Dedicated vs Shared vCPU: What the Premium Actually Costs in 2026
On Vultr, a dedicated vCPU costs 2.8 times as much as a shared one. Also on Vultr, it costs 12.5% more. Both figures come off the same price list on 13 September 2026: $10 against $28 for a server with 1 vCPU and 2 GB of RAM, and $640 against $720 for one with 24 vCPUs and 96 GB.
Providers sell "dedicated CPU" as one upgrade with one meaning. The price attached to it swings by a factor of 2.5 depending on who you buy from and how big you go. We took every shared and dedicated plan that has the exact same vCPU count and RAM at the same provider and found 18 matching pairs across four hosts. The premium ranged from 1.13x to 2.80x. At Vultr it shrinks as the server grows, while DigitalOcean and Linode lock it to a fixed multiple. And at Linode, once you need eight cores or more, the dedicated line is the cheaper way to buy them.
What a shared vCPU actually gives you
A shared vCPU is a slice of a hardware thread that other customers can be scheduled on too. DigitalOcean's documentation says so plainly: on a Basic Droplet, "the hyper-thread allocated to the Droplet may be shared between multiple other Droplets," and when those neighbours are busy, "a Droplet could receive fractions of hyper-threads instead of dedicated access to the underlying physical processors." Dedicated CPU Droplets "have guaranteed access to the full hyper-thread at all times" (DigitalOcean docs).
Akamai, which runs Linode, puts a number on it. Shared CPU cores "can be used at 100% for short bursts, but should remain below 80% sustained usage on average" (Akamai shared CPU docs). Dedicated instances are pitched for "sustained full-duty workloads (100% CPU all day, every day)" (Akamai dedicated CPU docs).
That 80% line is the most useful sentence any host has published on the subject. A shared vCPU is a core you may spike but should not pin.
How to see contention from inside the server
Time the hypervisor hands to someone else shows up as steal. In top it is the st field, which the man page defines as "time stolen from this vm by the hypervisor" (top(1)). vmstat reports the same figure in its own st column.
vmstat 1 30
# read the "st" column while your app is under loadA value that stays near zero while your service is busy means the neighbours are quiet. A value that climbs at the same moment your response times do means you are paying for their workloads in latency, and that is the evidence worth acting on.
Burst credits are shared CPU with a written limit
Some providers replace the vague fair-share arrangement with a meter. AWS burstable instances earn CPU credits below a baseline and spend them above it. A t3.micro's baseline is 10% per vCPU, and in Standard mode an instance that runs out of credits "gradually comes down to baseline CPU utilization and cannot burst above baseline until it accrues more credits." In Unlimited mode it keeps going and bills the surplus (AWS docs). Kamatera's Type T uses the same idea in plain dollars: "Exceeding an average CPU usage of 10% will be extra charged" (Kamatera FAQ).
The advantage of a meter is that you can read the ceiling before you hit it. With an ordinary shared vCPU you find the ceiling by hitting it.
How we matched the plans
- Prices are from our nightly catalog read of 13 September 2026. The same day we checked Linode against Akamai's pricing page and public API, Vultr against its public plans API, DigitalOcean against its Droplet pricing page, and Hetzner, Contabo and Kamatera against their own pages.
- The class comes from the product line, because a plan's CPU label almost never says "shared". DigitalOcean Basic Droplets count as shared; General Purpose and CPU-Optimized as dedicated. Linode Shared CPU against Dedicated CPU. Vultr Cloud Compute against Optimized Cloud Compute, which Vultr's docs call "dedicated virtual machines" (Vultr docs). Contabo Cloud VPS against Cloud VDS.
- A pair means the same provider, the same number of vCPUs and the same RAM. Where several plans in one class fit a shape, the cheapest counts.
- Monthly list prices in the currency the host bills. Contabo's figures are its 24-month-term rates. Both Contabo lines use the same term discounts, so the ratio is identical on a monthly contract (€25.00 against €69.00 for the 12-core pair).
- Not in the pair table: Hetzner and Kamatera, which have no matching shapes and get their own section below. Fly.io is out too, because its shared machines start at 256 MB per vCPU and its performance machines at 2 GB, so nothing lines up. So is DigitalOcean App Platform, a PaaS rather than a server.
