VPS with Most Storage in 2026 — I Ran 72-Hour Write Tests So You Don't Burn Through Yours

The IOPS Throttling Trap Nobody Talks About

Every VPS review site will tell you Contabo offers 400GB for $6.99. They will show you a spec sheet, maybe a quick benchmark, and slap an affiliate link at the bottom. What they will not tell you is that the 25K IOPS on that spec sheet is a burst number. It is the performance you get for the first few minutes of heavy I/O. After that, the throttle kicks in.

I discovered this the hard way. I was migrating a 180GB PostgreSQL database to a Contabo VPS S and the initial import was screaming along at 450 MB/s sequential write. Beautiful. Then around the 12-minute mark, throughput dropped off a cliff. Down to 180 MB/s. The import that should have taken 25 minutes took nearly an hour. The database was not the bottleneck. The disk was being throttled.

This is not a Contabo-specific problem. It is an industry-wide pattern. Providers allocate burst IOPS from an SSD cache or I/O credit system. Short benchmarks — the kind most reviewers run — never deplete the credits. They tell you the sports car does 0-60 in 4 seconds without mentioning the engine overheats at mile 3.

So I did something that takes actual patience: I ran fio sustained write tests for 72 hours on every provider on this list. Not 30 seconds. Not 5 minutes. Seventy-two hours. Because that is how you find out what your storage actually delivers when the burst credits are long gone and the marketing numbers have stopped meaning anything.

The results changed my recommendations completely. Providers I expected to lead fell behind. Others I expected to struggle held steady. If you are spending money on storage, these are the numbers that should inform your decision — not the burst figures on a marketing page.

My fio Test Methodology

Transparency matters. Here is exactly what I ran so you can reproduce every number in this article on your own VPS. I used fio 3.36 on Debian 12 across all five providers, with direct I/O to bypass the OS page cache (because cached results are fantasy numbers).

fio --name=sustained-rw --ioengine=libaio --direct=1 \
    --rw=randwrite --bs=4k --iodepth=32 \
    --size=50G --runtime=259200 --time_based \
    --group_reporting --output-format=json \
    --log_avg_msec=60000 --write_iops_log=sustained

fio --name=seq-write --ioengine=libaio --direct=1 \
    --rw=write --bs=1M --iodepth=16 \
    --size=50G --runtime=259200 --time_based \
    --group_reporting --output-format=json \
    --log_avg_msec=60000 --write_bw_log=seqwrite

fio --name=mixed-rw --ioengine=libaio --direct=1 \
    --rw=randrw --rwmixread=70 --bs=4k --iodepth=32 \
    --size=50G --runtime=259200 --time_based \
    --group_reporting --output-format=json

Three tests. 72 hours each. Five providers. That is 45 days of cumulative benchmark time. The --direct=1 flag is critical — it bypasses Linux's page cache so you are measuring actual disk performance, not RAM performance disguised as disk speed. The --time_based flag ensures fio runs for the full duration even if the file size is exhausted. The 50GB test file is large enough to exceed any SSD cache on the host node.

I logged IOPS every 60 seconds (--log_avg_msec=60000) to produce time-series data showing exactly when and how throttling occurs. Full benchmark data for all providers is on our benchmarks page.

#1. Contabo — The Storage Giant with a Write Speed Asterisk

Let me be upfront: Contabo belongs on this list because nobody else gives you 400GB of SSD storage for $6.99 a month. At the top end, their Cloud VPS XL delivers 1600GB for $54.99. On pure capacity-per-dollar, Contabo is in a league of its own, and there are legitimate workloads where that is the only metric that matters.

But here is what my 72-hour test revealed.

The throttling pattern is consistent and predictable: burst IOPS depletes roughly 53% over 8 minutes, then stabilizes. It does not continue degrading. After the initial drop, 11,600 IOPS held rock-steady for the remaining 71 hours and 52 minutes. That is important. This is not a failing drive or random degradation. It is a deliberate I/O credit system, and the sustained floor is dependable.

11,600 sustained random write IOPS is not fast by NVMe standards. But let me contextualize that number. A Plex server serving a 4K movie reads sequentially in large blocks — it barely touches random IOPS. A Nextcloud server where 5 people sync files generates maybe 200-400 random IOPS during normal operation. A WordPress site with WooCommerce peaks at maybe 2,000 IOPS during a traffic spike. You would need to be running a moderately busy PostgreSQL instance writing WAL logs continuously to actually feel 11,600 IOPS as a bottleneck.

Where Contabo genuinely struggles: large bulk imports, database migrations, and any workload that writes continuously for hours. That PostgreSQL migration I mentioned earlier? It would have been fine on Hetzner. On Contabo, you wait. The full Contabo review covers CPU and network in detail.

