Managed Kubernetes
sans virtualisation

unStack is a cloud where every server is physically yours.
We run the Kubernetes; you get whole machines — no hypervisor, no noisy neighbours, no surprise bill.

100%
dedicated hardware
400k IOPS
NVMe storage
0
hypervisor
unstack — one command, real machines
$ unstack describe my-cluster
isolationhardware · dedicated
bare-metalfull machine · no virtualization
storageNVMe · 400k IOPS, in-machine
regionspans 2 AZs · both zones active
billingflat / month · 4 lines
hypervisornone · removed
✓ cluster ready — deterministic performance
How it
works
1
Reserve a cluster
Pick a size. One flat monthly price.
2
We dedicate real servers to you
Physical machines in two availability zones — yours alone.
3
Deploy with kubectl
Standard Kubernetes & S3. Nothing proprietary to learn.
00 / why

The cloud you rent was designed for problems you had 20 years ago.

Back then apps were CPU-bound and distributed systems barely existed, so mutualising a few big servers with a hypervisor made sense.
That trade-off is now baked into your bill — and it no longer holds.

err: performance
Non-deterministic
Overcommit from virtualization means the machine you pay for isn't the machine you get.
err: resilience
Regions aren't independent
Shared control planes turn a local incident into a global outage. You've seen the headlines.
err: security
Shared silicon leaks
Spectre-class side channels are structural to running strangers on the same CPU as you.
warn: complexity
Non-standard services
Hundreds of proprietary services with unclear value — and a lock-in you can't leave.
warn: pricing
Pay-per-everything
A confusing, variable bill with a poor performance-per-euro ratio, by design.
// clients want
// to take back
// control.
01 / product

Four standard parts.

Only the services that actually matter: Kubernetes delivered on hardware you don't share with anyone and S3.

/compute

Managed Kubernetes

Fully managed K8s clusters on dedicated microservers — no virtualization. Deterministic performance, hardware isolation, an immutable hardened Linux built for one job.

/hardware

Microservers

Bare-metal at the price of a VM, each dedicated to one client, only 5–30 watts.

cpuARM 8–12c
ram16–64GB
diskNVMe
net2.5–5GbE
/services

Managed services, inside your cluster

Multi-tenant cloud services are replaced by Kubernetes operators installed right inside your own cluster. We manage them for a familiar experience — you keep full control. The ecosystem is open; you pick your level.

/object storage

S3

Standard Simple Storage Service hosted and managed outside your cluster by unStack on a dedicated infrastructure.

02 / the unstacking

Kubernetes already gives you the resilience the hypervisor was invented for.
So we removed everything in between — three whole layers of overhead.

TODAY — virtualized cloud · follow one disk write ↓
1your application
2guest OS
3virtual machine← toll
4hypervisor← toll
5host OS← toll
6shared hardware← queues
7network → remote disk · your data is a cable away
7 stops — including a ~10 ms round trip on the network.
rm -rf
3 layers
+ the hop
UNSTACK — bare-metal Kubernetes · follow the same write ↓
1your application
2container · Kubernetes
3immutable Linux
4dedicated hardware — NVMe inside← data lives here
— nothing else —
4 stops. The disk is in the machine, ~0.1 ms away.
03 / performance

Databases are the cloud's worst case. They're our best one.

50–70% of cloud workloads are database-backed apps — ERP, CRM, HR, collaboration, custom LOB tools.

They're I/O-bound. In a traditional cloud the disk is a network hop away — remote block storage, behind the hypervisor. On unStack the NVMe lives inside the machine, direct-attached, so you get its native throughput instead of a throttled fraction of it.

And IOPS aren't even the whole story — latency is what kills you. A database query is thousands of small, sequential I/Os: at 10 ms each, your app spends its life waiting. Direct NVMe answers in ~0.1 ms.

Customers used to pay a fortune for catastrophic performance — most never realized it, because there was no alternative.

// until now.

bench — db workload (i/o-bound)
$ unstack bench --profile=postgres-oltp --metric=iops,latency
their writeappvmhypervisornetworkremote disk
our writeappNVMe
virtualized · remote block storage~20k IOPS
unStack · direct-attached NVMe400k IOPS
latency per I/O — what every query actually waits on
>10 ms
virtualized · disk across the network
~0.1 ms
unStack · NVMe in the machine
~20×
the IOPS
~100× lower
latency per I/O
= every run
deterministic
* illustrative — I/O-bound OLTP profile, direct NVMe vs remote block storage
04 / resilience

One cluster, stretched across two availability zones.