The premium, pair by pair
| Provider | vCPU / RAM | Shared plan | Dedicated plan | Premium | Disk (shared / dedicated) |
|---|---|---|---|---|---|
| DigitalOcean | 2 / 4 GB | Basic Regular, $24 | CPU-Optimized, $42 | 1.75x | 80 / 25 GiB |
| DigitalOcean | 2 / 8 GB | Basic Premium AMD, $42 | General Purpose, $63 | 1.50x | 100 / 25 GiB |
| DigitalOcean | 4 / 8 GB | Basic Regular, $48 | CPU-Optimized, $84 | 1.75x | 160 / 50 GiB |
| DigitalOcean | 4 / 16 GB | Basic Premium AMD, $84 | General Purpose, $126 | 1.50x | 200 / 50 GiB |
| DigitalOcean | 8 / 16 GB | Basic Regular, $96 | CPU-Optimized, $168 | 1.75x | 320 / 100 GiB |
| DigitalOcean | 8 / 32 GB | Basic Premium AMD, $168 | General Purpose, $252 | 1.50x | 400 / 100 GiB |
| Linode | 2 / 4 GB | Linode 4GB, $24 | Dedicated 4GB, $36 | 1.50x | 80 / 80 GB |
| Linode | 4 / 8 GB | Linode 8GB, $48 | Dedicated 8GB, $72 | 1.50x | 160 / 160 GB |
| Vultr | 1 / 2 GB | CC Regular, $10 | Optimized CPU, $28 | 2.80x | 55 / 25 GB |
| Vultr | 2 / 4 GB | CC Regular, $20 | Optimized CPU, $40 | 2.00x | 80 / 50 GB |
| Vultr | 4 / 8 GB | CC Regular, $40 | Optimized CPU, $80 | 2.00x | 160 / 75 GB |
| Vultr | 4 / 16 GB | CC High Frequency, $96 | Optimized General Purpose, $120 | 1.25x | 384 / 80 GB |
| Vultr | 8 / 16 GB | CC High Performance, $96 | Optimized CPU, $160 | 1.67x | 350 / 150 GB |
| Vultr | 8 / 32 GB | CC Regular, $160 | Optimized General Purpose, $240 | 1.50x | 640 / 160 GB |
| Vultr | 16 / 64 GB | CC Regular, $320 | Optimized General Purpose, $480 | 1.50x | 1,280 / 320 GB |
| Vultr | 24 / 96 GB | CC Regular, $640 | Optimized General Purpose, $720 | 1.13x | 1,600 / 480 GB |
| Contabo | 12 / 48 GB | Cloud VPS 12, €20.00 | Cloud VDS L, €55.20 | 2.76x | 400 / 360 GB |
| Contabo | 16 / 64 GB | Cloud VPS 16, €29.60 | Cloud VDS XL, €75.20 | 2.54x | 500 / 480 GB |
Seven of the 18 pairs land on exactly 1.50x. The outliers sit at the two ends of Vultr's range and at Contabo.
Contabo's ratio is high because its shared side is so cheap. Its basic Cloud VPS plans cost €1.00 to €2.18 per vCPU per month on a 24-month term, far below anything else in this set (see its price history). The Cloud VDS line, according to Contabo, gives you "dedicated CPU and RAM inside a virtualized environment" with resources "committed to your instance alone, not pooled across tenants like a shared VPS." The same page concedes that a VDS "is also hosted on a shared physical server" (Contabo VDS).
Small servers pay the steepest markup
Vultr shows the size effect most clearly. Its smallest dedicated plan, Optimized CPU with 1 vCPU and 2 GB, costs $28. The shared plan of the same shape costs $10. For $20 you could have the shared 2 vCPU, 4 GB plan instead, which is twice the cores and twice the memory for $8 a month less than the single dedicated core.
Further up, the markup falls away. Against the Regular plans it goes from 2.80x at 1 vCPU to 2.00x at 2 and 4, then 1.50x at 8 and 16, and finally 1.13x at 24 vCPUs, where dedicated adds $80 to a $640 bill. Dedicated has a price floor at the bottom, and Vultr's shared line gets more expensive per core as it grows: $10 per vCPU on the Regular plans in the table up to 4 cores, $20 at 8 and 16, $26.67 at 24. Optimized General Purpose holds at a flat $30 per vCPU from 1 core to 64.
DigitalOcean and Linode do something different and keep the ratio fixed. DigitalOcean's CPU-Optimized Droplets cost exactly 1.75x the Basic Regular Droplet of the same shape at every size we could match. General Purpose costs exactly 1.50x Basic Premium AMD. Linode lands on 1.50x because it prices both lines by memory: $6 per GB of RAM on shared plans (the $5 Nanode aside) and $9 per GB on dedicated plans, without a single exception on either list. If you want the per-GB view across the whole market, we ran it in our VPS price per GB of RAM index.
