Key takeaways
- Start with data, not servers. Classify data into tiers first, then decide where each workload runs.
- Use site-to-site VPN to get started, and move to dedicated private connectivity once traffic, latency sensitivity or compliance requirements justify it.
- DR across providers removes single-provider risk, but replicating regulated data offshore can trigger cross-border transfer obligations. Design DR around your data tiers.
- Kubernetes, infrastructure as code and federated identity are what make multi-cloud manageable for a small platform team.
- A Vietnam cloud partner should complement your hyperscaler, not replace it.
Vietnam has become one of Southeast Asia’s most dynamic digital markets. With a population of over 100 million, fast-growing internet penetration and strong government support for digital transformation, the country is attracting foreign SaaS companies, AI startups, fintech platforms, gaming studios and regional enterprises.
Entering Vietnam, however, is not simply a matter of launching servers in the nearest hyperscale region. Infrastructure decisions now shape regulatory exposure, latency, user experience and long-term scalability. That is why most foreign companies expanding into Vietnam in 2026 adopt a hybrid cloud model: they keep their global cloud provider and extend it with a local infrastructure layer built for Vietnam-specific requirements.
This guide goes beyond the “why” and covers the four decisions that determine whether a hybrid cloud strategy in Vietnam actually works:
- Data tiering: which data must stay onshore and which can remain on your global platform.
- Connectivity: how to link AWS, Azure or Google Cloud with a Vietnam cloud through VPN, Direct Connect, ExpressRoute or Interconnect.
- Disaster recovery: how to design DR between a hyperscaler and a domestic cloud without creating new compliance risk.
- Multi-cloud operations: how to run two or more clouds with one team, one toolchain and one security model.
Why hybrid cloud is the default model in Vietnam
Foreign companies entering Vietnam usually already run on AWS, Microsoft Azure or Google Cloud. These platforms power identity, analytics, CI/CD pipelines and multi-region deployments. Serving Vietnamese users from a nearby region such as Singapore looks efficient at first, but three structural forces make it less suitable over time.
Latency and user experience
Singapore is geographically close, but cross-border routing still adds latency and variability. For gaming, payments, media streaming, real-time dashboards and AI inference APIs serving local users, that variability directly affects perceived product quality.
Data regulation
Vietnam’s legal framework for data has matured quickly, with the Personal Data Protection Law, the Law on Data and data localization requirements under the cybersecurity framework. Requirements differ by sector and by data type, but hosting sensitive workloads offshore increasingly means more paperwork, more assessments and more uncertainty.
People also read: Cloud Compliance in Vietnam: What Businesses Must Know
AI workload growth
As companies add machine learning, generative AI and large-scale analytics to their products, demand for GPU capacity and high-performance compute rises. Relying only on offshore GPU regions can introduce cost volatility, capacity shortages and data transfer overhead.
The reference architecture: three layers
Every chapter in this guide builds on the same reference model:
- Global layer (hyperscaler): global services, corporate identity, internal tooling, centralized analytics on non-sensitive data, CI/CD and multi-region failover for global products.
- Vietnam layer (domestic cloud): latency-sensitive APIs, AI inference endpoints, user data storage and regulated workloads.
- Regional layer (optional): infrastructure in a neighboring Southeast Asian country, such as Thailand, for redundancy, disaster recovery or regional expansion.
The layers are connected by private or encrypted network links and operated with a shared toolchain. The four chapters below explain how to decide what goes in each layer, how to connect them, how to recover when one fails and how to run them day to day.
Data tiering - deciding what stays in Vietnam
The most common mistake in hybrid cloud planning is to start with compute: “Which services should we move to Vietnam?” The better starting point is data: “Which data are we allowed to process where, and with what controls?” Once data is classified, workload placement usually follows naturally.