Contabo Storage Tiers

The Bottom Line on Contabo Storage

If you are reading more than writing — media server, file hosting, archive — Contabo is arguably the best value on this entire page. 200GB at $6.99 with 32TB bandwidth is absurd value. But if your workload writes heavily and continuously, the sustained numbers matter, and you should scroll down to Hetzner. See our NVMe SSD VPS guide for providers where raw I/O speed is the priority.

#2. BuyVM — The $0.005/GB Secret That Sells Out in Hours

I need to structure this review differently than the others because BuyVM's storage story is not about the VPS itself. It is about the block storage slabs you bolt onto it. The base VPS ships with 10-80GB of local SSD — unremarkable. The magic is the $1.25 per 256GB block storage slab that turns a $3.50/mo VPS into a multi-terabyte storage server for pocket change.

How the Slab System Works

You order a KVM Slice (starting at $3.50/mo for 1 vCPU, 1GB RAM, 10GB SSD). Then you add block storage slabs at $1.25 per 256GB. They mount as regular block devices via /dev/vdX. Format with ext4 or XFS, mount wherever you want. Stack up to 16 slabs on a single VPS for 4TB total. Total cost for 1TB: $3.50 (VPS) + $5.00 (4 slabs) = $8.50/month.

That pricing is not a typo. It is not an introductory rate. It has been $1.25/slab for years.

There is something oddly refreshing about BuyVM's block storage performance. The numbers are not impressive — 4,800 random write IOPS is below every other provider on this list. But the curve is flat. No burst. No throttle. No surprises. Hour 1 looks identical to hour 72. The SSD cache absorbs frequently accessed data, and everything else hits enterprise HDD arrays at their honest, unaccelerated speed.

For the workloads that justify 1TB+ of storage, this performance profile is actually ideal. Plex reads large sequential files — 175 MB/s supports 7 simultaneous 4K direct-play streams. Nextcloud bulk syncs are sequential. Backup jobs are sequential. The only storage-heavy workload where 4,800 random IOPS hurts is a database, and if your database is larger than 80GB, you should probably be on dedicated database hosting anyway.

The Availability Problem

BuyVM's biggest flaw has nothing to do with performance. They sell out. Constantly. Block storage slabs in their Las Vegas, New York, and Miami locations go out of stock for weeks at a time. There is no waitlist. You either catch availability by checking their stock page daily, or you wait. I set up a monitoring script that pinged their order page every 15 minutes for three weeks before I was able to grab a Las Vegas slab for testing. If you see stock available and know you want it, do not deliberate. Order now, evaluate later, cancel within 7 days if it does not work.

Unmetered bandwidth is the other piece that seals the deal for media servers. On Vultr, serving 10TB of Plex streams per month would cost $100+ in bandwidth overages. On BuyVM, it costs $0. Read the full BuyVM review for CPU and network benchmarks.

#3. Hetzner — Where I/O Consistency Justifies the Premium

If the previous two providers taught you that cheap storage comes with compromises, Hetzner is where you learn what "no compromises" costs. And honestly, it costs less than you would expect.

Hetzner's storage architecture gives you two tiers on every server: local NVMe for your OS, databases, and anything I/O-sensitive, plus expandable Volumes at $0.052/GB for everything else. A CX42 plan with 320GB local NVMe and a 500GB attached Volume gives you 820GB of usable storage for $58.49/mo. Not the cheapest option on this list. Not even close. But the I/O performance makes it a completely different class of product.

Look at that local NVMe number. 48,200 sustained random write IOPS with less than 3% variation across 72 hours. No burst. No credits. No throttle. That is 4.2x Contabo's sustained performance and 10x BuyVM's block storage. The 0.65ms average latency means your database queries are not waiting for disk. Ever.

The Volumes tell a different story — 15,200 IOPS is respectable for network-attached storage but about a third of local NVMe. This is where Hetzner's two-tier architecture shines. You put PostgreSQL on the local NVMe. You put your Nextcloud data directory, media library, and log archives on the Volume. Each tier operates at its natural speed. The database never competes with media file I/O for the same disk bandwidth.

Live Volume Resizing

I tested expanding a 100GB Volume to 500GB via the Hetzner API while a database was actively reading from it. The API call returned in 3.2 seconds. The Volume was immediately usable at its new size. I ran resize2fs on the live filesystem — completed in under a second with zero I/O interruption. No reboot. No unmount. No application restart. Try doing that on Contabo (spoiler: you cannot; you upgrade the entire plan through a support ticket).

Hetzner is where I host anything that writes continuously: production databases, CI/CD build servers with heavy artifact generation, log aggregation pipelines. The per-GB cost is higher. The per-useful-IOPS cost is the lowest on this list. Read the Hetzner review.

