Weighing the trade-offs of neoclouds and sovereign clouds

HPC on steroids

For organisations looking to deploy high-end AI capabilities, neoclouds offer an attractive proposition by offloading incredibly complex underlying hardware requirements.

“It’s like HPC on steroids, where the interconnection between GPUs, the network, and connection to storage needs to be highly optimised,” Wong said, referring to high-performance computing. “Otherwise, you’re starving your GPUs and you’re not maximising your return on investment. So, there’s a level of value that they provide from that perspective.”

The raw infrastructure capabilities of neocloud providers can bring cost advantages. Many provide cost savings of up to 60–70% compared with hyperscaler GPU instances, while offering near-instant access to the latest hardware generations, according to Mike Dorosh, senior director analyst at Gartner.

However, unlike hyperscalers like Google Cloud with Vertex AI and AWS with Amazon Bedrock, neoclouds don’t typically offer managed AI services.

“I can take advantage of AI capabilities from hyperscalers and use them in my organisation; I don’t necessarily need to be a data science expert,” Wong said. “It can be a lot harder when you’re starting off with a blank page. That’s the struggle that we’re seeing in the neocloud market at the moment.”

Spotty availability

Despite heavy investments from neoclouds and sovereign providers, the global availability of state-of-the-art GPUs remains inconsistent outside of the US.

Wong pointed to Nvidia-backed CoreWeave as an example. While it is one of the biggest players in the neocloud space, its GPU footprint outside the US is limited to markets such as Canada, Sweden, Spain and Norway. Similarly, European provider Scaleway has built out regions in Paris, Amsterdam, and Warsaw, but GPU availability is limited to one or two datacentres within those regions.

“They’re trying to build things out, but it’s very spotty,” said Wong. “That’s going to be very jarring for an enterprise used to dealing with a hyperscaler. If you’re looking at a neocloud or sovereign cloud provider, it’s significantly more inconsistent, and there’s a lot less documentation and transparency about what’s going to be available.”

Hyperscalers aren’t about to cede the market to neocloud and sovereign cloud providers without a fight. For one thing, they already provide services such as AWS European Sovereign Cloud and Oracle EU Sovereign Cloud to meet the demand for sovereign services.

Wong noted that sovereignty exists on a spectrum, and hyperscalers are trying to hit every point along it – from restricted cloud regions used by governments and the defence sector to on-premises deployments via services like Oracle Cloud Infrastructure Dedicated Cloud region, as well as hybrid environments via offerings such as AWS Outposts or Azure Local.

Ultimately, true sovereignty often comes down to geopolitics and business goals. While organisations in markets like Australia continue to rely heavily on US hyperscalers due to historically strong ties, those in Europe or the Greater China region are increasingly evaluating sovereign cloud services out of geopolitical caution.

Still, Wong advised that any move to a neocloud or sovereign provider must be grounded in business strategy and internal technical capabilities.

“If I'm hoping to take advantage of a whole bunch of capabilities, and then they aren’t even offered in a neocloud or a sovereign cloud provider, that can rule things out,” Wong said. “Some specialised cloud providers, neoclouds, and sovereign clouds just aren’t equipped or used to serving the needs of enterprise customers.”