Two related launches from Southeast Asia this August describe the same structural move from different companies: telecom operators positioning themselves not just as the network layer beneath AI workloads, but as the compute provider running those workloads directly. Ooredoo Group, alongside Nvidia and Nokia, is leading a neocloud GPU-as-a-Service platform branded Zankore, targeting 1 gigawatt of AI capacity across Indonesia and Southeast Asia over three years. Indosat Ooredoo Hutchison has separately launched its own branded variant, Zankore by Indosat, also with Ooredoo Group, Nokia, and Nvidia, delivering AI infrastructure specifically through Indosat’s Indonesian network infrastructure.
The Capacity and Capital Commitment Behind the Zankore Platform
The disclosed numbers give a sense of scale rarely available for operator-led AI infrastructure plays: 200 megawatts of the targeted 1 gigawatt is slated for delivery in the first half of 2027, sourced from multiple Indonesian data centres rather than a single facility, with blue-chip customers already secured ahead of that initial capacity coming online. Ooredoo holds a 49% stake in the platform and is committing $800 million over five years, funded through a combination of equity and debt, a capital structure that mirrors the infrastructure-scale financing patterns increasingly used across AI compute buildouts generally, rather than a conventional telecom capex model. The stated technical goal is using Nvidia‘s technology to optimise GPU fleet utilisation, targeting up to 40% more effective compute from the deployed hardware, a meaningful efficiency claim if delivered, since GPU utilisation efficiency is frequently the difference between a commercially viable neocloud platform and one that struggles to compete on price against hyperscaler alternatives.
Why a Telecom Operator Is a Plausible GPU-as-a-Service Provider
The logic for a telecom operator moving into GPU-as-a-Service is more coherent than it might first appear. Operators already own or control large-scale power, cooling, physical security, and network connectivity infrastructure, the same foundational assets a data centre and compute business requires, and in markets like Indonesia and the wider Southeast Asian region, where hyperscaler data centre presence is less dense than in North America, Europe, or Northeast Asia, an operator with existing regional infrastructure and government relationships has a genuine capacity gap to fill rather than competing head-on against an already-dominant incumbent. Pairing that infrastructure with Nvidia‘s compute technology and Nokia‘s networking expertise gives the platform technical credibility it wouldn’t have relying purely on the operator’s own infrastructure experience.
It’s also a hedge against the risk telecom operators generally face as connectivity itself becomes increasingly commoditised: rather than competing purely on network price and coverage, an operator that also controls a regional compute layer has a second, potentially higher-margin revenue stream sitting directly on top of infrastructure it already owns. That’s a materially different growth strategy than the capacity-driven connectivity investment most operators still default to, and it mirrors, at a regional platform level, the same reallocation toward AI infrastructure showing up in operator capital planning globally this year.
What Changes for a Buyer When the Network Vendor Also Sells Compute
For an enterprise or industrial buyer already sourcing connectivity from Ooredoo or Indosat, or evaluating them for a private network or industrial AI deployment, this development changes the shape of the vendor relationship rather than the immediate procurement decision. A single vendor now potentially supplies both the network layer an AI workload runs over and the GPU compute layer the workload actually executes on, which offers a genuine simplification, one commercial relationship, potentially better-integrated network-to-compute performance since both layers are controlled by the same provider, alongside a genuine concentration risk, a single vendor now has visibility and control over a larger share of the buyer’s technical stack than a pure connectivity relationship would create.
That trade-off is directly analogous to the build-versus-buy-versus-bundle decision already playing out at the model layer, where operators building their own AI models, buying models as a managed service, or bundling a third party’s consumer AI product into their offering represent three fundamentally different vendor relationships with different lock-in and control implications. GPU-as-a-Service from a connectivity provider is the same decision one layer down the stack, at compute rather than at the model, and it’s worth evaluating with the same explicit question: does bundling network and compute under one vendor improve the buyer’s actual outcomes in performance, cost, and simplicity, or does it just concentrate dependency without a corresponding benefit.
An Early-Stage Platform Worth Watching, Not Yet Acting On
Zankore’s initial capacity isn’t due until the first half of 2027, and the platform, in both its Ooredoo Group and Indosat-branded forms, is still in build-out rather than full commercial operation. For buyers in Indonesia and the wider Southeast Asian region, particularly those already working with Ooredoo Group operators or Indosat, this is worth tracking closely as a potential compute sourcing option as it matures, with the underlying evaluation question, the network-compute bundling trade-off, worth thinking through now, ahead of the platform reaching full commercial availability, rather than encountering it for the first time inside a live procurement conversation once named blue-chip customers are already locked into early capacity.
Questions Worth Asking Before Treating a Carrier as a Compute Vendor
Buyers considering Zankore or a comparable carrier-delivered GPU-as-a-Service platform once it reaches commercial availability will find it useful to think through a small set of concrete questions alongside the capacity and partnership details already announced: what service-level commitments apply to the compute layer specifically, separate from any existing network SLA, since GPU availability and performance guarantees are a different commitment than connectivity uptime; what happens to committed compute capacity and pricing if the buyer’s underlying connectivity relationship with the same carrier changes or is renegotiated separately; and what exit and portability provisions exist if the buyer later needs to migrate workloads to a different compute provider, given that GPU-as-a-Service platforms, like any compute platform, can create meaningful data-egress and workload-migration friction once a deployment has been running on them for a period of time. Answering these questions clearly before committing is the practical way to capture the genuine simplification benefit of a bundled network-and-compute relationship without absorbing more concentration risk than the buyer actually intends to take on.
Explore the full TeckNexus Intelligence Platform — independent, buyer-neutral tools for private network and industrial AI decisions. https://tecknexus.com/intelligence/
