IPv6-Only VPS in 2026: When It Works, When It Breaks
IPv6-only plans cut 30-50% off your VPS bill. A field guide to the workloads they fit, the ones they wreck, and the workarounds that hold up.
The discount is real, and so is the catch
An IPv4 address now trades for $30 to $50 on the secondary market. A full /24 block costs $7,500 to $12,000. Every dual-stack VPS you rent has that number baked into the price, quietly, forever. Which is why the IPv6-only tiers showed up on every price list at once: same vCPU, same RAM, same NVMe, 30 to 50 percent cheaper, because the IPv4 address is gone.
Sometimes that trade is a straight win. Sometimes it turns a $4 server into a weekend of wondering why half the internet cannot reach your app. The difference between the two outcomes is not luck. It is knowing which side of the network boundary your workload lives on.
A single IPv4 address is worth more than six months of rent for a $5 VPS. That is the whole reason this product category exists.
What the three plan types actually give you
Price tables flatten the distinction, so here it is plainly. Three kinds of VPS networking exist in 2026, and they are not interchangeable.
- Dual-stack - one public IPv4 plus a routed IPv6 block, usually a /64. Reachable from essentially 100 percent of the internet. Costs the most, because the provider is renting you a scarce address.
- IPv6-only - a routed IPv6 /64 (sometimes /48) and nothing else. Directly reachable from roughly 60 to 65 percent of the internet today. Outbound to IPv4-only services needs translation.
- NAT IPv4 - a shared IPv4 with a handful of forwarded ports, sometimes no dedicated address at all. Cheapest of all. Fine for a toy, frustrating for anything serious.
The 60 to 65 percent figure deserves a closer look, because averages hide the spread. Mobile carriers in the US run 70 to 85 percent IPv6. Residential ISPs sit around 40 to 60 percent. Enterprise networks, still stuck on legacy gear, sit at 20 to 30 percent. If your audience is consumers on phones, an IPv6-only server loses fewer visitors than you'd guess. If your audience is corporate networks, it loses most of them.
Where IPv6-only works without drama
The pattern behind every successful IPv6-only deployment is the same: the public internet is not the client. Something you control sits in front, or the workload was never meant to be public in the first place.
- Backend services - database replicas, Redis caches, internal APIs. Your frontend servers talk to them, you control both ends, and IPv4 never enters the conversation.
- CI and build runners - they pull from GitHub, which has full IPv6 support, and push to targets you control. CI machines are the single best fit for this pricing.
- VPN endpoints and bastion hosts - WireGuard and OpenVPN run over IPv6 without complaint. Your devices connect by address, not by DNS discovery.
- Monitoring agents and log shippers - outbound-only connections to a collector you also control.
- Staging and labs - environments nobody outside your team should touch anyway.
One operator running nine instances across four providers moved three backends to IPv6-only and saves $3 a month each, $108 a year. Not life-changing. But it funds an entire extra small VPS, and removing public IPv4 from services that were never supposed to be public also removes an attack surface. The cleaner architecture is the underrated part of that deal.
Where it breaks
Now the failure list, because the folklore is not entirely wrong. It is just badly aimed.
- Public websites - 35 to 40 percent of visitors on IPv4-only networks cannot connect, period. For a commercial site this is disqualifying on its own.
- Mail servers - plenty of receiving MTAs still do IPv4 MX lookups, and IPv6 PTR records are not guaranteed upstream. Deliverability suffers quietly.
- Game servers - Minecraft and friends still default to IPv4 clients. You lose a chunk of the player base before anyone says a word.
- Calling IPv4-only APIs - some payment processors and legacy services expose IPv4 endpoints only. Your IPv6-only box cannot reach them without a translation layer.
- External uptime monitors - a number of them still probe IPv4 only, so your server reads as down when it is fine.