A four-tier data model
| Tier | Examples | Recommended location | Key controls |
|---|---|---|---|
| Tier 1: Regulated and sensitive data | Sensitive personal data, financial and identity records, data classified as important or core under Vietnamese law, sector-regulated data | Vietnam layer | Encryption at rest and in transit, keys managed onshore, strict access control, audit logs retained locally |
| Tier 2: Customer operational data | Transactions, user profiles, sessions, user-generated content, AI prompts and inference logs | Vietnam layer (preferred) | Replicate outward only in anonymized or aggregated form; document any cross-border flow |
| Tier 3: Derived and anonymized data | Aggregated metrics, anonymized analytics datasets, business KPIs | Global layer allowed | Verify anonymization is irreversible; keep the transformation pipeline onshore |
| Tier 4: Non-personal global data | Source code, container images, configuration, model weights trained without local personal data, system telemetry without personal identifiers | Global layer | Standard security controls |
The regulatory anchors
Your legal team should own the final classification, but most foreign companies need to account for three areas of Vietnamese law:
- Personal data protection: the Personal Data Protection Law (Law No. 91/2025/QH15) and its implementing Decree No. 356/2025 set obligations for processing personal data, including impact assessments for cross-border transfers.
- Data governance: the Law on Data introduces categories such as core data and important data, with additional conditions for transferring them abroad.
- Data localization: the cybersecurity framework, including Decree No. 53/2022/ND-CP, requires certain domestic and foreign service providers to store specific types of data in Vietnam.
Sector regulators, especially in banking and financial services, may add further requirements. This guide is not legal advice; treat it as a framework for structuring the conversation with your legal and compliance teams.
For a deeper look at sovereignty criteria, explore From data sovereignty to sovereign AI: What it takes for enterprises in Vietnam ebook.
AI workloads need special attention
AI systems create data that traditional classification often misses. Prompts, chat histories, retrieval-augmented generation (RAG) documents, vector embeddings and inference logs can all contain personal data, even when the model itself does not. A practical rule: if an AI feature serves Vietnamese users, keep the inference endpoint, the vector database and the request logs in the Vietnam layer, and send only aggregated usage metrics to the global layer.
Dive deeper in to Sovereign AI Cloud Architecture: Data Control and Responsibility Delineation in AI Operations
How to put data tiering into practice
- Inventory: list every data store and data flow that touches Vietnamese users.
- Classify: assign each data store to a tier and tag resources accordingly in both clouds.
- Map flows: draw every flow that crosses the border and mark its direction, frequency and data tier.
- Default to one-way outbound: let anonymized or aggregated data flow from Vietnam to the global layer, and avoid replicating Tier 1 and Tier 2 data offshore.
- Review regularly: new features, new AI use cases and new regulations change the map, so revisit it at least twice a year.
Connecting your hyperscaler to a Vietnam cloud (VPN, Direct Connect, ExpressRoute and Interconnect)
Once you know which data lives where, the next question is how the layers talk to each other. Your choice of connectivity affects latency, throughput, security posture and, often overlooked, your data transfer bill.
The connectivity options
| Option | How it works | Performance | Setup time | Best for |
|---|---|---|---|---|
| Public internet with TLS | Services call each other over public endpoints with encryption at the application layer | Variable latency, shared bandwidth | Immediate | Low-volume API calls, early testing |
| Site-to-site IPsec VPN | Encrypted tunnel between a hyperscaler VPN gateway and the Vietnam cloud network, over the internet | Better security, still internet-dependent | Hours to days | Starting a hybrid deployment, moderate traffic, backup path for private links |
| Dedicated private connectivity (AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect) | A private circuit from the hyperscaler’s connection point to your network, usually delivered through a carrier or connectivity partner | Consistent latency, high bandwidth, no public internet | Weeks | Production traffic, data replication, latency-sensitive integrations |
| Managed cloud-to-cloud links (for example AWS Interconnect – multicloud, Google Cross-Cloud Interconnect) | Provider-managed private links between clouds | Consistent, managed by providers | Varies by provider and location | Teams already running across multiple hyperscalers |
What “Interconnect” means in this context
The word “interconnect” is used loosely. In chip design it refers to the wiring between components on a processor. In cloud networking it usually means a private connection between networks. Cloud providers also use it as a product name: Google Cloud Interconnect, AWS Interconnect and GreenNode Interconnect are different services. When you compare options, check what each product actually connects (your data center, another cloud, or networks inside the same provider) rather than relying on the name.