#4. InterServer — Price-Locked Storage That Won't Creep Up on You

I almost dropped InterServer from this list during the rewrite. Their storage numbers are not exceptional: 30-480GB SSD, 35K IOPS on the spec sheet, no block storage add-on. But then I pulled up my cost spreadsheet for a project I have been running for 28 months, and InterServer was the only provider where the monthly bill had not changed once since signup. Not a single line item increase. That is their actual competitive advantage, and for long-running storage workloads, it matters more than you think.

The throttling pattern is gentler than Contabo's. InterServer drops from 33.4K to 22.1K over about 20 minutes — a 34% reduction versus Contabo's 53%. And 22,100 sustained IOPS is solidly usable for database workloads. Not great. Not embarrassing. Just honest mid-range SSD performance from a company running its own hardware in its own Secaucus, NJ datacenter.

The 3-Year Cost Argument

Here is where InterServer's price lock becomes tangible. A storage-heavy VPS is typically a long-term commitment — you are not spinning up a Plex server for a weekend. Let me run the numbers for 480GB over 36 months:

Okay, the raw numbers do not favor InterServer for pure storage capacity. Contabo gives you more storage for less money at every tier. But InterServer offers something you cannot see in a pricing table: 24/7 human support from technicians in the same building as your server, Windows VPS availability for NTFS-dependent workloads, and the guarantee that your $96/mo will still be $96/mo when everyone else has bumped prices 15-20%. For compliance-sensitive workloads that need a US-only provider with no surprises, that matters. See the InterServer VPS review.

#5. Vultr — When You Need Storage as Programmable Infrastructure

Vultr is the most expensive storage option on this page by a wide margin. At $0.10/GB for block storage, a 1TB volume costs $100/month — literally 20 times what BuyVM charges per gigabyte. I am not going to pretend that is good value for a Plex library. It is not. If you are reading this page because you want the most storage for the least money, Vultr is the wrong answer.

But Vultr is the right answer for a specific use case that none of the other providers can serve: programmable storage infrastructure. Let me explain what that means with a real scenario.

The Scenario That Justifies Vultr

I manage a data pipeline that processes satellite imagery. Every Tuesday, a 400GB dataset arrives. The pipeline needs to: create a block storage volume, attach it to a processing VPS, write the dataset, run analysis for 6-8 hours, export results to S3, and destroy the volume. The entire lifecycle is managed by a Terraform module.

Weekly cost: 400GB × $0.10/GB × (8 hours / 720 hours per month) = $0.44. Yes, forty-four cents. Because Vultr bills by the hour, temporary storage workloads cost almost nothing. Try doing that on Contabo where the minimum commitment is monthly and there is a setup fee.

28,400 sustained IOPS with no throttling on block storage is impressive. Vultr backs their volumes with NVMe (unlike BuyVM's HDD arrays), so you get genuine SSD-class performance on network-attached storage. The 1.1ms average latency is about half of BuyVM's, which matters for database-on-volume configurations.

The API That Makes It Worth It

Vultr's block storage API supports: create, attach, detach, resize, snapshot, and clone — all programmatically. You can snapshot a 200GB database volume before a risky migration, clone it to a test server in a different US city, validate the migration on the clone, then delete it. All from a CI/CD pipeline. 40TB maximum per volume. 10 volumes per server. 9 US datacenter locations. This is infrastructure-as-code storage, and it is where Vultr's premium buys you operational capabilities that BuyVM and Contabo simply do not offer.

For persistent, always-on storage needs, use someone cheaper. For ephemeral, programmable, snapshot-heavy storage workflows, Vultr is purpose-built. Full details in the Vultr review.

72-Hour Sustained Write Results — The Complete Picture

This is the table that should inform your purchase decision. Not burst speeds. Not marketing specs. These are the numbers each provider delivers after the SSD cache is exhausted, the I/O credits are depleted, and the disk is doing exactly what it will do for the next 12 months of your subscription.

Notice the pattern: NVMe providers (Hetzner, Vultr) show zero throttling. Their burst and sustained numbers are identical because NVMe does not rely on I/O credit systems in the same way SATA SSD does. BuyVM also shows zero throttling, but for the opposite reason — HDD arrays do not have burst capability, so the floor is the ceiling.

Storage Comparison Matrix

Which Provider for Which Workload

After running these tests, I no longer think of storage VPS as a single category. The right answer depends entirely on your I/O pattern. Here is the decision framework I use:

Still not sure? Use our VPS size calculator to estimate your storage and IOPS requirements based on your specific workload, or check the price comparison table for up-to-date pricing across all providers.

Frequently Asked Questions