IDC: 50% of Large Retailers to Expand Computer Vision by 2028, Cutting Shrinkage 40% — Which GPU Cloud Powers Vietnam’s Retail Security?

Shrinkage is once again a top-of-mind problem for retailers worldwide, especially as supermarket chains and convenience stores keep expanding self-checkout lanes. According to IDC research cited by BizTech Magazine, 63% of retailers now rank fraud and loss at checkout — including self-checkout — among their top investment priorities. More strikingly, IDC predicts that 50% of large retailers will expand computer vision for in-store monitoring by 2028, cutting shrinkage by as much as 40%.

Meanwhile, global AI compute demand keeps climbing. Research firm TrendForce (via iconnect007) estimates that combined AI inference computing power among the top five North American cloud service providers will grow roughly 122% in 2026, far outpacing training-infrastructure growth. Inference — exactly the type of workload retail computer vision depends on — is becoming the dominant driver of GPU consumption, pushing up GPU costs and supply pressure for businesses trying to self-build this infrastructure.

Why smart retail surveillance needs GPU Cloud, not just storage

In an earlier article, Viettel IDC covered how the global video surveillance market reached US$27 billion, driving up camera-data storage costs. But storage is only half the equation. Turning hundreds of hours of raw security footage into real-time fraud alerts requires an entirely different processing layer: computer vision models (object detection, behavior analysis, motion tracking) that run continuous inference on GPUs, not CPUs or disk capacity.

That is exactly why more retail chains are renting GPU Cloud capacity instead of building physical GPU server farms — avoiding heavy upfront capital spend while still being able to scale elastically during seasonal peaks, such as Lunar New Year or year-end sales.

A note for India-based GCCs and IT outsourcing teams in Vietnam

This use case is also directly relevant to India-headquartered Global Capability Centers (GCCs) and IT outsourcing teams running delivery operations out of Vietnam. Many of these teams build computer-vision and retail-analytics products for global retail clients, and typically need benchmarkable GPU specs, predictable price-performance, and infrastructure that keeps client video and transaction data inside Vietnam for data-residency reasons. A locally hosted GPU Cloud with transparent NVIDIA GPU tiers (A30, L40S, H200) gives offshore development teams a way to prototype and run production inference workloads without routing sensitive retail data through a third country.

Concrete use cases for retail in Vietnam

  • Self-checkout fraud detection: computer vision models flag scan-avoidance (items that bypass the scanner), barcode switching, or items hidden in bags — alerting staff before the transaction closes.
  • Customer behavior analytics and heatmaps: tracking foot traffic and high-density zones to optimize store layout and staffing.
  • Automated shelf-inventory monitoring: cameras paired with AI detect empty shelves and trigger real-time restocking alerts, reducing lost sales from out-of-stock items.
  • After-hours security monitoring: detecting break-ins or intrusion into restricted areas outside operating hours.

Technically, these models share the same foundation as the computer vision Viettel IDC has already deployed for manufacturing — covered in AI defect detection reaching 95% accuracy — and the same real-time inference category as the real-time fraud detection for banks and fintechs Viettel IDC covered previously. The common thread: all of them need low inference latency, stable processing throughput, and the ability to scale elastically with video data volume — advantages a GPU Cloud delivers far more easily than fixed physical servers.

Viettel Cloud GPU: infrastructure ready for retail workloads

Viettel Cloud GPU currently offers NVIDIA A30, L40S, and H200 GPUs, with L40S and H200 as the flagship tiers optimized for both real-time AI inference and graphics/video processing. The infrastructure comes pre-installed with OS, NVIDIA Driver, and CUDA Toolkit, and fully supports popular AI/ML frameworks such as TensorFlow and PyTorch — letting a retail chain’s engineering team deploy computer vision models quickly without building infrastructure from scratch.

The vGPU feature virtualizes GPU cards, letting teams right-size resources for each store or camera cluster, optimizing long-term operating costs compared with buying a full physical GPU card per site. Businesses can also request a proof-of-concept trial of up to 7 days to evaluate performance before committing to a full rollout.

The entire infrastructure runs on data centers certified to ANSI/TIA-942-B Rated 3 — the highest and, currently, the only such certification for Constructed Facilities in Vietnam — alongside ISO 27001:2022 and ISO 27017 information-security certifications, ensuring customer imagery and transaction data are protected at an enterprise level, backed by an SLA of up to 99.99% service availability.

Conclusion

With IDC forecasting that half of the world’s large retailers will expand computer vision by 2028, Vietnamese retail chains don’t need to wait that long to get started. Internationally certified, locally hosted GPU Cloud infrastructure lets businesses deploy fraud-monitoring and operations-optimization models right away — without trading off between upfront investment and speed of deployment.

References:

  • IDC, cited in BizTech Magazine (April 2026) — statistics on retail fraud-prevention investment and computer-vision expansion forecasts.
  • TrendForce, cited in iconnect007 — 2026 AI inference compute growth figures.

For a tailored GPU Cloud quote and consultation on retail loss-prevention and surveillance infrastructure, please contact our Sales team via Hotline/Zalo/WhatsApp: 0966707701 or Email: dungnh1@viettelidc.com.vn