Summary

  • Indosat Ooredoo Hutchison launched Zankore by Indosat on 6 August with Ooredoo Group, Nokia and NVIDIA.
  • The partners set a deployment target of 1GW of NVIDIA DSX AI Factory capacity; they did not report live, funded or sold capacity.
  • Zankore says it intends to deliver approximately 200MW in the first half of 2027 using NVIDIA GB300 NVL72 systems.
  • Ooredoo is lead investor and platform sponsor, Indosat handles local execution, NVIDIA supplies compute and software, and Nokia supplies networking.
  • A claimed MaxLPS benefit of up to 40% more compute within the same power budget is a design objective, not a measured Zankore result.
  • Sites, grid counterparties, permits, construction status, total capital, financing, contracts, prices and utilisation remain undisclosed.

The launch creates an operator, not an operating gigawatt

The hard fact is institutional. A named platform has been launched, Ulf Ewaldsson has been appointed chief executive, and a board will include representatives of Ooredoo and Indosat plus an independent director. Those decisions allocate authority. They do not show that one gigawatt has been connected, financed, built or contracted.

That distinction matters in a power-constrained market. A capacity ambition can coordinate land, electricity, equipment and customers, but the economic asset appears only as each site clears those dependencies. On 6 August, Zankore disclosed no address, utility agreement, permit, construction milestone or commissioning sequence. The appropriate starting value for live capacity is therefore zero, not the headline target.

The 200MW tranche is the first useful clock

The nearer promise is approximately 200MW in the first half of 2027. It is still a target, yet it supplies a date and a denominator against which execution can be tested. Delivering that tranche would represent one fifth of the longer 1GW ambition; missing it would expose the distance between partnership design and physical delivery.

Even 200MW is ambiguous without a stated power boundary. The release calls it AI capacity but does not say whether that means gross facility power, contracted utility load or IT power available to accelerators. Investors and customers should resist converting it into a server count or revenue estimate. No utilisation, tariff or contracted tenant mix is given.

Four partners divide the stack and the risk

Ooredoo is described as lead investor and platform sponsor. Indosat contributes local position, infrastructure and execution. NVIDIA supplies accelerated computing, software, access to next-generation GPUs and an ecosystem; Nokia supplies AI-native networking. The structure assembles complementary capabilities without disclosing who signs each capital obligation.

That omission is economically important. Sponsor capital, construction finance, power commitments, GPU procurement and customer credit are different risks. A project may possess strong technology partners while remaining short of bankable demand or grid access. Until contracts appear, the announcement identifies roles but not the loss-sharing rules if demand, electricity or equipment arrives late.

Power is the binding input hidden behind the compute label

A gigawatt-scale AI platform is first a power-allocation programme. Accelerators can be ordered globally; energised sites require local generation, transmission, substations, permits and cooling. Indonesia’s local execution advantage may help assemble that chain, but no grid counterparty or physical site has been named.

The 2027 timetable therefore transfers attention from chip availability to electrical readiness. If power arrives in stages, compute must follow it. If the project reserves power before customers sign, the sponsors carry utilisation risk. If capacity is pre-sold, customers may bear take-or-pay obligations. None of those contractual choices is public, so the capital burden cannot yet be assigned.

MaxLPS addresses scarcity but does not create megawatts

The partners intend to use NVIDIA DSX and its MaxLPS power-orchestration approach. The release says the design can enable up to 40% more compute inside the same power budget. That is potentially valuable where interconnection capacity is scarce, because the control system can coordinate workload demand rather than provision every rack for a simultaneous peak.

But “up to” is a design claim, not an observed Zankore benchmark. It does not mean the project has 40% more electrical capacity, consumes 40% less energy or produces 40% more revenue. The realised benefit will depend on workload variability, service commitments, thermal limits and how aggressively customers permit jobs to be scheduled. A field measurement with a defined baseline is still required.

Hardware specificity does not establish a customer book

Zankore has named NVIDIA GB300 NVL72 systems for the first tranche. That gives the platform a technical direction and may improve procurement coordination. It does not reveal how many systems have been ordered, their delivery dates, the price paid or the share already reserved by customers.

The customer side is the missing commercial denominator. A platform can install scarce accelerators and still earn poor returns if contracted prices fail to cover power, financing, depreciation and operations. Conversely, pre-committed customers could make staged capacity bankable before construction completes. The launch discloses neither condition, so demand should not be inferred from the prominence of the partners.

Governance is visible; capital discipline is not

Naming a chief executive and board is more consequential than a loose memorandum because it creates a vehicle able to negotiate with utilities, suppliers and buyers. Ooredoo’s lead-investor role also places a party with regional operating experience near the capital decision. Indosat’s execution role gives the plan a local operating surface rather than a purely offshore sponsor structure.

Still, governance only becomes discipline when it controls staged spending. Useful evidence would separate land and grid development from shells, network, cooling and compute, then condition each tranche on permits and contracted demand. With no budget, financing terms or construction status, readers cannot judge whether the 1GW number is a funded plan or a long-dated ceiling.

The alternative is smaller, staged and more reversible

The realistic alternative to a 1GW programme is not doing nothing. It is deploying smaller sites or leased capacity as customers materialise, preserving the option to change hardware generations and locations. That approach sacrifices some purchasing scale but reduces the cost of being early in a fast-changing accelerator market.

Zankore’s announced 200MW tranche may itself be such a stage. Its value will depend on whether the platform can secure power and customers without locking sponsors into obsolete equipment or idle capacity. The fact that another Indonesian operator has separately announced a different programme does not combine the two: Zankore’s ownership, timetable and capacity must be judged on their own record.

Sources