WSS: Neocloud momentum accelerating on development and infrastructure side as sector transitions continue
A busy week saw more activity from the emerging neocloud space. Momentum is accelerating and providers like CoreWeave have moved to the head of the pack as early leaders. Success with a service provider operation is about differentiation and value-add, and the likes of CoreWeave often have to decide between building or partnering with third parties. This is a decision being made when it comes to data centre infrastructure – should I self-build of lease third party data centre capacity – but also comes into play when building management systems, along with features, products and services. Priorities, cost and capabilities all have to be considered. CoreWeave, for example, has looked to offload some of its cloud storage development to Backblaze and this week confirmed a new agreement that is worth $335m over five years.
The neocloud space continues to see new entrants. NVIDIA has seeded an ecosystem that continues to grow and AMD is also getting its service provider channel off the ground. A lot of the early stage development has been focused on training workloads, but the pivot to inference has started. NVIDIA realized it had to upgrade its capabilities in this area and effectively acqui-hired the team that built Groq and its LPU chips, while licensing the technology. With that business being absorbed into NVIDIA’s orbit, Groq is now looking to focus on the cloud infrastructure delivery business GroqCloud that it had already started building. To fund the operation and its expansion plans, Groq recently raised $650m and is targeting a data centre infrastructure footprint of 200MW by the end of 2027. GroqCloud will be looking to capture demand exemplified by the Anthropic cloud deal with Akamai, which is geared to inferencing requirements.
Neoclouds continue to need capacity to support cloud infrastructure builds in the face of meaningful demand levels. Is SpaceXAI going to be a neocloud business? That is not necessarily clear over the medium- or long-term. But what it is definitely doing is using its data centre capacity and silicon allocations to provide cloud infrastructure services to hyperscalers and AI labs. In the last several weeks, SpaceXAI has signed cloud deals with Anthropic, Google Cloud and just recently, open source AI lab Reflection. The capacity being contracted for is significant, but there are clauses for early termination. Meanwhile, neoclouds like Fluidstack continue to procure infrastructure for growth initiatives. Fluidstack has taken down hundreds of megawatts with multiple data centre colocation providers and recently committed to lease capacity from Hut 8 in west Texas.
Transformation across the infrastructure service provider landscape is ongoing. Rackspace, which has been around since the sector’s earliest days, basically came up with the managed hosting model, which evolved over the years into private cloud infrastructure. These services essentially pre-dated the rise of hyperscale public cloud, but when that space hit critical mass, Rackspace pushed into delivering managed public cloud services. Things have not stood still and another pivot is underway at Rackspace. It is always going to have a value proposition built around being a managed service provider, but is now looking to take that and focus on private AI infrastructure environments for regulated industries. A partnership with AMD is going to provide the technology foundation, and a 30MW deployment is being built to house these services. Rackspace has been relatively asset-light for a while, but with this new emphasis, is going to be a meaningful multi-MW data centre colocation customer again.
At the end of the month, we publish our monthly insights and touch on various themes around hyperscale demand and the competitive landscape. There continues to be steady and consistent momentum, and a steady stream of new company formation and establishment of consortiums and JVs, that speak to the optimism felt across the sector.
or