Table of Contents
- How this page ranks
- 32 GB plans ranked by monthly price
- Who needs 32 GB, and which plan fits
- Three price tiers for the same memory
- The two ratios that decide whether 32 GB fits
- EC2 and the hyperscalers at 32 GiB
- 32 GB VPS or a dedicated server
- Moving up from 8 or 16 GB
- What this page could not verify
- FAQ
- Sources
How this page ranks
One rule, fixed before the prices were read: the monthly price, as printed by the provider, of its smallest plan with at least 32 GB of memory in a US datacenter. Where a provider's ladder skips 32 GB, the next size up is used and its memory shown (Contabo's 48 GB). Where the only printed monthly figure is a multi-year price, it is shown with its term and renewal and flagged (Hostinger). Prices come from the sources verified on each provider's review: Vultr's and Linode's plan APIs, OVHcloud's US catalog API, Hetzner's price API, and the pricing pages of the rest. An earlier version of this page ranked providers by per-GB cost and reported Elasticsearch, Redis and Java heap tests with figures that could not be sourced; it has been replaced.
32 GB plans ranked by monthly price
| Provider | Plan | vCPU / RAM / disk / transfer | Monthly | Terms and notes | US locations |
|---|---|---|---|---|---|
| Hostinger | KVM 8 | 8 vCPU / 32 GB / 400 GB NVMe / 32 TB | $25.99 | 24-month term only; renews at $49.99; list $73.99 | Boston, Phoenix |
| Contabo | Cloud VPS 12 | 12 vCPU / 48 GB / 400 GB / unlimited | $30 | Introductory for 24 months; euro list price €25.00 with the dollar figure at Contabo's own 1.20 pairing; Cloud VPS 8 (24 GB) $16.80; a US location fee may apply | Carlstadt NJ, St. Louis, Seattle |
| InterServer | 16 slices | 8 cores / 32 GB / 640 GB SSD / 32 TB | $48 | Monthly, no contract; support help begins at 8 slices; the page states there is no trial period and no money-back guarantee | New York City, Dallas, Los Angeles |
| OVHcloud | 2027 Elite 8-32-160 | 8 vCore / 32 GB / 160 GB / unmetered | $74.80 | US catalog API; its configuration names US-EAST-VA and US-WEST-OR; $68.81 committed for 12 months and billed monthly, $65.38 paid upfront for 12; not one of the four sizes on the US VPS page, which stops at 24 GB | Vint Hill VA, Hillsboro OR |
| Linode | g7-highmem-2 (Linode 48GB) | 2 vCPU / 48 GB / 40 GB / 6 TB | $120 | Hourly; the memory arrives without proportional cores or disk; g6-standard-8 (8 vCPU, 640 GB, 16 TB) $192; g6-dedicated-16 (16 vCPU) $288 | Nine US regions; not Fremont |
| Hostwinds | Unmanaged Linux 32 GB | 8 CPU / 32 GB / 400 GB / 6 TB | $124.99 | Monthly; snapshots included on the unmanaged line; managed line higher | Dallas, Seattle |
| AWS Lightsail | Memory-optimized 32 GB | 4 vCPU / 32 GB / 320 GB SSD / 6 TB | $144 | Fixed bundle, hourly and capped; the standard 32 GB bundle (8 vCPU, 640 GB, 7 TB) is $164 | N. Virginia, Ohio, Oregon |
| Vultr | vc2-8c-32gb | 8 vCPU / 32 GB / 640 GB SSD / 6 TB | $160 | Hourly; API; eight US regions for this plan; vhf-8c-32gb (512 GB NVMe) is $192 in five US regions | Eight US regions |
| DigitalOcean | Memory-Optimized 32 GiB | 4 vCPU / 32 GiB / 100 GiB NVMe / 6,000 GiB | $168 | Hourly; its plan data also lists a Premium AMD Basic 32 GiB at 8 vCPU and 400 GiB for the same $168; General Purpose 32 GiB $252, Storage-Optimized $262, CPU-Optimized from $336; the Memory-Optimized line starts at $84 for 16 GiB | NYC 1 to 3, San Francisco 2 to 3, Atlanta, Richmond, Kansas City |
| Hetzner | CPX51 (US) | 16 vCPU / 32 GB / 360 GB NVMe / 1 TB | $279.49 | US price since June 15, 2026 per the Hetzner price API and its docs; IPv4 extra; third-party price aggregators still show the pre-June figure (about $70), which no longer applies to new orders | Ashburn, Hillsboro |
Not in the table: Kamatera, which sells 32 GB through its configurator without a printed price for that size; RackNerd, whose standard KVM ladder stops at 12 GB on the page this site read; BuyVM, whose slice ladder tops out below 32 GB; and Google Cloud and Azure, which are metered per machine type (an Azure B8as v2 or Google e2-standard-8 would be priced from their calculators, not from a plan). Hostinger's US VPS locations are Boston and Phoenix per its support page.