How private connectivity typically reaches Vietnam
Hyperscaler private connection points are concentrated in regional hubs. For workloads in Vietnam, a common pattern is to terminate the hyperscaler’s private connection in a nearby hub and extend it into the Vietnam data center through a carrier or network partner. When evaluating this path, ask:
- Where is the nearest connection point for your hyperscaler region?
- Who provides the cross-border and last-mile segments, and what SLA do they offer?
- Are there two physically diverse paths, or a single point of failure?
- What is the measured round-trip latency between your hyperscaler region and the Vietnam data center?
Design principles for hybrid connectivity
- Start with VPN, graduate to private links. Site-to-site VPN lets you launch quickly. Move to dedicated connectivity when traffic volume, latency requirements or compliance reviews call for it, and keep the VPN as a failover path.
- Plan IP addressing early. Overlapping CIDR ranges between your hyperscaler VPCs and your Vietnam network are one of the most expensive mistakes to fix later.
- Encrypt even on private links. Private does not automatically mean encrypted. Use IPsec, MACsec or application-level TLS for Tier 1 and Tier 2 data.
- Design DNS deliberately. Decide how services in each layer resolve each other, and how DNS behaves during failover.
- Model data transfer costs. Hyperscalers typically charge for data leaving their network. Private connectivity often carries lower per-GB transfer rates than internet egress, which can offset its fixed port costs at higher volumes.
Connectivity on GreenNode
GreenNode provides several building blocks for hybrid networking:
- VPN Site-to-Site creates encrypted tunnels between GreenNode and your hyperscaler VPCs or on-premises networks.
- Interconnect moves data through GreenNode’s internal network without using the public internet.
- Cross Connect routes traffic between VPCs.
Disaster recovery between a hyperscaler and a domestic cloud
Hybrid cloud gives you two independent infrastructure providers. Used well, that is a resilience advantage. Used carelessly, it creates a DR design that looks robust on paper but fails compliance review or fails under real load.
Start with RPO and RTO
Two numbers drive every DR decision:
- Recovery Point Objective (RPO): how much data you can afford to lose, measured in time.
- Recovery Time Objective (RTO): how long the service can be unavailable.
Set these per workload, not per company. A payment API and an internal reporting tool should not share the same targets or the same budget.
Three DR patterns for hybrid cloud in Vietnam
| Pattern | Primary site | Recovery site | Best for | Watch out for |
|---|---|---|---|---|
| A. Onshore recovery | Hyperscaler region | Vietnam cloud | Companies that must keep a recoverable copy of data in Vietnam while still running mainly on a hyperscaler | Primary processing still happens offshore, so Tier 1 data may need to move to the Vietnam layer as primary instead |
| B. Offshore recovery | Vietnam cloud | Hyperscaler region | Companies that want to reuse existing hyperscaler capacity for DR | Replicating Tier 1 and Tier 2 data offshore can trigger cross-border transfer obligations; limit replication to permitted tiers or confirm the legal basis first |
| C. Regional recovery | Vietnam cloud | Second country with the same provider (for example Thailand) | Companies that want geographic separation with one toolchain and one support contract | Still a cross-border transfer for regulated data; validate with legal and size the link for replication traffic |
Choosing a DR strategy per workload
- Backup and restore: lowest cost, RTO in hours. Suitable for internal tools and non-critical systems.
- Pilot light: core data replicated continuously, compute started on demand. RTO in tens of minutes to hours.
- Warm standby: a scaled-down copy of the full stack runs continuously and scales up on failover. RTO in minutes.
- Active-active: both sites serve traffic. Near-zero RTO, highest cost and complexity, and the hardest to reconcile with data residency rules.
