the past week, Meta launched a new initiative called Meta Compute that will build out gigawatt-level infrastructure. Is that all for internal infrastructure purposes or will Meta consider opening it up to extermal customers? The latter seems like a reasonable possibility. If Meta can scale quickly, it could potentially have enough resources to offer infrastructure up for external usage. And it might make strategic sense to begin offering compute to build out an AI ecosystem around its properties. Energy is of course a bit part of building out compute infrastructure at scale, and Meta also made moves with that goal in mind. Meta signed a number of deals to procure up to 6.6GW of energy from nuclear sources by 2035. Google is also planning ahead and partnering directly with energy providers for gigawatt-level resources.
Earnings eason is just around the corner and Applied Digital has a slightly different reporting cycle and released its latest results last week. Applied Digital confirmed plans to spin out the cloud infrastructure business to focus on data centres and is starting to book revenue from its Polaris Forge 1 campus in North Dakota. A second hyperscale client has been signed and Applied Digital is getting closer to signing a third. We have details in our analysis of Applied Digital results.
Applied Digital got the ball rolling with hyperscale and neocloud tenants with the signing of a 200MW deal with CoreWeave for the Polaris Forge 1 site. The neocloud part of the market remains busy. CoreWeave’s recently CEO shared some data points about customer cconcentration, while reporting came out that Lambda is working on another capital raise as it gets closer to a possible IPO.
Elsewhere, Applied Digitial was not the only operator making strategic moves. Singapore-based DayOne and UK-based Global Technical Realty both raised in the area of $2b for growth and expansion initiatives, while maincubes also garnered resources and Cloudflare made another technology-oriente acquisition with the purchase of Replicate.
or