Who needs 32 GB, and which plan fits
Memory is the resource that cannot be borrowed: a server short of CPU gets slow, while a server short of memory swaps and then kills processes. The plans in the table are also not interchangeable at this size, because each workload wants a different shape of machine. Reading the table by workload, with the memory fixed at 32 GB or the nearest larger size:
| Workload | What it needs besides memory | Plan that fits | What to check first |
|---|---|---|---|
| Self-hosted PostgreSQL or MySQL with a hot set of tens of gigabytes | Cores for concurrent queries and checkpoints, plus disk for the data, the WAL and a backup copy | InterServer 16 slices ($48: 8 cores, 640 GB), or Linode g6-standard-8 ($192: 8 vCPU, 640 GB) when the API matters | RAM per core. A database on 2 cores and 48 GB holds the buffer pool and queues the queries; measure concurrent connections before choosing. |
| Elasticsearch or OpenSearch | The index is held in memory by design, and the heap is usually set to half the RAM | Contabo Cloud VPS 12 (48 GB, $30, 12 vCPU) is the most memory per dollar here; DigitalOcean Memory-Optimized ($168) if the index is small enough for its 100 GiB disk | Disk size against index size. The index on disk is typically larger than the memory image, so the volume caps the data before the 32 GiB of RAM does; DigitalOcean ships 100 GiB for the memory, InterServer 640 GB. |
| A JVM or .NET application with a heap in the tens of gigabytes | Heap plus headroom for metaspace, threads and direct buffers; cores matter less | Vultr vc2-8c-32gb ($160) if the environment is rebuilt from code; InterServer 16 slices if it is not | Heap sizing. A 32 GB plan runs a 20 to 24 GB heap comfortably; setting the heap at 32 GB leaves the operating system and the runtime nowhere to live. |
| Redis or Memcached holding a whole catalogue or session store | Memory is the product; disk and transfer barely register | Linode g7-highmem-2 (48 GB, $120) or Contabo Cloud VPS 12 (48 GB, $30) | Whether anything else lives on the box. Both plans are memory-first: 40 GB and 400 GB of disk respectively, and neither ships proportionate cores. |
| A Docker or Kubernetes node running a dozen services | Memory per service plus the runtime and orchestrator overhead | InterServer 16 slices (8 cores) or OVHcloud Elite 8-32-160 ($74.80, 8 vCore) | Cores per container. Eight services each with a small CPU share saturate 4 vCPU long before they exhaust 32 GB. |
| A build or CI host compiling large projects | Cores first, memory second, NVMe third | Vultr vhf-8c-32gb ($192, 512 GB NVMe) or Hostinger KVM 8 ($25.99 on a 24-month term, 400 GB NVMe) | Whether the budget is better spent on cores. Builds are CPU-bound, and the cheap plans here trade cores for memory in the other direction. |
| A website | Nothing on this page | 4 to 8 GB, on the 8 GB page | Whether the memory is actually the constraint. An uncached query or a runaway process looks like a memory problem and is not one. |
Memory is the resource that cannot be borrowed. A server short of CPU gets slow, while a server short of memory swaps and then kills processes. So the question at 32 GB is whether the working set, the data the workload touches repeatedly, is that large. It is for: a relational database whose hot tables and indexes run to tens of gigabytes (PostgreSQL's shared_buffers and the OS page cache, MySQL's innodb_buffer_pool_size, all want to hold the hot set); a search index, where the vendors' own sizing guidance keeps the index in memory; a Java or .NET application with a heap sized in the tens of gigabytes; an in-memory cache (Redis, Memcached) holding a whole catalogue or session store; a Docker or Kubernetes node running a dozen services each with its own runtime; and a build or CI host compiling large projects in parallel. It is not for a website. A WordPress, Node or Django site with a page cache and an object cache serves heavy traffic on 4 GB, and the usual reason a site "needs more RAM" is an uncached query or a runaway process, which 32 GB hides for a while and then does not. Measure before buying: free -m for memory actually used, and the database's own cache-hit statistics. If the number is under 8 GB, the 8 GB page is the one to read.
