WSS: The end of hyperscale has come and GPUs and AI are driving the shift in end user profile for data centre colocation
Another busy week saw more capital raising to support expansion activity as the profile of data centre colocation end users continues to expand and begs the question: is the end of hyperscale (as it relates to dominating data centre colocation demand) upon us and is it a good thing? The answer should be obvious. We look at why the end of hyperscale dominance is a good thing against the backdrop of more activity in the GPU cloud space. This demographic is increasingly known as the neoclouds. Not a great name, but one that seems to be catching on.
Speaking of neoclouds, Crusoe, Nebius and Lambda were busy in recent weeks. Crusoe secured a $750m credit facility from Brookfield Asset Management and leased 12MW from Polar in Norway where it plans to house its cloud infrastructure service Crusoe Cloud. Lambda has been busy as well and stood up GPU-based infrastructure built on NVIDIA in Cologix’s hyperscale data centre in Columbus, Ohio. The service provides on-demand AI infrastructure for enterprises that want to avoid CapEx and procure in a service-based delivery model. Meanwhile, Nebius in the Netherlands continues to expand its footprint, with a next stop in the UK. Nebius raised $1b in convertible note Nebius Group raises $1b in convertible notes for expansion activitys for its expansion initiatives.
There were other developments in the GPU cloud space. OpenAI continues to diversify its hyperscale cloud partnerships. It started with Microsoft Azure and is now using Oracle Cloud quite extensively, along with CoreWeave. OpenAI added Google Cloud to that list in recent weeks as it looks to add even more capacity. Interestingly, OpenAi will run on both Google Cloud’s platform, along with the infrastructure it is making available through CoreWeave’s cloud.
GPU cloud expansion activity has been driven by the NVIDIA ecosystem, but there has been more diversification of late. In the past week, we saw DigitalOcean roll out a new GPU cloud service based on AMD Instinct chips, while Crusoe is doing the same for its cloud infrastructure service.
The activity across AI infrastructure marks an ongoing shift in the hyperscale data centre colocation business, which has been focused on a very small group of buyers that everyone has been in pursuit of. That target customer profile has constantly tried to reduce the number of colocation vendors it works with (with varied degrees of success), while also going it alone and self-building. Those paths make it harder for hyperscale colocation to be a part of the value chain. The sector has responded by moving into other places within the value chain, such as the pre-development phase and real estate-oriented leasing of powered shells or build-to-suit arrangements. But over the long-run, it was always going to be a good if there was more demand out there for hyperscale-level colocation, and not just a ‘big three’ or ‘big four’. And that is what is finally happening as the hyperscale category has added the likes of CoreWeave and literally more than a few dozen companies from the GPU cloud space (and other segments like social media and AI technology companies) have emerged that could realistically take down multi-MW capacity with a colocation provider.
BtyeDance is one of those emerging as a legitimate global hyperscaler, with a demand profile potentially on par with the biggest hyperscale clouds, but are there others? NVIDIA seems to be the one that could step up and reach that status. The recent launch of a marketplace that will see it resell GPU cloud service inventory is a small step in that direction. NVIDIA quietly acquired a company called Lepton in recent months and that brought in the technology to enable the marketplace and other reselling and service aggregation capabilities. How might this play out? We have discussion and details in our commentary.
Needless to say, hyperscale continues to be the engine of growth behind the sector. We saw in earnings season how CapEx is not being dialled back despite various uncertainties in the macro environment, and the likes of AWS and Microsoft are continuing to invest heavily in date centre infrastructure, skills and training. AWS is set to invest another $20b in Pennsylvania and the data centre infrastructure is going to be enabled by nuclear energy. AWS acquired a data centre site from Talen Energy last year and after hitting some regulatory hurdles, completed a new PPA that will get it nearly 2GW of carbon-free nuclear energy to the site acquired from Talen and two additional ones in PA. A three-AZ region in PA looks to be in the works and the scale and scope look to be at another level entirely, boosted by the coming requirements around AI. There were also other developments around nuclear energy in recent weeks. Meta signed a PPA for nuclear energy with Constellation and Google and Elementl partnered for nuclear site development as well.
Finally, there was plenty of capital raising and financing activity in the past week from a number of operating companies. This included Vantage Data Centers, DayOne and Ada Infrastructure, along with Crusoe and Nebius (mentioned above). CapEx investments remain on its current course and capital continues to pour into the sector, confirming bullish sentiments and dismissing any notion that the sector is on the cusp of an overbuild.
or