Hostinger deserves its own sentence here: it shipped no IPv6 support at all in 2026, which makes it the outlier in the opposite direction. If IPv6 matters to your stack for any reason, that provider is simply off the list, benchmarks or no benchmarks.
The workarounds, honestly assessed
Cloudflare in front
Point a proxied DNS record at your IPv6-only origin. Visitors connect to Cloudflare over IPv4 or IPv6; Cloudflare fetches from your origin over IPv6. For a public website this gets you effectively universal reachability for free. What you added is a dependency on a third party's edge, which is a real trade, not a magic checkbox. For low-stakes sites it is an easy call.
NAT64 and DNS64
These solve the outbound direction: an IPv6-only server reaching IPv4-only services. Public NAT64 relays exist and work for casual use. Teams that want control run Jool or Tayga on a $1 nanode, allocate a /96 from their own space, and get per-customer accounting and abuse logs on top. Setup is under 30 minutes. What NAT64 does not do is make IPv4-only clients able to reach you inbound, which is the direction most public-site owners actually care about.
A dual-stack jump host
One cheap dual-stack VPS fronting any number of IPv6-only backends, with WireGuard or plain routing between them. This is the fleet pattern: one IPv4 rental, N workers behind it. IPv4 becomes a compatibility shim you pay for exactly once.
What the money looks like in October 2026
Concrete plans, checked this month. Prices move weekly in this market, so treat these as anchors, not gospel.
- Vultr - the $2.50 sandbox tier is IPv6-only: 1 vCPU, 512 MB, 10 GB SSD, 0.5 TB transfer, capped at two instances, no API automation. A testing tier, priced like one.
- BuyVM - IPv6-only slices from $2 a month versus $3 dual-stack, and every plan gets a routed /48. That is 1.2 trillion addresses per box if you enjoy that kind of thing.
- HostBrr - about $10.50 a year for 1 GB RAM, 15 GB NVMe, a /64 and 2.5 TB on 1 Gbps, in Germany or Finland. The IPv4-less annual price floor.
- DMIT Hong Kong Eyeball - IPv6-only tier from $8 a month on quarterly billing: 1 EPYC vCPU, 512 MB, 10 GB SSD, 10 Gbps port on CMI routing, 2 TB traffic. Direct-connect Hong Kong bandwidth at that price only exists because you gave the address back.
- Hostynet - French IPv6-only line at up to 28 percent under the matching dual-stack plan, 4 GB from about 6 euros a month.
- KhanWebHost - NAT VPS at $6.99 a year, 256 MB, 20 forwarded ports. Listed for completeness. Know what NAT means before you buy.
The other half of the story is what dual-stack now costs. On 1 October 2026 OVHcloud unbundled the public IPv4 from its Generation 3 Public Cloud instances: the address is now its own line at about 1.97 euros a month, attached or not. Hetzner's June repricing lists cloud plans excluding IPv4. The industry is making the IPv4 surcharge visible on purpose, and once you can see the line item, the incentive to delete it writes itself.
A decision procedure, not vibes
- Who initiates connections to this box? Only you and your own servers: IPv6-only is fine. The general public: keep reading.
- Is there something dual-stack in front that you control, or a CDN you are willing to depend on? If yes, IPv6-only origin plus proxy works for public sites.
- Does anything it calls outbound sit on IPv4 only? Check payment APIs specifically. If yes, budget a NAT64 path or a jump host before you commit, not after.
- Does it need to receive email or serve game clients? Stop here. Buy dual-stack.
- Still unsure? Buy dual-stack and spend the difference on doing something useful instead of debugging routing at midnight.
Bottom line
IPv6-only VPS is a specific tool for a specific network boundary, and in 2026 the boundary is easier to see than ever: public-facing, dual-stack; private and backend, IPv6-only and cheaper. The mistake to avoid is treating it as an invisible discount on a normal VPS. Test from outside your own network, put a real firewall on the box, and be honest about what you added when you added a proxy. The savings compound across a fleet. So do the shortcuts.