Three price tiers for the same memory
The table spans $25.99 to $279.49 for 32 GB, and the spread is not noise; it is three products.
Under $50: terms and introductory prices. Hostinger's $25.99 exists only on a 24-month commitment and becomes $49.99 at renewal; Contabo's $30 is an introductory price (its review covers the renewal and the possible US location fee); InterServer's $48 is the plain one, month to month, and support help starts at 8 slices. All three are shared vCPUs, monthly billing, no API, no hourly.
Because two of those three are conditional, the three-year arithmetic is worth doing before the monthly comparison decides anything. Hostinger's two years cost $623.76 at $25.99, and the third year costs $599.88 at the $49.99 renewal, so three years come to $1,223.64. InterServer at a flat $48 is $1,728 over the same three years, and Contabo is $720 for the first two with the third year unprinted. The term price therefore survives its own renewal at this size, and the reason to prefer the flat plan is the ability to leave rather than the total.
$75 to $170: hourly platforms and bundles. OVHcloud's Elite 8-32-160 from its catalog, Linode's 48 GB high-memory plan at $120, Hostwinds at $124.99, Lightsail's memory-optimized bundle at $144, Vultr at $160 and DigitalOcean's Memory-Optimized 32 GiB at $168. Here you are paying for hourly billing, an API, images and snapshots, more regions (Vultr's plan is available in eight US regions), or, at Hostwinds, a US company with its own datacenters and phone support.
Above $190: cloud brands and dedicated cores. Linode's standard 32 GB (g6-standard-8) at $192, its dedicated 32 GB at $288, Hetzner's US CPX51 at $279.49 (16 vCPU, but a US price set well above Hetzner's European one and with only 1 TB of transfer), and EC2 below. The memory figure is the same across all of them; what the extra money buys above this line is reserved cores on the dedicated plans, a larger platform, and, at Hetzner, a smaller transfer allowance than the plans below.
The two ratios that decide whether 32 GB fits
Memory is the headline number on a plan page and, at this size, the least informative one. Two ratios do most of the deciding: gigabytes of memory per core, and gigabytes of disk per gigabyte of memory. DigitalOcean publishes the ratios it builds its own families to, and they are a useful yardstick for reading anyone else's plan: Memory-Optimized 8 GB RAM per vCPU, Storage-Optimized 8 GB per vCPU with 146 to 225 GB of SSD per vCPU, General Purpose 4 GB per vCPU, CPU-Optimized 2 GB per vCPU.
| Plan | RAM per core | Disk per GB of RAM | What the ratio costs |
|---|---|---|---|
| Linode g7-highmem-2, $120 | 24 GB (48 GB on 2 vCPU) | 0.8 GB (40 GB) | The extreme end of a memory-optimized plan. It is a cache or a Redis instance and not much else: two cores cannot feed 48 GB of database, and 40 GB of disk is smaller than the memory above it. |
| DigitalOcean Memory-Optimized, $168 | 8 GB (32 GiB on 4 vCPU) | 3.1 GB (100 GiB) | DigitalOcean's own stated ratio for the family, so the cores are proportionate. The disk is thin for the money: 100 GiB holds a cache comfortably and a database only while its data, indexes and log together stay under that. |
| OVHcloud Elite 8-32-160, $74.80 | 4 GB (32 GB on 8 vCore) | 5 GB (160 GB) | Balanced cores, thin disk. Enough for an application server or a container host; a database holding 32 GB of data needs more room for indexes, WAL and backups than 160 GB gives. |
| InterServer 16 slices, $48 | 4 GB (32 GB on 8 cores) | 20 GB (640 GB) | The most disk per gigabyte in the table (20 GB against Hostwinds' 12.5 and OVHcloud's 5). This is the shape a self-hosted database wants: proportionate cores and room for the data, the indexes and a backup copy. |
| Hostwinds 32 GB, $124.99 | 4 GB (32 GB on 8 CPU) | 12.5 GB (400 GB) | Same ratio as InterServer at a different price, with 6 TB of transfer instead of 32 TB. |
| Hetzner CPX51, $279.49 | 2 GB (32 GB on 16 vCPU) | 11 GB (360 GB) | Cores-heavy: 16 vCPU on 32 GB suits parallel batch work rather than memory-resident data, and 1 TB of transfer is the smallest allowance in the table on a server that could be serving data. |
The cores ratio costs you differently from the disk ratio. Too few cores shows up under concurrency: the working set fits in memory and the service still slows when several clients arrive, because the queries queue for a core. Too little disk shows up on the first night the dataset grows past it, and the failure is harder to recover from than a slow query, because there is nowhere to write the new data. Both are reasons the memory figure alone cannot rank these plans.
