Global Cloud Repatriation Wave: 89% of Enterprises Plan to Bring Data Back On-Premises — Why Egress Fees Are the Last Straw

Cloud Repatriation: Why Global Enterprises Are Turning Their Backs on Hyperscalers

For nearly a decade, “cloud-first” was the default playbook for digital transformation: push everything to AWS, Azure, or Google Cloud and figure out the cost later. But heading into 2026, a reverse wave — known as cloud repatriation — is spreading fast enough to make CIOs rethink their infrastructure strategy.

According to the Cloud Data Repatriation Survey published by Cloudian in April 2026 (fielded by Centiment among 212 senior IT decision-makers), 89% of enterprises plan to expand on-premises or private cloud infrastructure over the next two years, and 75% admit they have already quietly moved some workloads back to their own infrastructure in the past 24 months — even though more than three-quarters of them still keep over half their workloads in public cloud today.

Egress Fees: The “Hidden Tax” Eating IT Budgets

The top driver isn’t performance — it’s unpredictable cost. The same survey found that 84% of enterprises are running over their cloud storage budgets, with nearly 1 in 5 exceeding it by more than 30%. The three most-cited causes: data egress fees — the cost of pulling data out of the cloud (46%), costs that scale with data volume (45%), and premium pricing for residency-compliant regions (43%).

Egress fees are especially painful for AI/ML projects, where data constantly moves between clouds, regions, or GPU training clusters. According to an analysis by Akave, at a typical egress rate of around $0.09/GB for the first 10TB per month on major hyperscalers, an AI pipeline that re-reads a 10TB training dataset weekly across cloud boundaries can generate roughly 47TB of egress traffic per month — equivalent to over $4,000/month, or about $51,000/year, just for moving data, before storage costs are even counted.

AI Is the Main Driver of the Repatriation Wave

Cloudian’s survey also found that 85% of enterprises confirm AI infrastructure requirements are pushing them toward on-premises/private cloud: 55% say public cloud cannot consistently meet AI inference latency needs, while 52% must keep AI training data on their own infrastructure for security and compliance reasons. This is exactly the moment when unstructured data — images, video, logs, datasets — is exploding faster than ever, a trend Viettel IDC previously analyzed in its article on unstructured data growing 57% in one year in the Vietnamese market.

The Same Problem Is Quietly Building in Vietnam

Vietnamese businesses are no exception to this trend. As data volumes grow rapidly, many organizations are realizing that storing data on international hyperscalers doesn’t just risk egress fees whenever data needs to be pulled back for local processing — it also raises data sovereignty and compliance concerns. Long-term record-retention requirements such as those under Circular 37/2026/TT-NHNN, or the need to defend against ransomware with immutable object storage, all point to the same conclusion: large-scale, long-term data storage needs a platform that is cost-optimized, based in Vietnam, and easy to integrate with existing systems.

Viettel Cloud Object Storage (vCOS): A Way Out of the Egress Fee Spiral

This is precisely the problem Viettel Cloud Object Storage (vCOS) is built to solve. As an advanced object storage platform compatible with the globally popular S3 standard, vCOS lets Vietnamese businesses migrate or back up data from hyperscalers to domestic infrastructure without rebuilding their entire application architecture.

  • Unlimited capacity, scaling to petabytes — the larger the storage footprint, the more optimized the operating cost per unit of data.
  • No limit on data transfer volume, no limit on upload/download frequency — removing the “every data pull costs money” anxiety that is the top reason global cloud budgets spiral out of control.
  • Scale without service interruption or reconfiguration — built for the pace of AI/ML data growth, training datasets, and model checkpoints.
  • Data stored in Vietnam, on data center infrastructure certified to ANSI/TIA-942-B Rated 3, with ISO 27001, ISO 27017, and ISO 27018 (personal data protection on cloud) certifications — meeting both security and data sovereignty requirements.
  • SLA commitment of up to 99.99% availability, suitable for both frequently accessed operational data and long-term, low-access archival data.

Real-World Applications for Vietnamese Businesses

  • Storing AI training datasets, model checkpoints, and AI/ML system logs without incurring egress fees when training or exchanging data between domestic GPU clusters.
  • Long-term compliance backup and archiving for banking, insurance, and healthcare, in line with regulatory retention requirements.
  • Storing large-volume media, images, and video for retail, manufacturing, and media companies — scaling flexibly during peak seasons.
  • Combining with immutable storage features to defend against ransomware and ensure data recoverability in the event of an incident.

Where Should Vietnamese Businesses Start?

Before the global cloud repatriation wave triggers another round of egress fee hikes, Vietnamese businesses should proactively review three things: (1) calculate the true total cost of ownership (TCO), including potential egress fees when data needs to be pulled back for domestic processing; (2) assess sector-specific data sovereignty and compliance requirements; and (3) prioritize S3-compatible infrastructure so that any future data migration goes smoothly, without having to rewrite applications from scratch.

References:

To get advice on an Object Storage solution tailored to your data scale and compliance needs, please contact our Sales team via Hotline/Zalo/WhatsApp: 0966707701 or Email: dungnh1@viettelidc.com.vn.