Emerging MarketsData CentersAI InfrastructureSoutheast AsiaLiquid Cooling
AI racks bring disputes to Southeast Asia data center contracts
US hyperscalers have spent $1.1 trillion since 2023, while five-year AI refresh cycles and 50kW rack demands create contract disputes across Southeast Asian data centers.
Emerging MarketsWhy it matters
- Four largest US hyperscalers have spent $1.1 trillion since 2023 on data centers, chips and power.
- McKinsey expects AI training and inference to reach roughly half of Asia-Pacific data center demand by 2030, from about 30%.
- AI servers may need significant upgrades within five years, versus seven- to 10-year PE holds and 15- to 20-year debt useful lives.
- Neocloud racks can exceed 50kW, against 8-15kW assumptions in standard powered-shell agreements.
- Hyperscale campuses take two to four years to commission, while Southeast Asian power connection lead times can exceed two years.
The story
The four largest US hyperscalers have spent $1.1 trillion in capital expenditure since the start of the AI boom in 2023. The money covered data centers, advanced chips and the power required to run them.
Southeast Asian projects now sit where that investment meets a faster hardware cycle. Andrew Raymond, legal director in Singapore at Charles Russell Speechlys LLP and a co-writer of the industry view, said: "Technological obsolescence is not a future risk for Southeast Asia's data center sector – it is generating disputes today."
What changes in Asia-Pacific demand?
McKinsey research puts traditional compute, storage and cloud workloads at more than 70% of Asia-Pacific data center demand. AI training and inference contribute approximately 30%. The research expects a roughly equal 50/50 split by 2030.
That mix changes facility specifications, not merely software workloads. Facilities designed for one generation of computing are being asked to support the next across hyperscale and neocloud projects before their construction schedules are complete.
Why do investment horizons collide with refresh cycles?
Servers powering AI workloads may require significant upgrades or replacement within five years. Private equity sponsors typically underwrite data centers over seven- to 10-year holds. Debt financing assumes useful lives of 15 to 20 years.
An investor expecting steady-state returns can face an operator request for hundreds of millions in unbudgeted capital expenditure to keep the platform competitive and current. The industry view says partnership and shareholder agreements often leave refresh obligations, trigger events and cost allocation unstated. Those omissions turn planned upgrades into formal disputes.
What changes when neoclouds arrive?
Neocloud providers, specialised cloud platforms for high-performance computing, emerged in 2023-2024 and gained traction through 2025. Their infrastructure demands rack densities above 50kW, liquid cooling rather than air cooling, and power redundancy for continuous high-intensity computation.
The industry view describes neocloud tenants as both the highest-value customers and the hardest to serve. Standard enterprise powered-shell agreements assume 8-15kW per rack and air-cooled environments. Operators are competing for these relationships across the region.
A tenant requiring 50kW or more exposes every mismatch between an 8-15kW air-cooled contract and a liquid-cooled, high-intensity computing estate serving modern AI accelerator workloads. The commercial question is who pays for the cooling retrofit. The legal question is whether the operator must deliver next-generation capability or only the signed specifications.
How does a construction brief become obsolete?
A hyperscale campus takes two to four years from design freeze to full commissioning. In Southeast Asia, power connection lead times alone can exceed two years. Development timelines therefore stretch beyond the original construction plan.
An EPC contract signed in 2024 may specify 20kW average rack density. A customer agreement signed in 2026 may require 50kW with liquid cooling. The contractor built to the first brief. The operator now serves the second.
The resulting three-way tension pits the developer's original specification against evolving tenant demand and the contractor's position that material changes fall entirely outside the contracted scope. Some owners claim defective design or inadequate design life. EPC contractors answer that they delivered the brief and later technology changes were not deficiencies. Such claims are becoming more common.
Where do the contract disputes emerge?
Four contractual interfaces now attract claims:
- Investors and operators: Capital-expenditure obligations for technology refresh, especially where the investment thesis assumed stable infrastructure and steady-state returns.
- Operators and tenants: Whether colocation or powered-shell agreements require upgrades for next-generation chip architectures, including liquid-cooling retrofits.
- Developers and EPC contractors: Claims that designs should anticipate higher power densities, or that later changes constitute compensable variations, plus design-life and OCP Ready scope disputes.
- EPC and customer contracts: Operators can sit between an EPC brief reflecting the original design and a tenant agreement reflecting newer, incompatible obligations.
How should the contract suite absorb change?
Three changes can put refresh risk before it becomes a dispute:
- Address obsolescence explicitly. Specify power-density escalation from 20kW to 35kW to 50kW per rack, with defined triggers and cost-sharing. Identify the density threshold that requires liquid cooling and who pays for it.
- Build flexibility into EPC contracts. Design-life warranties should cover functional relevance, not only structural longevity. Add defined variations for power density, cooling and connectivity. Place OCP Ready certification or equivalent standards within scope when anticipated.
- Align finance with technology cycles. Partnership and shareholder agreements should treat refresh as foreseeable capital. Reserved matters, capital calls and return calculations should include periodic refresh, while depreciation reflects five-year technology replacement rather than 20-year building lifecycles.
Why use arbitration for technical claims?
The industry view recommends arbitration for disputes involving PUE metrics under varying AI workload profiles. Other issues include liquid-cooling specifications, OCP Ready certification and engineering standards required for hyperscale operation.
SIAC arbitration can appoint arbitrators with engineering or technology expertise, allowing direct examination of rack densities, cooling coefficients, power redundancy configurations and design-life calculations themselves. The authors argue this offers an advantage over national courts, which may not provide judges with data center infrastructure expertise.
That comparison remains advocacy rather than measured evidence. No case outcomes, filing volumes or success rates accompany the recommendation.
Raymond said: "Arbitration's ability to appoint technically qualified arbitrators and engage with complex engineering evidence makes it the forum of choice for these disputes."
The commercial mismatch is already visible in rack densities, construction dates and refresh periods. The response must be contractual, not retrospective. Charles Russell Speechlys also invites operators to review obsolescence provisions and calibrate dispute clauses to these technical claims.
Through 2030, operators should expect AI workloads to approach half of Asia-Pacific demand, making five-year refresh funding and staged density clauses central to new projects.
Also reported
Source: datacenterdynamics.com
More from Daniel Okafor
Show full bio
Senior reporter covering regulation, policy and their impact on Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.
546 articles