Who each ratio hits is a question about how many people depend on the machine. One person running a cache, a queue or a single-tenant database can size by the ratio and adjust when it hurts, because the worst case is a restart on a machine nobody else is using. A team whose customers reach the service through that database does not have that slack: a resize at 32 GB is a migration, with a window and a rollback plan, and the ratio has to be right at the purchase rather than discoverable afterwards. For that buyer the columns that decide the purchase are cores and disk, and the memory figure is the one they settled before they started looking.
EC2 and the hyperscalers at 32 GiB
From AWS's published on-demand price feed for US East (N. Virginia), Linux, at 730 hours a month: t4g.2xlarge (8 vCPU Arm, 32 GiB, burstable at a 40% baseline) $196.22; t3a.2xlarge $219.58; m7g.2xlarge (8 vCPU Arm, not burstable) $238.27; t3.2xlarge $242.94; m7i.2xlarge $294.34; m7a.2xlarge $338.49. Add gp3 storage at $0.08 per GB-month, a public IPv4 at $3.65 a month, and $0.09 per GB out beyond 100 GB (the Lightsail vs EC2 page has the full line items). Lightsail's memory-optimized $144 bundle is already in the VPS table; the standard $164 bundle is the other AWS 32 GB product. EC2 competes with Linode's dedicated line and above. Google Cloud and Azure price 32 GB machine types in their calculators; the Google Cloud review and Azure alternatives page cover how those bills are built.
32 GB VPS or a dedicated server
At the bottom of the table, no dedicated server competes: nothing physical rents for $30 to $48 a month. At the top, the question is real. Vultr's bare-metal API listed dedicated servers from $120 a month in September 2026 (4 cores, 32 GB, in its Silicon Valley region, per the San Francisco page), which is less than Vultr's own 32 GB VPS and gives you the whole machine. Linode's Dedicated 32GB at $288 buys 16 reserved vCPUs on a shared host; a dedicated box at a similar price buys the host. The trade is operational: a VPS resizes, snapshots and reprovisions in minutes and a dedicated server does not. For a database that will live at 32 GB for years, price the dedicated option; for anything that may grow, shrink or move, stay virtual.
Moving up from 8 or 16 GB, and what sits above this page
Most readers arrive here from a smaller size rather than from nothing, and the path between sizes differs by provider. At InterServer the ladder is linear and public: 8 slices are $24 for 16 GB, 12 slices are $36 for 24 GB, and 16 slices are $48 for the 32 GB on this page, so moving up is a re-price at the same rate per slice rather than a migration to a different product. The hourly providers' resize mechanics are not stated here, because this page did not read them; check the provider's own documentation before relying on a resize.
In the other direction, if the working set will fit in half the memory, the 16 GB page and the 8 GB page cover the sizes below this one. At InterServer the step from 16 GB to the 32 GB on this page doubles the price along with the memory, from $24 to $48, so the smaller size is worth ruling out on the working set rather than on the budget. Above this page, two things change rather than one: at dedicated CPU the cores stop being shared, and at bare metal the whole machine is yours, which is the alternative to compare against the $192 to $290 rows here. If the reason for 32 GB is a provider rather than a size, the Contabo alternatives and Linode alternatives pages work through what a move would cost.
What this page could not verify
- Contabo's US location fee and its renewal price at 48 GB, and Hostinger's price after the second renewal: see their reviews. Both renewals are outside what the provider pages print.
- Kamatera's 32 GB price: configurator only.
- Any performance figure for any plan: none measured, none quoted.