Cross-provider DR pitfalls
- Image and format lock-in: provider-specific machine images and managed services do not move between clouds. Containers and infrastructure as code make recovery portable.
- Database replication: prefer engine-native replication (for example PostgreSQL or MySQL replicas) over provider-specific features that only work within one cloud.
- DNS and traffic failover: test how quickly clients actually switch to the recovery site; DNS caching often adds more downtime than expected.
- Failback costs: moving data back to the primary site after an incident can generate large data transfer charges. Include failback in your cost model.
- Untested plans: a DR plan that has not been rehearsed is an assumption. Run failover drills at least twice a year.
DR on GreenNode
GreenNode supports DR designs with DR Center for replication and failover, Backup for automated encrypted backups, and Snapshots for incremental disk snapshots and fast rollback. With infrastructure in both Vietnam and Thailand, GreenNode can also host pattern C within a single provider.
Read more: GreenNode & NVIDIA Unveil Thailand’s First Hyper-scale AI GPU Cluster at STT Data Center
Chapter 4: Operating multi-cloud without doubling your team
Hybrid cloud fails most often not because of architecture, but because of operations. Two consoles, two sets of credentials, two monitoring stacks and two billing models quickly overwhelm a small platform team. The goal is to make the second cloud feel like another region, not another company.
Infrastructure as code as the single source of truth
Define infrastructure in both clouds with the same tooling, such as Terraform or OpenTofu, in the same repository and through the same review process. Every change to either cloud goes through a pull request. This keeps environments consistent, makes audits easier and turns DR rebuilds into a repeatable pipeline rather than a manual runbook.
Kubernetes as the portability layer
Running applications on Kubernetes in both layers lets you deploy the same container images with the same manifests. Combine it with GitOps tools such as Argo CD or Flux so each cluster pulls its desired state from Git. On GreenNode, VKS provides fully managed Kubernetes for the Vietnam layer.
One identity model
- Federate access to every cloud from a central identity provider using SAML or OIDC, so joiners and leavers are managed in one place.
- Apply least-privilege roles consistently across clouds, and map them to your data tiers.
- Use GreenNode IAM to control access to GreenNode resources.
Unified observability
Standardize on OpenTelemetry for metrics, logs and traces so data from both clouds lands in one view. Keep logs that contain Tier 1 or Tier 2 data in the Vietnam layer and forward only metrics or redacted logs to global dashboards. vMonitor provides monitoring for GreenNode resources.
Security and compliance as code
- Express guardrails such as encryption requirements, allowed regions and tagging rules as policy as code, and enforce them in CI.
- Manage encryption keys for regulated data in the Vietnam layer.
- Retain audit logs for regulated workloads onshore and define retention periods with your compliance team.
FinOps across clouds
- Enforce a common tagging scheme (team, product, environment, data tier) in both clouds.
- Track cross-border data transfer as its own cost line; it is often the fastest-growing item in a hybrid bill.
- Compare committed-use discounts on your hyperscaler with local pricing for steady workloads such as inference and storage.
Support and accountability
Agree on an escalation matrix that covers both providers and any connectivity partners. Local-language support in the same time zone shortens incident resolution, especially when an issue sits at the boundary between two networks.
A practical hybrid deployment scenario
Consider a SaaS company expanding into Vietnam with AI-powered analytics for local enterprise customers.
- Data tiering: customer records, uploaded documents, prompts and inference logs are classified as Tier 1 and Tier 2 and stay in Vietnam. Aggregated usage metrics are Tier 3 and flow to the global analytics platform on AWS.
- Connectivity: the company starts with site-to-site VPN between its AWS VPC and the Vietnam environment, then adds dedicated private connectivity once replication traffic grows, keeping the VPN as a backup path.
- Vietnam layer: customer-facing APIs, the vector database and AI inference endpoints run on GreenNode in Vietnam, on VKS and GPU instances.
- Disaster recovery: critical services use a warm standby in GreenNode’s Thailand region after legal review of the replication scope, while internal tools use backup and restore.
