Dedicated Servers and Bare Metal
Dedicated Server Hosting and Bare Metal Pricing in 2026, Managed and GPU Servers, and How to Get a Better Rate
Dedicated and bare metal servers run from about $39 a month to tens of thousands for GPU servers. Send us your configuration for a Free Server Price Check, and you’ll hear back within 24 hours.
Get My Free Server Price CheckBare metal servers, also sold as dedicated server hosting, are dedicated physical machines rented from a provider: one customer per server, no hypervisor, no shared hardware. You get every core, all the memory and the full network port, and the provider owns the hardware, runs the facility and replaces parts when they fail. That puts bare metal between public cloud, where you rent a virtual slice of a machine, and colocation, where you own the hardware and rent the space. Bare metal server pricing in 2026 starts at about $0.08 an hour or $39 a month for an entry machine on phoenixNAP's Bare Metal Cloud, runs roughly $97 to $750 a month for standard dedicated servers on Hivelocity's price list, and reaches tens of thousands of dollars a month for eight-GPU servers. The bigger differences between providers are often not the server price but bandwidth, egress, term discounts and support, and that is where a comparison pays for itself.
Prices checked September 2026. Sources: AWS data transfer pricing (MCA calculation at $0.09 per GB to 10 TB and $0.085 beyond); Rackdog provider comparison, July 2026 for Hivelocity port pricing and Latitude.sh's included transfer. Allowances and port prices vary by provider and plan. Free to cite with a link to this page.
CPU or GPU, cores, memory, storage, bandwidth, location, how many servers and for how long, and what you run on today. Rough numbers are fine.
A first read on which kinds of providers fit your configuration, what it typically runs, and whether managed, unmanaged or colocation is the better fit.
If cloud or what you run today is the better answer, we will tell you so, and it will have cost you one email.
The provider you choose pays us from its channel budget, and your rate is not marked up.
Or email what you need to protect to contact@metrocoloadvisory.com
Dedicated Server and Bare Metal Pricing in 2026
What a bare metal server costs depends on four things: the hardware, how you are billed, how much traffic it sends, and the term. These are the current reference points.
| Tier | Typical price | What it is | What moves the price |
|---|---|---|---|
| Entry bare metal cloud | From about $0.08 an hour, or $39 a month | A single-processor server provisioned by API in minutes, billed hourly | Hourly versus monthly or reserved billing; reservations are discounted |
| Standard dedicated servers | About $97 to $750 a month | Single and dual-processor servers for web, application and database tiers | Cores, memory, NVMe storage, included transfer, location |
| High-memory and storage servers | Hundreds to low thousands a month | Up to 1 TB of RAM or large drive arrays for databases, analytics and storage | Memory and drive configuration; custom builds are quoted |
| Single-GPU dedicated servers | From about $250 a month for older or smaller GPUs | One GPU in a dedicated server for inference, rendering or development | GPU model and memory; data center GPUs cost far more |
| Eight-GPU servers | About $19,000 (H100) to $38,000 (B200) a month at medians | HGX servers for training and large-model inference | GPU generation, term, and whether capacity is in stock. See GPU rental |
| Bandwidth | Included allowance, metered overage, or an unmetered port (about $199 a month for 1 Gbps, $2,999 for 10 Gbps on one list) | The line that most often decides the cheapest provider | Allowance size, overage rate, port speed |
Sources: phoenixNAP Bare Metal Cloud; GPU dedicated server comparison, September 2026 for Hivelocity's list range and single-GPU entry pricing; Rackdog, July 2026 for port pricing; GPU server figures are MCA calculations from the medians on our H100, H200 and B200 pages. List prices are starting points; volume and term pricing is negotiated.
