Introduction
I am thrilled to announce that Nscale has raised $155m in Series A funding led by existing investor Sandton Capital Partners, with participation from Kestrel 0x1, Blue Sky Capital Managers Ltd, and Florence Capital.
At Nscale, our mission has always been to provide sustainable, cost-effective, AI infrastructure at scale. Our unique fully vertically integrated approach enables us to design and deploy large-scale compute clusters at our own AI-ready DCs to ensure we deliver unrivalled speed, time to market, and value to our customers.
Since launching from stealth in May 2024, Nscale has experienced insatiable demand for AI infrastructure. To capitalise on this vast and growing opportunity, we have expanded our pipeline of greenfield data centres across Europe and North America from 300MW to 1.3GW.
This oversubscribed funding round is a major milestone for Nscale. Not only is it one of the largest Series A rounds in UK/European history, it also demonstrates strong confidence in our fully vertically integrated approach and signals investors' belief in our mission of building the AI hyperscaler, purposely engineered for AI.
Why We Are Different
In the process of building Nscale in Q3 2023, initially, my intention was to build out a large data centre business utilising a number of low-cost power assets in Europe and North America that we had acquired over the preceding 2 years. As such, I spent a lot of time profiling the market for AI Customers in addition to the emergence of a new type of potential DC customer, the “Neoclouds”.
As I looked further into the Neocloud model it felt like an incredibly risky business from a counterparty perspective. Broadly what we found is that the average Neocloud is 80-90% leveraged with an average cost of capital of 15+%. Most Neoclouds do not own their own data centres and as such pay high data centre costs. Many Neoclouds do not own their own cloud software or orchestration software and license this from third parties. Lastly, a large number of Neoclouds rely on spot instance or merchant market compute and so in spite of all of these high costs, they do not have consistent revenue.
Viewing this from a customer perspective these businesses are thin margin, high risk and have too many segments of the value chain that present deal risk to a customer wanting a guaranteed quality of service. Looking objectively at this back in Q3 2023 I came to the conclusion that Neoclouds presented significant challenges. However if this AI Demand curve is indeed the fourth industrial revolution like many claim, then how would you build a resilient AI Cloud Platform that is highly profitable, downside protected and provides the lowest cost service to the customer at scale?
After investigating this, I believe the ONLY way to do so is to build a fully vertically integrated platform. One that owns the Data Centre, the GPU cluster and the software. Building this vertically integrated structure is the antithesis to the Neocloud model, and can be characterised as the “Hyperscale Model”.
Nscale's vertical integration enables more than just the security of power supply and data centre capacity in a highly competitive market. Deploying these superclusters within our own custom-designed facilities in the lowest cost power markets, with lower leverage (60-70% compared to the market average of 80-90%), ensures we can deliver a bespoke cluster at almost any scale, all while delivering far superior unit economics to our customers. This cost advantage is becoming very important as we enter the large cluster market in 2025, where customers are faced with both the scarcity of large contiguous power availability and a more competitive GPU landscape. Cloud competitors that do not have control over their power, data centre availability and their value chain will fall behind in 2025.
By contrast, Nscale has full control over every segment of the value chain, and as such our cost of production for compute is at least 10% lower than our competitors. Enabling Nscale to always offer a lower-cost service at any scale.
Nscale is the hyperscaler engineered for AI, the antithesis to the Neoclouds and we are here to provide a higher quality, lower cost service to the market at scale.
The Opportunity Ahead
As we look ahead into 2025, we see two key trends playing out in this market. The first is the size of the clusters required by the larger customers is increasing exponentially, which gives rise to the scarcity of large contiguous power availability in 2025. The second is the fundamental change to the density requirement at the rack level with the new NVIDIA GB200, B200 and AMD MI325X GPU releases that will render a large number of available DCs unsuitable due to the low rack density and lack of liquid cooling.
Both of these trends result in a significant increase in scarcity for those that require large clusters of new-generation GPUs in 2025 and 2026.
Many, if not all, of Nscale's competitors deploy AI infrastructure at third-party colocation centres. They do not build, own, and operate their own data centres. This means they pay expensive rack/colocation fees which are increasing 13% YoY according to CBRE and JLL and cannot accommodate large 10,000+ GPU clusters which require 20 MW of contiguous capacity that can support high-density (150kW per rack) GPU clusters. This means they are completely beholden to the availability and good graces of the colocation market in order to service customer needs.
Nscale is prepared for this, and has spent the last 3 years consolidating a pipeline of 1.3 GW of greenfield data centre sites in some of the lowest-cost power markets in Europe and North America, Nscale is now uniquely positioned to capitalise on this rapidly growing market segment for large 10,000+ GPU clusters. We are uniquely positioned in that we can design these sites from the ground up to support large, bespoke superclusters for global hyperscalers, LLM platforms, and governments requiring large-scale sovereign compute capacity. These sites will leverage closed-loop direct liquid cooling to deliver maximum performance and efficiency and will benefit from the lowest cost power on earth, enabling us to deliver bespoke superclusters at 100,000+ scale while delivering superior unit economics to our customers.
What’s Next
The capital from this raise will accelerate the deployment of another 4,000 GPUs at our Glomfjord data centre in Norway, and the definitive design of 290 MW of purpose-built AI DCs across Europe and North America. 60MW in Glomfjord Norway (2025), 180MW in Texas USA (2025), 50MW in London United Kingdom (2025).
Nscale expects to develop 500 MW of AI Cloud capacity over the next 3 years and deploy 50,000 GPUs by the end of 2025, 150,000 by the end of 2026, and 350,000 by the end of 2027 as it builds the full-stack hyperscaler, engineered for AI.
Other things you can expect from us in 2025 are the launch of our inference service in January, enabling consumers to run inference services directly from the Nscale platform. Nscale’s Public Cloud service is launching in February 2025.
If you’re innovating at the forefront of AI, please get in touch here.