- Operations: one Terraform repository, GitOps deployments to Kubernetes clusters in both clouds, SSO from the company’s central identity provider and a single observability stack.
- Global layer: authentication, CI/CD, internal systems and global analytics remain on AWS.
The result: compliance alignment, lower latency for Vietnamese users, efficient AI workloads and a path to regional expansion, without dismantling global infrastructure.
Checklist: evaluating a cloud partner for the Vietnam layer
- Onshore data centers in Vietnam with relevant certifications for your industry.
- Hybrid connectivity options: site-to-site VPN, private connectivity paths to your hyperscaler and redundant network routes.
- AI and GPU capacity for training and inference, with high-throughput storage.
- DR capabilities: backup, replication and a second region for geographic separation.
- DevOps compatibility: managed Kubernetes, APIs and infrastructure-as-code support.
- Identity and monitoring that integrate with your central tools.
- Regional presence to support expansion beyond Vietnam.
- Local support in your time zone with clear escalation paths.
For a broader comparison of providers, see Top Cloud Providers in Vietnam.
Why GreenNode fits a hybrid cloud strategy
GreenNode does not position itself as a replacement for hyperscalers. It acts as a performance and compliance layer that complements your global infrastructure.
- Local infrastructure in Vietnam for latency-sensitive applications and onshore data storage.
- Regional presence in Thailand for redundancy, disaster recovery and Southeast Asia expansion without switching providers.
- AI-optimized infrastructure, including GPU instances, scalable compute and high-performance storage.
- Hybrid building blocks across networking, DR, managed Kubernetes, IAM and monitoring.
- Certified operations, including ISO 27001, ISO 27017, ISO 27018, SOC 2 and Tier III data centers.
Frequently asked questions
Do foreign companies have to store all data in Vietnam?
Not necessarily. Requirements depend on the type of data, the industry and the services you provide. Many companies keep regulated and personal data in Vietnam while running non-personal workloads on their global cloud. A data tiering exercise with your legal team is the best way to decide.
Can I connect AWS, Azure or Google Cloud to a cloud provider in Vietnam?
Yes. The quickest option is a site-to-site IPsec VPN. For production traffic, many companies use dedicated private connectivity such as AWS Direct Connect, Azure ExpressRoute or Google Cloud Interconnect, typically extended to Vietnam through a carrier or connectivity partner.
What is the difference between Direct Connect, ExpressRoute and Cloud Interconnect?
They are the dedicated private connectivity services of AWS, Microsoft Azure and Google Cloud respectively. Each provides a private link between your network and the provider’s network without using the public internet, with differences in locations, port speeds, pricing and partner ecosystems.
Is it legal to use a Singapore or other offshore region for disaster recovery?
It can be, but replicating personal or regulated data abroad may trigger cross-border transfer obligations under Vietnamese law. Limit offshore replication to permitted data tiers or confirm the legal basis before implementing it.
Where should AI inference run when serving users in Vietnam?
Ideally in Vietnam. Running inference locally reduces latency, and keeping prompts, retrieved documents and inference logs onshore simplifies compliance, because these often contain personal data.
How do small teams manage multiple clouds?
By standardizing: infrastructure as code for both clouds, Kubernetes with GitOps for deployments, federated identity from one provider and a single observability stack. The aim is to operate the second cloud like another region.
Conclusion
Vietnam’s cloud market in 2026 is shaped by stronger data regulation, rising AI adoption and growing demand for local performance. For foreign companies, relying only on offshore hyperscale regions is often not enough, and hybrid cloud has become the practical answer.
The difference between a hybrid strategy that works and one that stalls lies in four decisions: how you tier your data, how you connect your clouds, how you recover from failure and how you operate day to day. Get those right, and choosing the local partner becomes a matter of matching capabilities to a clear architecture.
If you are planning a hybrid deployment in Vietnam, our team can help you map your data tiers, connectivity and DR design. Book a consultation with our expert now.