Where dedicated is the cheaper way to buy cores
Linode's per-GB pricing has a side effect worth knowing about. As shared plans grow, they add memory faster than cores: 2 GB per vCPU at the bottom, 4 GB per vCPU at 8 and 16 cores, 6 GB per vCPU at 32. The dedicated line stays at 2 GB per vCPU all the way to 48 cores. Priced per core, the two lines cross between 6 and 8 vCPUs.
| vCPUs | Cheapest shared plan | Shared $/vCPU | Cheapest dedicated plan | Dedicated $/vCPU |
|---|---|---|---|---|
| 2 | Linode 4GB (4 GB), $24 | $12.00 | Dedicated 4GB (4 GB), $36 | $18.00 |
| 4 | Linode 8GB (8 GB), $48 | $12.00 | Dedicated 8GB (8 GB), $72 | $18.00 |
| 6 | Linode 16GB (16 GB), $96 | $16.00 | No 6-core plan | n/a |
| 8 | Linode 32GB (32 GB), $192 | $24.00 | Dedicated 16GB (16 GB), $144 | $18.00 |
| 16 | Linode 64GB (64 GB), $384 | $24.00 | Dedicated 32GB (32 GB), $288 | $18.00 |
| 32 | Linode 192GB (192 GB), $1,152 | $36.00 | Dedicated 64GB (64 GB), $576 | $18.00 |
At 32 vCPUs the dedicated plan costs half as much. You give up 128 GB of RAM to get there, and that is the whole point. If your build farm or encoder is CPU-bound and fits in 64 GB, the shared plan charges you $576 a month for memory you will never touch.
One caveat from outside our tracked catalog. Akamai's North America pricing page, checked on 13 September 2026, describes both the shared plans and the G6 dedicated plans above as running on legacy hardware, and lists a newer G8 Dedicated line at higher prices: $50.40 for 2 vCPUs and 4 GB, $194.40 for 8 vCPUs and 16 GB (Akamai pricing). On G8, 8 and 16 cores cost about the same as the shared plans with the same core count. The dedicated line only pulls clearly ahead at 32 vCPUs, $777.60 against $1,152.
Dedicated plans usually ship with less disk
The premium column hides a second cost. In 16 of the 18 pairs, the dedicated plan comes with less storage than the shared plan it replaces. Linode is the only one of the four that keeps disk identical across both lines.
- DigitalOcean at 2 vCPUs and 4 GB: 80 GiB on Basic, 25 GiB on CPU-Optimized, for $18 more per month.
- Vultr at 4 vCPUs and 16 GB: 384 GB on High Frequency, 80 GB on Optimized General Purpose.
- Vultr at 24 vCPUs and 96 GB: 1,600 GB on Cloud Compute Regular, 480 GB on Optimized.
Some of that gap is a change of drive type. Vultr's Regular plans and Contabo's Cloud VPS list plain SSD, while their dedicated lines list NVMe. But if you are sizing a database, read the disk column before the CPU label, because extra block storage is a separate monthly charge that no premium ratio includes.
Hetzner and Kamatera don't fit the grid
Hetzner makes more sense matched by memory
Hetzner has no shared and dedicated cloud server with the same cores and memory. Its dedicated CCX line gives 4 GB of RAM per vCPU; the shared CPX line gives 2 GB. The cheapest shared lines, CX and the Arm-based CAX, were also not orderable when we read the catalog on 13 September 2026: 8 of Hetzner's 14 shared cloud plans were marked unavailable. (Our ARM VPS guide covers what CAX-style Arm servers are good for.) That leaves CPX as the shared line you can actually buy.
Per vCPU, CPX costs €8.12 to €11.49 a month and CCX costs €16.67 to €21.50, before VAT and the IPv4 address. That reads like a 2x premium. Match the plans by RAM instead and most of it disappears.
| RAM | Shared (CPX) | Dedicated (CCX) | Premium |
|---|---|---|---|
| 8 GB | CPX32, 4 vCPU, €35.49 | CCX13, 2 vCPU, €42.99 | 1.21x |
| 16 GB | CPX42, 8 vCPU, €69.49 | CCX23, 4 vCPU, €85.99 | 1.24x |
| 32 GB | CPX62, 16 vCPU, €129.99 | CCX33, 8 vCPU, €138.49 | 1.07x |
So the real Hetzner question is whether half as many dedicated cores beat twice as many shared ones with the same memory. For a Postgres server with 32 GB, eight dedicated cores for €8.50 more than sixteen shared ones is an easy call.
