Zankore AI Compute Platform Targets Southeast Asia With 1 GW of Capacity

Ooredoo Group is leading Zankore, a new AI compute platform in Indonesia developed with Nvidia, Nokia, and Indosat Ooredoo Hutchison (IOH). The platform plans to contract 1 GW of AI capacity over three years, with roughly 200 MW already secured for the first half of 2027. The practical takeaway for AI builders is straightforward: Southeast Asia is getting another large-scale option for renting GPUs instead of sourcing and operating all the hardware independently.

Zankore combines GPU rental with a regional infrastructure layer

Zankore is designed as an AI neocloud platform. Its GPU-as-a-Service model lets enterprises rent compute over the internet for model training, deployment, and inference. Initial capacity will come from multiple Indonesian data centers, rather than one disclosed campus.

The platform will use a multi-tenant orchestration layer to allocate GPUs across users and clusters. Zankore says this can recover stranded power in the GPU fleet and deliver up to 40% more usable compute. That is a vendor-reported efficiency claim, not an independently verified benchmark, and the real benefit will depend on workload scheduling, networking, storage, and utilization.

Ooredoo is committing $800 million over five years and will hold a 49% stake as founder and lead investor. IOH, Ooredoo's Indonesian affiliate, is expected to own and operate about 28 MW of Nvidia-powered AI cloud capacity by the third quarter. IOH reported $33 million in neocloud revenue during the first half of 2026, while citing $1.2 billion to $1.5 billion in contracted future revenue.

Why this matters for AI product teams

The important shift is regional availability. Founders and enterprise teams building agents, retrieval systems, or model-serving workflows may eventually be able to place workloads closer to Southeast Asian users, potentially simplifying latency, connectivity, and regional procurement decisions.

The partnership also separates responsibilities across the stack. Nvidia supplies the compute and AI architecture, Nokia contributes networking expertise, and IOH provides local telecom and infrastructure experience. That is different from simply buying GPUs: the product is meant to package hardware, networking, orchestration, and access into a managed service.

For builders, the decision will still come down to workload economics. A managed GPU service can reduce upfront capital expense and operational burden, but teams will need to compare hourly or usage pricing, queue guarantees, supported frameworks, data residency, retention policies, model access, and observability against hyperscalers and specialist GPU clouds.

The projections are ambitious, not yet operating results

Zankore projects about $13 billion in revenue and $9 billion in cumulative EBITDA over five to six years. Those figures are projections tied to planned capacity and customer contracts, not reported platform performance. The 1 GW target also depends on financing, data-center availability, power, regulation, and deployment execution.

Exact facility locations, pricing, service-level commitments, and contractual terms have not been disclosed in the supplied material. AI teams should therefore treat Zankore as an infrastructure option to monitor rather than an immediately selectable replacement for an existing cloud provider. Its value will be proven when capacity is online, workloads can be scheduled reliably, and customers can verify the cost and performance trade-offs.