Beyond AI-RAN: Why APAC Telcos Are Piling Into AI Infrastructure Now

In early August 2026, RCR Wireless News ran a roundup under the headline “Thursday (telco diary) | APAC telcos pile into AI infrastructure.” It was a short brief, but it packed a significant signal: some of the largest and most influential telecom operators in Asia Pacific and the Middle East are no longer merely experimenting with AI as a network tool. They are becoming AI infrastructure providers in their own right.

Chunghwa Telecom in Taiwan is building AI infrastructure capabilities. SoftBank is making AI a central pillar of its growth strategy, moving beyond mobile into AI data centers and compute platforms. Indosat Ooredoo is working with Nokia and Nvidia on a “full-stack” AI infrastructure platform. And e& in the UAE is pursuing sovereign AI infrastructure with Core42.

For a telecom executive watching the AI-RAN hype cycle from the sidelines, it’s easy to dismiss these as isolated experiments. But they aren’t. They represent a structural shift in how telecom operators are thinking about their role in the AI economy. This article walks through what these announcements mean, and why they matter even if you care primarily about Open RAN.

Q1: Why should I care? This is about data centers, not radio networks.

That’s exactly the point. The telecom industry has spent the last few years debating AI-RAN — putting AI on the radio access network to improve spectral efficiency, energy consumption, and automation. That’s a perfectly valid use case, but it’s also a defensive one. AI-RAN makes your existing network work better. AI infrastructure, by contrast, is an offensive play. It creates new revenue streams from compute, storage, and AI services that ride on top of your network.

When SoftBank talks about AI as a central pillar beyond its “traditional mobile business,” it is signaling that the future growth engine isn’t just connectivity. It’s the ability to host AI workloads for enterprises, governments, and even other telecoms. Chunghwa Telecom building AI infrastructure in Taiwan is similarly about positioning itself as a regional AI hub. Indosat Ooredoo with Nokia and Nvidia is explicitly targeting “regional demand” for a full-stack platform that combines accelerated computing, cloud, and network infrastructure.

If you are an Open RAN operator, the rise of AI infrastructure should matter because it changes the economics of your vendor ecosystem. The same companies building your radio network — Nokia, Ericsson, Nvidia — are also building general-purpose AI infrastructure. The more revenue they get from data centers, the more they can reinvest in R&D for both AI-RAN and traditional RAN. But it also means more competition for your capex dollars.

Q2: What exactly are each of these operators doing?

The RCR roundup doesn’t provide detailed project plans, but the general direction is clear:

• Chunghwa Telecom is building AI infrastructure capabilities in Taiwan. Taiwan is already a critical hub for AI hardware manufacturing – TSMC, after all, makes the chips that power everything. Chunghwa Telecom, as the incumbent carrier, is well positioned to offer AI cloud services, edge compute, and possibly sovereign AI capabilities. The move suggests a deliberate strategy to capture more of the AI value chain beyond just providing the pipes.

• SoftBank is making AI a central pillar of its growth strategy, extending beyond its “traditional mobile business” into AI data centers and compute platforms. SoftBank has long been an AI heavy investor through the Vision Fund. Now it’s turning its own balance sheet toward building the physical infrastructure. We can expect Japanese enterprises and international customers to rent GPU capacity from SoftBank’s data centers.

• Indosat Ooredoo is working with Nokia and Nvidia to develop a “full-stack” AI infrastructure platform. This is particularly interesting because it combines two vendors that often sit on opposite sides of the telecom AI debate. Nokia brings the network and cloud orchestration expertise. Nvidia brings the accelerated compute. Together, they are building something that isn’t just a data center or a RAN, but a platform that can serve regional demand across Southeast Asia – one of the hottest growth markets for AI services.

• e& (formerly Etisalat) is pursuing “sovereign AI infrastructure initiatives with Core42.” Sovereign AI is a term that goes beyond simply building data centers. It means retaining control over data, compute, and AI models within national borders. For a telco like e&, this is a strategic move because many governments in the Middle East want AI infrastructure that is not dependent on hyperscalers or foreign clouds. e& is positioning itself as the trusted national champion.

Q3: How does this differ from the AI-RAN work we keep hearing about?

Good question. AI-RAN is about using AI to improve the radio access network. It includes things like NVIDIA’s AI Aerial platform, Ericsson’s Intelligent Automation Platform, and the AI-RAN Alliance. The goal is to make the RAN more efficient, more intelligent, and eventually more autonomous. It’s a network-side transformation.

AI infrastructure, as described in the APAC telco moves, is about turning the telco into a provider of AI compute and services. The network is still important, but not as the destination for AI workloads. Instead, it becomes the distribution channel. The telco builds a data center filled with GPUs, adds cloud management software, and then sells AI capacity to customers. The network is how customers reach that capacity.

There is some overlap, of course. An AI-RAN deployment can also be a way to monetize idle compute. If you have GPUs in your radio network, you can sell excess capacity. That is part of the “AI-RAN” value proposition. But the APAC telcos are not just doing that. They are building dedicated AI infrastructure with the expectation that this becomes a major revenue line.