Bare Metal vs Cloud, VM, VPS and Colocation
A virtual machine is a slice of a server shared through a hypervisor. Bare metal is the whole server. Colocation is your own server in someone else's building. Each step down the list costs less per month for a steady workload and asks more commitment in return.
| Cloud VM | Bare metal | Colocation | |
|---|---|---|---|
| What you rent | A virtual slice of a shared machine | The whole physical machine | Space, power and cooling for hardware you own |
| Who owns the hardware | The provider | The provider | You |
| Performance | Hypervisor overhead and variation from other tenants | Full, consistent performance; no noisy neighbors | Full, on your own hardware |
| Billing | Per second or hour | Hourly, monthly or reserved | Monthly per cabinet or kW, plus the hardware you buy |
| Time to run | Seconds | Minutes for stock configurations, days for custom builds | Weeks to months, including hardware delivery |
| Capital | None | None | The hardware, amortized over three to five years |
| Scales down | Yes, instantly | At the end of the billing period or term | No |
| Hardware failure | Rescheduled automatically | Provider replaces the part, usually within hours | Your spares and remote hands |
| Bandwidth | Egress metered per GB | Often generous allowances or unmetered ports | Carrier circuits you contract |
| Best for | Spiky, short-lived or managed-service workloads | Steady workloads without buying hardware | Steady workloads at scale over three years or more |
Bare metal beats cloud when
The workload runs steadily around the clock (databases, application tiers, analytics, batch), it sends a lot of traffic out, it is sensitive to latency or noisy neighbors, and you can predict it for a year. The cloud vs colo calculator models your numbers.
Cloud beats bare metal when
Load spikes many times above its baseline, the stack depends on managed services, you need presence in many regions at once, or the footprint is small enough that operating servers costs more than it saves.
Colocation beats bare metal when
The fleet is large and steady enough to justify owning the hardware for three to five years. At that point renting the machine costs more than buying it and renting the space. See cloud repatriation and wholesale colocation.
Dedicated server vs VPS. A VPS, or virtual private server, is the budget version of a cloud VM: a fixed slice of a shared machine at a low monthly price. It suits small sites and test environments. A dedicated server gives you the whole machine, so performance does not depend on what other customers are running, and it costs less per unit of compute once a workload is steady and large.
Send us what runs there today. We tell you which workloads would cost less on dedicated servers, find the providers that fit and negotiate the terms. See also FinOps consulting and the cloud vs colo calculator.
Managed vs Unmanaged Dedicated Servers
The biggest choice after the hardware is who runs it. An unmanaged dedicated server is the machine, the network and the power; everything above the hardware is yours. A managed dedicated server adds a team that runs the operating system for you, which is why managed buyers are usually businesses without a full-time systems administrator.
Unmanaged
The provider keeps the hardware, network and power running and replaces failed parts. You install and patch the operating system, secure it, back it up and fix what breaks. Lowest monthly price; needs in-house skills.
Managed
The provider also installs and patches the operating system, monitors the server, hardens security, runs backups and responds when something fails, usually for a monthly management fee per server. Higher price; far less work for your team.
What to check
Exactly what “managed” covers, response times in writing, whether backups and restores are included, and who handles security patches. The word means different things at different providers.
Tell us what the servers run and what you need handled. We find managed dedicated server providers whose service covers it, and get the response times and backups written into the contract.
Dedicated GPU Servers and GPU Server Rental
A GPU dedicated server is a bare metal server with one or more GPUs, rented the same way: the whole machine, by the month or the term. It is the most common way to run inference and fine-tuning without buying hardware, and the fastest way to get a single eight-GPU node without a cluster contract.
| Server | Typical monthly cost | Good for |
|---|---|---|
| Single consumer or workstation GPU | From about $250 | Development, rendering, small-model inference |
| Eight H100s | About $19,000 at the $3.25 median, less on a reserved term | Fine-tuning and inference on models that fit in 80 GB per GPU |
| Eight H200s | About $26,000 at the $4.48 median | Large-model inference; 1,128 GB of pooled GPU memory |
| Eight B200s | About $38,000 at the $6.49 median | Training and FP4 inference at the highest throughput per server |
MCA calculations from September 2026 on-demand medians, running around the clock. More than one server? See GPU cluster rental; for the specialist GPU clouds, see neocloud providers.
