When to move from VPS to a dedicated server: the four signals

You have been running your application on a VPS for months or years. It worked fine during the early growth phase. Now you are seeing performance symptoms that no amount of tuning fixes. The CPU graph stays pinned at 90% during business hours. Database queries that used to return in milliseconds now take seconds because the disk is fighting for attention. Or your licensing costs for a per-core product have crossed the point where a physical machine would be cheaper.

These are not random problems. They are signals, four specific signals, that tell you exactly when a VPS is no longer the right tool and a dedicated server is the next step. This article walks through each one with real-world scenarios so you can decide based on data, not a gut feeling.

Signal 1: Sustained CPU ceiling

Every VPS runs on a hypervisor that divides the physical CPU cores among multiple tenants. If you rent a VNLite plan on a standard Vietnam VPS infrastructure (1 vCPU, 2 GB RAM, 20 GB NVMe), that single vCPU maps to one logical thread on the host. You get the full thread when the host is quiet. When neighbours are busy, you share it.

Your monitoring shows CPU averaging 85-95% for hours at a stretch. You have already done the obvious, reviewed your code for inefficiencies, added basic caching with Redis, considered switching to a 4vCPU or 8vCPU plan. But the problem is not the number of vCPUs; it is that the workload is consistently compute-bound. Disk reads and network I/O are not the bottleneck. The CPU itself is saturated.

A dedicated server gives you every physical core to yourself. No noisy neighbour. No CPU steal time (the metric Linux reports as steal in /proc/stat and top). If you are regularly seeing steal above 5%, that is strong evidence that the host is overcommitted and a dedicated machine would eliminate the contention entirely.

When to act: CPU steal exceeds 5% consistently, or load average stays above 80% of the vCPU count for more than four hours a day, even after you have optimised the software stack.

Signal 2: I/O contention and unpredictable storage latency

Shared storage is the second reason you outgrow a VPS. Even NVMe-backed VPS, which is far faster than spinning disks, still runs on a storage pool shared across all tenants on the host. Your iostat output shows high await times (above 10 ms for NVMe is a red flag) and elevated svctm. More importantly, the latency varies wildly. A query that took 2 ms at 3 AM might take 40 ms at peak hours.

Database workloads and high-transaction applications are the first to feel this. Web servers with well-cached pages barely notice because most content comes from RAM. But if your application does frequent writes, session storage, queued jobs, logging, e-commerce order processing, the storage latency directly affects response time.

A dedicated server uses drives that belong to you and only you. The same NVMe model that gave you 50,000 random-read IOPS on a busy host might deliver 80,000 to 100,000 IOPS consistently when it is not shared. The latency also stabilises. Your monitoring graphs go from jagged noise to a flat line.

How to verify: Run fio --randrw=8k --direct=1 --ioengine=libaio --runtime=60 --numjobs=4 on your VPS at peak and off-peak hours. If the difference in throughput or latency exceeds 30%, your storage is oversubscribed.

Signal 3: Licensing economics, the crossover point

This signal is purely financial and often overlooked. Many enterprise software products, databases like Microsoft SQL Server, monitoring tools like Datadog, even some SaaS platforms, license per core or per vCPU. A VPS with 8 vCPUs on a 16-core host uses 8 of those 16 cores. You pay licensing for 8 cores. A dedicated server with 8 physical cores also costs licensing for 8 cores. But the monthly rental of a dedicated server is usually 2 to 4 times the rental of an equivalent VPS.

The crossover happens when your software licensing dwarfs the server rental. If your VPS costs $50/month and your per-core license costs $40/core/month, an 8-core VPS means $320/month in licensing plus $50 in hosting, total $370/month. A dedicated 8-core server might rent for $150/month, but the licensing is the same $320/month: total $470/month. No savings yet.

