Microsoft gives data centre hopefuls a reality check
Nothing interferes with a company’s plans to go public like a large customer getting cold feet.
Neocloud Nscale, which is targeting a New York listing at a $35bn valuation, will need to rethink its assumptions after Microsoft pulled the plug on a deal to lease computing capacity at a UK data centre.
The US tech giant called time on its deal because the Essex data centre — a smallish part of its total spend with Nscale — is unlikely to be supplied with power until after 2030.
That is not just a blow to Nvidia-backed Nscale, which counts former Meta executives Sir Nick Clegg and Sheryl Sandberg as directors, but to its peers too, as it exposes the multiple faultlines running through this fast-growing slice of the AI sector.
Neoclouds rely on multiple moving parts: a power supply is a must for securing fat long-term contracts, but so are chips, infrastructure and multi-faceted financing, much of which is required upfront.
Once all those components are in place, the resulting agreements can run into big numbers: the value of Nscale’s contracted revenue, in effect its order book, has crossed $100bn, roughly 40 times the value of active business as at end-August.
There is a lot of execution risk nestled in that figure. No power or chips; no dice. Those contracts rely on the deployment of 461,000 GPUs; or almost 20 times what Nscale currently has.
Chips in today’s warp speed technology have a limited shelf life. Nscale, like its peers, reckons it can recontract obsolete chips to smaller enterprises, say, or to those looking to handle more mundane tasks. But there is no precedent for this.
With an eye to bottlenecks and supply constraints, clients have been careful to tie down their termination rights. Frontier lab Anthropic, which together with Microsoft fills out near90 per cent of Nscale’s backlog, gives only “limited relief” in the event of supply-chain delays.
Its tranche-by-tranche delivery schedule allows for termination at regular steps if deadlines are missed. Some other contracts are on-demand, and could vanish at any time.
Still, none of these risks are especially new; Nscale delineated as many in its pre-listing filing. So assuming its proposed IPO goes ahead — bold call as that would be given the number of data centre listings that have fallen by the wayside — how should investors value it? Modestly, is the answer.
Revenue clocked in at $141mn in the first half; assuming its stonking first-half growth rate continues for the rest of 2026 gives full-year sales of $420mn. Putting that on the same multiple as peer Nebius would imply a valuation of just over $6bn — and not even that if using CoreWeave’s measly multiple. The US neocloud’s shares have halved in the past six months.
Neoclouds’ failure to flick all the necessary switches will shunt more sales over to their bigger brethren — data centre builders and operators with capacity already available, or sufficient clout and capital to get things done on time.
For hyperscalers such as SpaceX, this is a boon. It’s not just the meteorological clouds that carry a silver lining.
Microsoft gives data centre hopefuls a reality check
Nothing interferes with a company’s plans to go public like a large customer getting cold feet.
Neocloud Nscale, which is targeting a New York listing at a $35bn valuation, will need to rethink its assumptions after Microsoft pulled the plug on a deal to lease computing capacity at a UK data centre.
The US tech giant called time on its deal because the Essex data centre — a smallish part of its total spend with Nscale — is unlikely to be supplied with power until after 2030.
That is not just a blow to Nvidia-backed Nscale, which counts former Meta executives Sir Nick Clegg and Sheryl Sandberg as directors, but to its peers too, as it exposes the multiple faultlines running through this fast-growing slice of the AI sector.
Neoclouds rely on multiple moving parts: a power supply is a must for securing fat long-term contracts, but so are chips, infrastructure and multi-faceted financing, much of which is required upfront.
Once all those components are in place, the resulting agreements can run into big numbers: the value of Nscale’s contracted revenue, in effect its order book, has crossed $100bn, roughly 40 times the value of active business as at end-August.
There is a lot of execution risk nestled in that figure. No power or chips; no dice. Those contracts rely on the deployment of 461,000 GPUs; or almost 20 times what Nscale currently has.
Chips in today’s warp speed technology have a limited shelf life. Nscale, like its peers, reckons it can recontract obsolete chips to smaller enterprises, say, or to those looking to handle more mundane tasks. But there is no precedent for this.
With an eye to bottlenecks and supply constraints, clients have been careful to tie down their termination rights. Frontier lab Anthropic, which together with Microsoft fills out near90 per cent of Nscale’s backlog, gives only “limited relief” in the event of supply-chain delays.
Its tranche-by-tranche delivery schedule allows for termination at regular steps if deadlines are missed. Some other contracts are on-demand, and could vanish at any time.
Still, none of these risks are especially new; Nscale delineated as many in its pre-listing filing. So assuming its proposed IPO goes ahead — bold call as that would be given the number of data centre listings that have fallen by the wayside — how should investors value it? Modestly, is the answer.
Revenue clocked in at $141mn in the first half; assuming its stonking first-half growth rate continues for the rest of 2026 gives full-year sales of $420mn. Putting that on the same multiple as peer Nebius would imply a valuation of just over $6bn — and not even that if using CoreWeave’s measly multiple. The US neocloud’s shares have halved in the past six months.
Neoclouds’ failure to flick all the necessary switches will shunt more sales over to their bigger brethren — data centre builders and operators with capacity already available, or sufficient clout and capital to get things done on time.
For hyperscalers such as SpaceX, this is a boon. It’s not just the meteorological clouds that carry a silver lining.
Comments
Post a Comment