Dedicated Server Configurations: AMD EPYC, 10 Gbps, Unmetered Bandwidth and Storage Servers
Four choices decide most dedicated server bills after the processor count.
| Choice | What it means | When it matters |
|---|---|---|
| AMD EPYC processors | AMD’s server processors; the current EPYC 9005 generation reaches 192 cores in a single processor | Dense compute, virtualization hosts and databases, where more cores per server means fewer servers |
| 10 Gbps ports | A 10 gigabit network connection to the server instead of the common 1 Gbps | Streaming, large file transfer, backups and busy APIs that would saturate a smaller port |
| Unmetered bandwidth | A flat price for the port with no per-terabyte charges, usually under a fair-use policy | Traffic-heavy workloads; metered plans include a set number of terabytes and bill the overage |
| Storage servers | Servers built around many drive bays, for hundreds of terabytes of hard disk or NVMe | Backup targets, archives, media libraries and data lakes |
Bandwidth is where dedicated servers most often beat cloud: a major cloud bills outbound data per gigabyte, while a dedicated server typically includes a port and a traffic allowance, or unmetered traffic, in the monthly price. For backup storage specifically, see backup as a service.
Bare Metal as a Service
Bare metal as a service, also called bare metal cloud, is bare metal with a cloud-style front end: servers provisioned by API or Terraform in minutes, billed by the hour, the month or on reservation, and released when you are done. It keeps the performance and single tenancy of a dedicated server with most of the convenience of cloud. Custom hardware builds are the exception; those are quoted and take days. For teams used to infrastructure as code, it is usually the easiest way into bare metal.
Bare metal Kubernetes
Running Kubernetes directly on bare metal removes the virtualization layer under your nodes, which gives containers the full performance of the hardware and makes GPU and high-throughput networking workloads simpler to run. The trade is that you, or a managed service on top, handle the control plane, load balancing and storage that a cloud's managed Kubernetes provides. Several bare metal providers offer Kubernetes-ready images or managed options; the choice usually comes down to how much of that layer your team wants to run.
Equinix Metal Alternatives
Equinix retired Metal, its bare metal as a service platform, on June 30, 2026; servers left running after that date are deleted, and console access for billing and history ends December 31, 2026. Former Metal customers usually land in one of three places: another bare metal cloud with API provisioning in the same metros, dedicated servers on a monthly or annual term, or their own hardware in colocation for steady workloads over three years. GPU workloads are usually better served by the specialist GPU clouds. The Equinix alternatives guide covers the full picture, and we place all three.
Bare metal for latency-sensitive workloads in the Northeast
For trading analytics, market data and other latency-sensitive work in the US Northeast, what matters is where the servers sit relative to the exchanges and your counterparties, not only the provider's footprint map. The strongest options are bare metal in or next to the Secaucus and Manhattan interconnection buildings, with a cross-connect or short metro circuit to the venues you need, and a provider that can provision there in days rather than weeks. Many buyers pair a small colocation footprint on the trading campus with bare metal nearby for the analytics tier. See Equinix NY4 and financial services colocation.
Were you on Equinix Metal? Send us what you ran there: server types, counts, locations and how you provisioned them. We find the bare metal providers that match it and negotiate the terms. Get your Free Server Price Check and you’ll hear back within 24 hours.
Dedicated Server and Bare Metal Providers
Bare metal providers differ more by footprint, provisioning model, bandwidth policy and support than by hardware, which is broadly similar. These are the kinds of provider a search compares.
| Type | Examples | Best for |
|---|---|---|
| Bare metal cloud, API-first | phoenixNAP Bare Metal Cloud, Latitude.sh (part of Megaport) | Teams that provision by API or Terraform and want hourly or reserved billing |
| Dedicated server specialists | Hivelocity, and others with custom builds and many locations | Custom hardware, predictable monthly pricing, direct support |
| Large global hosts | OVHcloud and similar | Wide location choice and standardized configurations |
| Hyperscaler bare metal | Bare metal instance types from the major clouds | Workloads that need bare metal inside an existing cloud environment |
| GPU clouds | The specialist neoclouds | GPU servers and clusters. See CoreWeave competitors |
Metro Colo Advisory summary, September 2026. We hold channel partner relationships with the bare metal and GPU providers we place, and the recommendation follows the requirement.