But once you scale to 16 cores, which you need because your workload demands it, the VPS rental jumps (say $150/month) and licensing jumps to $640/month. A dedicated 16-core server might rent for $250/month with the same $640/month in licensing. That is $790/month for the dedicated versus $790/month for the VPS, the crossover. Beyond 16 cores, the dedicated server becomes cheaper, because VPS rental scales linearly with vCPU count while dedicated server rental plateaus.

When to act: Run the numbers. If the difference between VPS hosting cost and dedicated hosting cost is less than 20% of your total licensing bill, switching to a dedicated server reduces or eliminates the hosting premium without changing the licensing bill.

Signal 4: Compliance isolation and audit boundaries

This is the least technical signal and the hardest to quantify, but it forces the move faster than any other reason.

If your company processes data that falls under regulatory frameworks, Vietnam's Decree 53 (data localisation), GDPR, PCI DSS, HIPAA, SOC 2, a shared hypervisor is a compliance risk. The data of multiple customers co-locates on the same physical hardware, even if logical isolation exists through the hypervisor. Your auditor may ask: "How do you guarantee that another tenant on the same host cannot access your data?"

The answer for a VPS is "hypervisor isolation," which passes most audits. But some frameworks, especially those governing financial data or state-related information in Vietnam, require physical separation. Decree 53, for instance, mandates that certain types of user data reside within Vietnam's borders on infrastructure that the service provider physically controls. The provider can still own the hardware (as in colocation or dedicated rental), but the hardware must not be shared at the hypervisor level with unrelated tenants.

When to act: When your legal or compliance team tells you that a shared environment is not acceptable for a specific data set. This is a binary signal, not a gradual one.

How dedicated server economics really work

A dedicated server on Vietnam infrastructure, such as the BizMaC network behind thueVPS, gives you an Intel E5-2670 or newer Xeon, 32 GB to 128 GB of RAM, and NVMe or SSD RAID storage, all on a 1 Gbps port. The domestic bandwidth is capped at 100 Mbps (the standard commit rate on that port), and international bandwidth is a shared pool of roughly 4 to 10 Mbps, sufficient for management traffic, updates, and API calls, but not for serving large files to global users. The machine comes with one dedicated IPv4 from the Vietnam range, free rDNS, and (for dedicated servers) IPKVM for out-of-band management.

Colocation packages typically offer a 100 Mbps commit on a 1 Gbps port with 400W of power per 1U. You bring the hardware; the datacenter provides the rack space, power, cooling, and the Viettel IDC or VNPT IDC, both Tier 3 facilities inside Vietnam, connectivity.

Compare this to a VPS on the same network: VNLite (1 vCPU / 2 GB / 20 GB NVMe) through VNx8 (8 vCPU / 16 GB / 100 GB NVMe), all KVM-virtualised with full root or Administrator access, monthly billing, snapshots, and backups. No data cap on bandwidth, same 100 Mbps domestic speed, same international pool. The VPS is local infrastructure built for workloads that touch Vietnam, serving Vietnamese users, integrating with Vietnamese payment gateways, running a Vietnamese-language website, or getting a Vietnam IPv4 for geo-targeting.

The jump to a dedicated server matters when the VPS runs out of headroom on CPU or I/O, when licensing arithmetic flips, or when compliance demands physical separation. All other workloads, and that is most of them, should stay on a VPS.

Making the decision

Run this checklist before you buy hardware:

• Is CPU steal time over 5% on a sustained basis?
• Does storage IOPS vary more than 30% between peak and off-peak?
• Does switching from VPS to dedicated reduce or flatten your total cost (hosting + licensing)?
• Does an audit or regulation require physical isolation of this workload?

If you answered yes to at least two of the first three, or yes to the fourth, it is time. If all four are no, keep your VPS, the dedicated server adds cost without measurable benefit.

FAQ

Related articles

• Sizing RAM for Windows Server VPS: why 2 GB is never enough
• Windows Server vs Linux for a small business VPS
• Self-hosted GitLab CE minimum viable server spec
• 10 cheap VPS Vietnam must-read before buy 2026 guide