Nokia and Nvidia are involved in both. On the AI-RAN side, they have partnered for years on GPU-accelerated RAN. On the AI infrastructure side, the Indosat Ooredoo deal shows them applying the same technology stack to general-purpose AI platforms. This is a classic case of “you can’t have one without the other” – the hardware, software, and orchestration tools are converging.

Q4: What is the “full-stack” approach that Indosat Ooredoo is taking with Nokia and Nvidia?

The term “full-stack” is deliberately vague, but in this context it means combining three layers:

1. Accelerated computing: Nvidia GPUs (and possibly DPUs) for training and inference workloads.
2. Cloud capabilities: The platform needs to offer self-service provisioning, orchestration, and billing – essentially a telecom-grade private cloud.
3. Network infrastructure: Connectivity is part of the platform, not an add-on. The network must deliver low latency and high bandwidth to the compute resources, especially for edge AI.

This is more than just building a data center. It’s about creating a seamless experience where a customer can request a virtual GPU cluster and have it deployed within minutes, with the network automatically configured to provide the right performance.

Nokia brings its experience in cloud-native network functions and orchestration. Nvidia brings the compute and the AI software stack. Indosat Ooredoo brings the customer relationships and the network footprint. If all three work as advertised, it could become a template for other telcos that want to enter the AI infrastructure business without building everything from scratch.

Q5: Are these investments wise?

The skeptics will point to the massive cost and uncertain returns. Building AI data centers is expensive – a single 100 MW facility can cost billions of dollars. The demand for AI compute is currently booming, but it’s also volatile. The recent debate about an “AI bubble” (even as covered on SiliconANGLE last week) suggests that some investors worry about an oversupply of AI capacity in the next few years.

However, there is a strategic calculus here that goes beyond simple ROI. For SoftBank, AI infrastructure is necessary to fuel its AI portfolio companies. For Indosat Ooredoo, it’s a way to capture more value from the explosive growth of AI in Southeast Asia. For e&, sovereign AI is practically a prerequisite for selling to governments. For Chunghwa, it’s about regional competitiveness.

None of these are pure financial investments. They are strategic bets on the future of the telecom industry. Even if the direct ROI takes longer than expected, they position these telcos as essential players in the AI economy. The risk is that they might be overbuilding, but the same was said about fiber networks in the early 2000s – and those eventually became critical infrastructure.

Q6: What does this mean for Open RAN?

Open RAN and AI infrastructure might seem like separate worlds, but they are intertwined in several ways.

First, the vendors are the same. Nokia, Ericsson, Samsung, and Nvidia are all active in both. If telecom AI infrastructure becomes a lucrative business, it will fund the development of Open RAN products. On the flip side, if AI infrastructure investments falter, it could drain R&D budgets away from Open RAN.

Second, the technology convergence is real. An AI infrastructure platform built on Nvidia GPUs and cloud-native software is very similar to the foundation of an AI-native Open RAN solution. In fact, one could argue that the “full-stack” approach for AI infrastructure will eventually merge with the “full-stack” approach for open networks. A telco that owns both can create synergies: the same compute pool can serve RAN workloads during the day and AI training jobs at night.

Third, operators are learning from these AI infrastructure projects how to deploy and operate cloud-native systems. That experience is directly transferable to Open RAN adoption. The more comfortable a telco is with containers, Kubernetes, and orchestration, the easier its transition to Open RAN will be.

Finally, there is a competitive dimension. If the APAC telcos become successful AI infrastructure providers, they will have more leverage in their negotiations with traditional RAN vendors. They won’t be pure connectivity providers. They will be technology companies. That shift in identity could drive them to push harder for open, disaggregated, and programmable networks – because they’ll already have the software skills to manage them.

Q7: What should a skeptical exec do now?

Don’t ignore this trend. Even if you don’t have the capex to build a massive AI data center, you can start preparing your network and your organization for the AI era.

• Study the full-stack template: You don’t need to build a national AI platform to understand how accelerated computing, cloud, and network infrastructure work together. Start with a small pilot – maybe a few GPU servers at the edge – and learn the operational challenges.

• Talk to your vendors: Nokia and Nvidia are not monolithic. Ask them how their AI infrastructure work relates to their Open RAN roadmap. There may be synergies you can exploit.

• Evaluate sovereign AI opportunities: If you are in a country where governments are concerned about AI sovereignty, that’s a market opening. e& and Core42 are showing how a telco can become a trusted partner for state AI ambitions.

• Don’t abandon Open RAN: The two trends are not mutually exclusive. If anything, AI infrastructure investment could be the most powerful argument for Open RAN. By decoupling compute from hardware and using standardized interfaces, you can build a more flexible foundation that serves both RAN and AI workloads.

• Watch the competitive landscape: Your competitors are moving. SoftBank, Chunghwa, Indosat, and e& are all making moves today. In five years, they might be selling AI services to businesses in your territory. You need a response.

The telecom industry has spent years talking about becoming “digital service providers” and “platform companies.” This week’s APAC news indicates that the talk is finally turning into action.

Sources

• Thursday (telco diary) | APAC telcos pile into AI infrastructure - RCR Wireless News