Dedicated Servers by Location: New York, Chicago, Dallas and Los Angeles
Where a dedicated server sits decides how fast it reaches your users, your exchanges and your clouds. Most providers offer the same hardware in several cities, so the choice is about network position rather than price.
| City | Where the networks meet | Typical fit |
|---|---|---|
| New York | 60 Hudson Street and 111 Eighth Avenue in Manhattan, with trading venues across the river in New Jersey | Financial services, media and East Coast users |
| Chicago | 350 East Cermak Road, one of the largest interconnection buildings in the US | Trading, Midwest users and a central US location |
| Dallas | The Infomart, the main interconnection hub of the central US | National reach from the middle of the country, at lower cost than the coasts |
| Los Angeles | One Wilshire, the landing point for many trans-Pacific routes | West Coast users and traffic to Asia |
How to Choose the Best Dedicated Server Hosting and Rent a Server
Size it on what you actually use
Measured CPU, memory, storage and traffic from your current environment, with headroom for a year's growth. Oversizing a server you pay for monthly is the easiest money to waste.
Price the bandwidth, not just the server
Included allowance, overage rate, or an unmetered port. For traffic-heavy workloads this line decides the cheapest provider.
Choose the billing model
Hourly for short or uncertain work, monthly for steady servers, reserved or one to two year terms for the lowest rate.
Check location and network
Where the server sits relative to your users and your other infrastructure, and whether private links to your cloud or colocation are available. See hybrid cloud colocation.
Read the support and replacement terms
Hardware replacement time, what support includes, and what costs extra.
Compare several providers on the same configuration
Rate cards are starting points. Volume, term and bandwidth are where a negotiated rate moves most.
How We Place Bare Metal
You tell us the configuration or the workload: CPU or GPU, cores, memory, storage, traffic, location, how many servers and for how long. We take it to the bare metal and GPU providers that fit, compare the full monthly bill including bandwidth and support, and negotiate the rate and term before you sign. If bare metal is the wrong answer, because the workload is too spiky or the fleet is big enough that owning hardware in colocation wins, we say so and price that instead. The provider you choose pays us from its channel budget, so it costs you nothing and your rate is not marked up. How a data center broker works.
Frequently Asked Questions
What is a bare metal server?
A dedicated physical server rented from a provider, used by one customer with no hypervisor and no other tenants. You get the whole machine's cores, memory and network; the provider owns the hardware, runs the facility and replaces failed parts.
How much does a bare metal server cost?
In 2026, entry bare metal cloud instances start around $0.08 an hour or $39 a month, standard dedicated servers run roughly $97 to $750 a month on one major provider's list, high-memory and storage servers run hundreds to low thousands, and eight-GPU servers run about $19,000 to $38,000 a month at current medians. Bandwidth, term and support can change the total as much as the server price.
What is the difference between bare metal and a virtual machine?
A virtual machine is a slice of a shared server, created by a hypervisor that splits the hardware among tenants. Bare metal is the whole server for one customer. Bare metal gives full, consistent performance with no noisy neighbors; a VM gives faster provisioning and resizing in seconds.
Is bare metal cheaper than cloud?
For steady workloads that run around the clock, usually yes, especially when they send a lot of traffic out: cloud internet egress runs about $0.09 per GB, while many bare metal plans include large allowances or unmetered ports. For spiky or short-lived workloads, cloud is usually cheaper because you stop paying when you scale down.
What is bare metal as a service?
