Nvidia CEO Jensen Huang said he had urged UK-based neocloud provider Nscale to “work around the clock” to get delayed data centers online. In a separate CNBC interview at GTC Berlin, he described demand for neoclouds as “off the charts” and pointed to working capital as a constraint. Nscale’s project delays show why confidence in demand is not the same as confidence that new capacity can be delivered on time.
49
54
Huang’s call for speed—and his case for neoclouds
Huang’s instruction to Nscale was a call to accelerate delivery after setbacks at some of its sites, according to Bloomberg.
49 In separate remarks to CNBC, he argued that demand for neocloud services—the AI computing capacity these providers rent out—was exceptionally strong, while working capital was in short supply.
54
Those points address different parts of the business. Strong demand may support the case for building more capacity, but operators still need to complete facilities, secure power and finance construction before they can supply that capacity.
Nscale’s delays are about delivering capacity
The Loughton project in Essex illustrates the gap. Reuters reported that Microsoft withdrew from its agreement to anchor the site after its expected power connection slipped into the 2030s. Microsoft described the decision as relating to that single site.
17 The report points to a power-delivery problem at Loughton; it does not establish that demand for AI computing has disappeared.
Nscale’s Norwegian flagship project faced a different set of execution challenges. Bloomberg reported that it was about six months behind schedule, adding to questions about how quickly the company could bring planned capacity online.
18
Demand does not remove financing risk
Investor skepticism adds another constraint. Bloomberg reported pushback against Nscale’s reported $30 billion IPO valuation target, with some investors suggesting a lower range of $16 billion to $19 billion. Those figures reflect reported expectations and investor views, not an achieved public-market valuation.
21
At the same time, Nscale and Anthropic were reported to be discussing a potential cloud-computing agreement worth almost $45 billion. The talks were not a completed deal.
20 The contrast is instructive: substantial potential demand can coexist with uncertainty about project delivery and how investors value the business.
What Nvidia’s DSX POD can—and cannot—address
At GTC Berlin, Nvidia presented DSX POD, a reference design for Vera Rubin AI factories that partners assemble and validate before delivery. Nvidia says that, once a site’s utilities are ready, the approach can reduce on-site deployment from about six months to as little as 14 days.
10
That is a claim about deployment at a prepared site—not a promise to supply missing grid power or remove every construction and financing obstacle. Faster installation can help with one part of the timeline, but a site still needs the infrastructure to operate.
The bottom line
Huang’s comments express confidence that neocloud demand is strong and urge Nscale to execute faster. The reported setbacks at Loughton and in Norway underline the caveat: demand must be matched by sites that can be powered, completed and brought into service. The available reporting documents that tension; it does not show that Nscale’s delays have been resolved.
17
18
49
54