The gap also moved this year. Hetzner's repricing of 15 June 2026 took CCX13 from €15.99 to €42.99 and CPX22 from €7.99 to €19.49, for new orders and rescales (Northflank's breakdown). Before it, a dedicated vCPU on those two plans cost exactly twice a shared one, €8.00 against €4.00. Afterwards it is 2.21x. Nothing about the hardware changed that morning, which tells you the premium is a pricing decision, and pricing decisions move. Hetzner's price trend page tracks where it goes next.
Kamatera sells four steps, and they are different units
Kamatera's headline gap is the widest we found. A server with 1 vCPU, 1 GB of RAM and 20 GB of NVMe costs $4 a month as Type A and $19 as Type D, which reads as 4.75x. The two types don't sell the same thing, though. From Kamatera's own definitions (Kamatera FAQ):
- Type A (Availability), from $4 a month: "a non-dedicated physical CPU thread with no resources guaranteed."
- Type B (General Purpose), from $9: "a dedicated physical CPU Thread with reserved resources guaranteed."
- Type T (Burstable): a dedicated thread, with usage above a 10% average billed on top.
- Type D (Dedicated), from $19: "a dedicated physical CPU Core (2 threads)."
Type D gives you two threads, which makes it $9.50 per thread, within 50 cents of Type B's $9 starting price. The step that buys a guarantee at Kamatera is A to B, about 2.25x on starting prices. Type D is for workloads that want both threads of a physical core to themselves. Kamatera's price history shows the current list.
These prices sit still
Nothing above comes from a flash sale. All 90 active DigitalOcean plans have shown one price each on every one of the 179 nightly readings we have taken since 12 February 2026. Linode's 39 plans have not moved across 169 readings since 24 February. Vultr's Cloud Compute and Optimized lines have held one price per plan since late February. The flat lines are visible on the DigitalOcean, Linode and Vultr trend pages. Hetzner, as covered above, is the exception.
That stability matters for the decision. On these three clouds the ratio has not budged in more than six months, so the premium is something you can plan a budget around.
When dedicated vCPU earns its price
Pay for dedicated when your CPU stays busy for hours at a time. CI runners, video transcoding, game servers and data analysis all appear on Akamai's list of dedicated workloads, and if your average sits above Linode's 80% sustained line, a shared plan is the wrong product at any price.
Latency-sensitive services are the second case. A database behind a checkout, or an API with a p99 target, suffers from a noisy neighbour long before the monthly CPU graph looks alarming. Watch st during your busiest hour. If it rises with your response times, the premium is buying you something.
Skip dedicated for blogs, low-traffic web servers and development environments, which is where DigitalOcean's own guidance places Basic Droplets. Staging boxes and internal tools belong in the same bucket. On small plans with a 2x or larger markup, my advice is to buy the next shared size up. Extra cores give a busy app headroom for less money, and if steal still shows up at that size, you will have the vmstat numbers to justify the upgrade.
What I would buy at each size
- WordPress site, staging box or side project: Vultr's shared 2 vCPU, 4 GB plan at $20. Hetzner CPX22 at €19.49 if you would rather be billed in euros.
- CI runner or encoder that needs 8 cores all day: Linode Dedicated 16GB at $144. It undercuts Linode's own shared 8-core plan by $48, and beats the nearest dedicated rivals, Vultr Optimized CPU at $160 and DigitalOcean CPU-Optimized at $168. If you are weighing the two US clouds head to head, the Linode vs Vultr page has the full catalogs.
- Database with 32 GB of RAM: Hetzner CCX33 at €138.49, €8.50 more than the shared CPX62.
- A 24 vCPU, 96 GB production box: pay Vultr the extra $80 for Optimized General Purpose ($720). At 1.13x it is the smallest markup among the 18 matched pairs.
And the one purchase I would avoid: a single dedicated core under $30, like Vultr's $28 Optimized CPU plan. Put $20 into two shared vCPUs and 4 GB, run vmstat under real traffic, and move to dedicated when steal shows up and not before. More plans sit in our VPS hosting and cloud hosting listings.