Bare metal with a cloud-style interface: dedicated servers provisioned by API or Terraform in minutes, billed hourly, monthly or on reservation, and released when you are done. It is also called bare metal cloud.
How much does a dedicated GPU server cost?
From about $250 a month for a single smaller GPU, to about $19,000 a month for eight H100s, $26,000 for eight H200s and $38,000 for eight B200s at September 2026 on-demand medians. Reserved terms and negotiation bring those down.
What are the best Equinix Metal alternatives?
Another bare metal cloud with API provisioning, dedicated servers on monthly or annual terms, or your own hardware in colocation for steady workloads over three years. GPU workloads are usually better served by specialist GPU clouds. Equinix retired Metal on June 30, 2026.
Should I run Kubernetes on bare metal?
If you want full hardware performance for containers, particularly GPU or high-throughput networking workloads, and your team is comfortable running the control plane, load balancing and storage, or buys a managed layer for them. If you want a fully managed service, a cloud's managed Kubernetes is simpler.
How fast can I get a bare metal server?
Stock configurations on bare metal clouds provision in minutes. Custom builds and large orders take days. Eight-GPU servers depend on what is in stock, which is why availability should be confirmed before you plan around a date.
When is colocation better than bare metal?
When the fleet is large and steady enough to justify owning hardware for three to five years. Renting a machine includes the provider's hardware cost and margin; at scale, buying the servers and renting the space costs less. See cloud repatriation and wholesale colocation.
Can you get me a better rate than the price list?
Often. Volume, term and bandwidth are where a negotiated rate moves most. Single entry servers have less room; fleets, GPU servers and multi-year terms have the most.
What does it cost to place bare metal through Metro Colo Advisory?
Nothing. The provider you choose pays us from its channel budget, the same way it pays its own sales team, and your rate is not marked up.
Is a dedicated server the same as a bare metal server?
Yes. Both mean a physical server rented to one customer. “Dedicated server” is the older hosting term; “bare metal” is more common when the server is provisioned on demand, by API or by the hour.
What is a managed dedicated server?
A dedicated server where the provider also runs the operating system: installation, patching, monitoring, security hardening, backups and response when something fails, usually for a monthly management fee per server. It suits businesses without a full-time systems administrator.
What is the difference between a dedicated server and a VPS?
A VPS is a fixed slice of a shared machine at a low monthly price; a dedicated server is the whole machine. Dedicated servers cost more but perform consistently and cost less per unit of compute once a workload is large and steady.
What is an unmetered dedicated server?
A server whose port is billed at a flat price with no per-terabyte charges, usually under a fair-use policy. It suits traffic-heavy workloads; metered plans include a set amount of traffic and bill the overage.
How do I choose the best dedicated server hosting?
Start with the workload, not the price list: the processor, memory, storage and bandwidth it needs, the city it should run in, and whether you want the server managed. Then compare the full monthly bill, including bandwidth, support and term, across providers that fit.
What is a Free Server Price Check?
Send us your configuration or describe the workload. Within 24 hours you hear back with a first read on which kinds of providers fit, what that configuration typically runs, and whether managed, unmanaged or colocation is the better fit. It costs nothing.
Get Your Free Server Price Check
Send the configuration or just describe the workload: what it runs, roughly how big, where, and what you are on today. You'll hear back within 24 hours. If bare metal is not the right answer, we will say so. No cost, no obligation, and the provider you choose pays us.
Once a provider has your requirement directly, we can no longer represent you on that deal.
GPU servers: GPU rental, H100 rental, H200 rental, B200 and B300 rental, GB200 NVL72, GPU cluster rental, neocloud providers and CoreWeave competitors.
Moving off cloud: cloud repatriation, the cloud vs colo calculator, hybrid cloud and data center migration.
Owning instead: AI and GPU colocation, wholesale colocation, the colocation pricing guide, data center cost and the provider comparison.
About us: how a data center broker works, Equinix alternatives and data center consulting.
Also see: VMware alternatives and VMware licensing and pricing.