A new class of provider: the “neocloud” is challenging hyperscalers and testing whether regional telecoms operators can reinvent themselves for the GPU era. With governments pushing sovereign AI strategies and enterprises racing to deploy models on local data, the question is not whether demand will arrive, but who will capture it. 

Neoclouds are distinct from traditional cloud providers. As Joseph Attwood, research analyst at Analysys Mason, explains, “we don’t define telcos as neoclouds, we consider neoclouds to be something slightly different. For us, a neocloud is a company that hasn’t been providing cloud services traditionally and then jumped straight into offering GPU as a service. The biggest examples are CoreWeave, Lambda, Nebius.” 

Their emergence has been driven by economics. Hyperscalers dominate conventional cloud, but the AI boom has created a supply-constrained GPU market where agility matters more than legacy scale. “Neoclouds’ main competitors are really going to be hyperscalers,” Attwood says. “Because of supply and demand economics, there’s a constrained supply of GPUs and very high demand – hyperscalers have fairly high prices. Neoclouds, looking to grow their businesses aggressively, are willing to undercut the prices and offer the best rates.” 

Unlike hyperscalers, neoclouds carry none of the overhead of decades of enterprise IT. “They have just the one business model,” Attwood notes, “so they’ve got low operational costs.” That lean structure has allowed dozens of start-ups to enter the market, yet it has also raised questions about sustainability. “There’s 200-plus potential neoclouds,” he says. “Over the next few years, we’re going to see consolidation. A lot of the business models are not sustainable.” 

Telcos: Trusted, but not natural neoclouds

For telecoms operators in the Gulf, the opportunity is clear, but the path is less so. Attwood argues that telcos are not natural neocloud leaders. “They don’t have the scale of hyperscalers, and they don’t really have the cost base of neoclouds. But what they do have is the trusted relationship with customers.” 

That trust can be crucial in highly regulated sectors. “For use cases where privacy or sovereignty is a big concern, a lot of companies aren’t going to want to work with neoclouds or hyperscalers because they won’t fulfil the requirements,” he says. “Telcos have been offering highly regulated services for decades, so they’re suitable players.” 

An analyst who works closely with neocloud operators in the Middle East is more sceptical. “Absolutely not, certainly not in the Middle East,” he says when asked if telcos are natural leaders. “Neocloud players need data centre capacity above all else, and there are hardly any telco players with DC expertise or capacity to cater to the needs of neoclouds.” 

AI infrastructure is stretching data centre design beyond historic norms. “Neoclouds need DC to support very high demands of AI compute, GPUs, and next-gen AI chip sets that require unheard-of kilowatts per rack that no DC operator ever thought would be necessary,” the analyst says. “Also, the heat considerations, air-cooled versus liquid-cooled, are a major variable.” 

Regional groups have already separated those capabilities. “The major regional telcos understand this dynamic and carved out their DC investments to create specialised entities,” he adds, pointing to center3 and stc in Saudi Arabia, and Khazna (G42) and e& in the UAE. “Neoclouds need DC first and foremost. It is a foundational infrastructure that enables neoclouds.” 

Infrastructure: Reuse has limits

Many operators hope to repurpose existing assets, but Attwood warns that AI is different from traditional hosting. “Some operators still have their data centre infrastructure, so there’s a case to deploy GPUs in this existing infrastructure,” he says. “There are limits, because AI data centres have greater cooling and power requirements. Even if you have floor space, you might need to refit the data centre.” 

Edge computing is often touted as an advantage, yet its role in generative AI remains uncertain. “For inference where latency might be a concern, there’s potential,” Attwood says. “But when you talk about large language models, the time to compute your token is several seconds anyway, so having a GPU slightly closer doesn’t really matter as much.” 

Who will buy?

Enterprises are expected to be the largest buyers of GPUaaS, but the Middle East adds a strong sovereign dimension. Attwood sees a mixed market. “Enterprises are the biggest because they’ve got larger budgets, but telcos will be targeting privacy and sovereign use cases – healthcare, financial services, banking.” 

Governments will remain pivotal. “Some are going to want sovereign cloud from the telco, and others will want sovereign solutions from hyperscalers,” he says. 

The regional analyst views the segmentation differently. “Neoclouds are emerging players focused on the AI stack, training and inferencing and SaaS. Telcos focus on connectivity infrastructure,” he says. Demand will come from “AI startups, academia and governments”. 

Sovereignty: Advantage or illusion?

Data residency has become a central plank of national AI strategies. Attwood believes it offers telcos a window, but not a permanent moat. “Sovereignty is going to be differentiated for telcos,” he says. “But when people talk about sovereign infrastructure, they mean different things.” 

Some enterprises simply require in-country GPUs, which hyperscalers can also build. “That might be a short-term differentiator,” Attwood cautions. The tougher bar is ownership. He cites concerns around the US Cloud Act: “There are players uncomfortable working with a cloud provider unless it’s headquartered in the country. Telcos have their advantage there.” 

The regional analyst agrees on the importance. “Yes, it is a major factor due to the ever-changing regulatory landscape as countries continue to isolate and build digital walls.” 

Barriers to entry

Both interviewees see formidable hurdles for operators. “Infrastructure costs are definitely going to be a big one,” Attwood says. Neoclouds funded by venture debt can spend aggressively, but “if a telco is doing that, it’s an unsafe business model for their shareholders.” 

Ecosystem maturity may be a greater challenge. “Enterprises already have a lot of their data on hyperscalers,” he notes. “There’s a real data gravity argument; if I’m using AWS to store my data, I might use AWS GPUs to train models because I don’t have to transfer it.” 

The regional analyst lists obstacles starkly: “All of the above. Traditional telcos must qualify customer demand -is a multi-tenanted DC or built-to-suit fit for purpose before making the enormous capex necessary.” 

Geography matters

Momentum is uneven across the Gulf. “UAE I’d say leads the way, closely followed by KSA,” the regional analyst says. “Qatar and others will face impossible odds to support neoclouds given market size and regulations.” 

Attwood also points to Saudi Arabia and the UAE as focal points, reflecting their rapid data centre build-outs and national AI programmes. 

Compete or collaborate?

Rather than a zero-sum fight, partnerships are emerging. Attwood notes that hyperscalers are already contracting neocloud capacity to meet demand. The regional analyst expects similar models in the Gulf: “One that collaborates with the regional players, to go solo will be a gamble. You need a distribution arm to promote neocloud services.” 

He sketches a division of labour: “The regional players would provide the underlying parts, the underlay and the neocloud would bring in their expertise, the GPUs, the technical people, they’ll build it, operate it and support it.” 

What success looks like

For Attwood, winning operators will bundle the GPU with broader transformation. “AI or GPU as a service isn’t going to be sold in isolation,” he says. “The most successful will link compute, data, AI platforms and professional services into a complete portfolio.” 

Neoclouds face a different test: speed and discipline. “The most successful are the ones that can scale quickly, deploy infrastructure and find customers quickly,” he says, while keeping “careful financial management.” 

Analysys Mason forecasts worldwide GPUaaS revenue rising from USD21 billion in 2024 to USD134 billion by 2030, with telecom operators growing from under USD100 million to USD9.6 billion. The Middle East will be a proving ground for that transition. 

Whether telcos become true neocloud champions or remain enablers may depend less on fibre and more on power, cooling and partnerships. As the regional analyst concludes, “Neoclouds need DC first and foremost.” In the Gulf’s AI race, the data centre, not the network, may hold the crown. 

RELATED STORIES

Capacity Middle East 2026: Your complete event guide

Why the Middle East is attracting major data centre investment