How fast is data centre capacity scaling and who is committing capital
Vietnam's operational data centre capacity expanded from 80 megawatts in 2023 to 350 MW in 2024, then to 525 MW by end-2025 — a sevenfold increase in two years [B-Company 2025]. KIRA projects capacity reaching 650 MW by end-2026 and crossing 1,000 MW by 2030 as committed projects complete construction and new licences are awarded. The aggregate AI-specific investment commitment now exceeds USD 7 billion [ERP.today 2025], with named projects including Viettel IDC's 140 MW hyperscale facility in HCMC, ST Telemedia's 60 MW build with VNG, and Saigon Asset Management's USD 1.5 billion, 150 MW campus. These are not speculative announcements: construction financing is confirmed, site acquisition is complete, and grid connection agreements are in progress for the largest facilities.
The demand driver is not simply cloud adoption — it is the intersection of AI model inference workloads with Vietnam's new Personal Data Protection Law, effective January 2026 [Govt 2025]. The law mandates domestic storage of personal data for Vietnamese citizens and imposes penalties of up to 5% of annual revenue for violations. For every international technology company with a Vietnamese user base — payments, ride-hailing, e-commerce, social — the law creates a non-discretionary capacity requirement that cannot be served from Singapore or Hong Kong data centres. This regulatory floor is structurally different from demand driven by cost arbitrage, which can reverse; compliance demand is sticky and difficult to offshore.
What does power density escalation mean for industrial park operators
Traditional co-location racks in Vietnam's early-generation data centres ran at 20–30 kilowatts per rack [B-Company 2025]. AI inference and training workloads are demanding 100–200 kW per rack — a four-to-tenfold increase in power density over the same physical footprint. This shift invalidates the infrastructure assumptions of almost every existing Vietnamese industrial park. Standard parks were designed around manufacturing tenants drawing 0.3–1.5 MW per hectare; a 10 MW data centre on two hectares of park land requires the equivalent power infrastructure of a 30-hectare manufacturing zone. Developers without existing high-voltage substation access or government relationships to fast-track grid connections cannot serve this tenant class at all, regardless of headline land availability.
The operational implications extend beyond power. Cooling infrastructure for AI-density racks typically requires liquid cooling systems — immersion or direct-to-chip — that most Vietnamese industrial buildings were not engineered to support. Water supply agreements, structural load ratings for densely racked server equipment, and fibre dark-fibre diversity into exchange points all require retrofitting or greenfield specification from the outset. Developers who begin specifying these requirements into master plans now — Becamex and Viglacera have both initiated pilot programmes [BCM AR 2025][VGC AR 2025] — will have a 24-to-36-month infrastructure lead-time advantage over developers who wait for confirmed tenant demand before speccing.
How do AI use cases beyond data centres affect the broader industrial ecosystem
Hyperscale data centres capture headlines, but AI's impact on Vietnam's industrial real estate market extends across six distinct use cases. Predictive maintenance for park infrastructure — deploying machine-learning models on sensor streams from power distribution, water treatment, and road networks — is demonstrating 30–50% reductions in unplanned downtime on pilots at parks above 200 hectares, with investment payback periods of 12–18 months [Kira estimates]. Smart-factory tenant services, where developers offer AI-enabled production analytics as a bundled amenity, are showing measurably longer lease terms and higher renewal rates; Viet Thang Jean and Orion Vina Food have published pilot outcomes [Govt SGGP 2025]. AI-assisted lead scoring deployed by tier-1 brokers — JLL, Cushman & Wakefield, Savills — is generating 15–25% conversion-rate improvements on large FDI mandates, though developer-side deployment lags by an estimated 12–24 months [Kira estimates].
Vietnam's enterprise AI adoption rate of 18% sits below Thailand (25%) and well below Singapore (40%) [RMIT 2025], but the cloud market underpinning AI deployment is growing from USD 1.24 billion in 2025 toward USD 2.5 billion by 2029. The AI software market is tracking at USD 932 million equivalent with a 26% compound annual growth rate through 2031. Industrial real estate operators that treat AI purely as a tenant-attraction tool — rather than an internal operational capability — will find themselves managing cost-inefficient parks against peers who have embedded predictive analytics into their asset management workflows within the next three to five years.
Methodology note: Capacity figures in megawatts represent total IT load (power delivered to servers), not gross building power draw, following the B-Company Vietnam Data Centre Outlook 2025 methodology [B-Company 2025]. Forecast data points for 2026F, 2028F, and 2030F are KIRA analyst interpolations from the B-Company 2030 target and named committed-project pipeline disclosed in ERP Today [ERP.today 2025]; they do not account for potential capacity delays from grid connection or permitting bottlenecks. The USD 7 billion aggregate commitment figure sums publicly announced investment values and may double-count joint-venture structures where both parties have separately disclosed the same facility.