On unStack a cluster isn't pinned to one availability zone. By default it spans two AZs, ~400 km apart (to protect against natural disasters).
A Kubernetes control plane keeps quorum across both — so an entire AZ can go dark and your cluster keeps serving.
It's the default shape of every cluster, not a disaster-recovery bolt-on.

1 Kubernetes cluster — one control plane, spanning both zones
Availability Zone Aphysical site 1
node-a1control-plane
node-a2control-plane
node-a3worker · dedicated
quorum400 km
Availability Zone Bphysical site 2
node-b1control-plane
node-b2worker · dedicated
node-b3worker · dedicated
T+0 — nominal
A
B
Both zones serve traffic. One control plane, quorum across the two sites.
T+1s — incident
A
B ✗
Zone B loses power — machines, network, everything at once.
T+90s — steady
A
B · offline
✓ Still serving. Quorum holds in A, pods reschedule automatically. Nothing to configure.
// default, not add-on
Every cluster spans two zones out of the box — no second cluster to run or sync yourself.
// independent regions
Regions share nothing. A failure in one can never cascade into another.
// S3 restore, outside
Clusters restore from a standard S3 service, one per region, hosted outside your clusters.
05 / energy

Every watt goes to your workload.

A microserver draws 5–30 watts, powered over Ethernet — a lightbulb, not a radiator. A virtualized rack server pulls hundreds of watts before your first request is served.

Unstacking is itself an energy plan: the hypervisor and host OS burned watts on bookkeeping, and overallocation kept oversized machines idling at full power. Remove the layers, and that tax disappears with them.

5–30 W per node
Zero watts spent running hypervisors
Right-sized nodes — nothing idles at full power

// efficiency is the architecture

power — audit
$ unstack power --audit --per-node
virtualized rack server~600 W
unStack microserver5–30 W
where the watts go — share of power doing useful work
virtualized~55% useful
useful workvirtualization taxidle overallocation
unStack~95% useful
~20×
less power per node
PoE
power + data
* illustrative — typical rack server vs microserver, useful-work share
06 / support

New to Kubernetes? We take you there.

Migrating to unStack is a project — we run it with you. Our engineers audit your current setup, plan the move and execute it by your side, then keep your cloud spend under control.

01/
Audit & migration plan
We analyze your workloads, your architecture and your current bill, then deliver a step-by-step migration roadmap with projected savings.
02/
Hands-on migration
Our engineers move your applications and databases to managed Kubernetes on unStack — zero-downtime strategy included — with knowledge transfer for your team.
03/
Cost optimization
FinOps review built in: flat monthly billing, no virtualization tax, no egress fees. We tune your clusters until the numbers drop.

engagements on quote

// how we work
your team unStack
01 · scoping & audit
02 · roadmap & quote
03 · build & migration sprints
04 · run, support & cost reviews
07 / pricing

A bill you can read in ten seconds.

A subscription, not metered pay-per-everything. A handful of line items instead of hundreds. You know what next month costs before it arrives.

An unprecedented price/security/performance ratio
Flat monthly subscription per cluster
A limited, fixed number of billing lines
Great performance-per-euro, by design
No egress traps, no lock-in
// legacy cloud invoice
compute-hrs
data-transfer-out
iops-provisioned
nat-gateway-hrs
load-balancer-lcu
snapshot-gb-month
api-requests-t2
inter-az-traffic
public-ipv4-addr
metrics-ingestion
dns-zone-queries
secrets-rotation
egress-acceleration
support-plan-3%
compute-hrs
data-transfer-out
iops-provisioned
nat-gateway-hrs
load-balancer-lcu
snapshot-gb-month
api-requests-t2
inter-az-traffic
public-ipv4-addr
metrics-ingestion
dns-zone-queries
secrets-rotation
egress-acceleration
support-plan-3%
…it keeps scrolling (+2,980 more)
// unstack invoice
cluster€ ———
nodes ×N€ ———
S3 storage€ ———
total€ flat
pricing published at public release · join the beta for early access
08 / sovereignty
European by design
Built and operated in Europe, for a market that needs a real alternative and strategic autonomy.
ISO 27001 / HDS
Certification in preparation — including health-data hosting.
Boundless network
Fully dual stack (IPv4 / IPv6), real IP traffic without protocol limitations, no ingress or egress fees.
// the cloud, after virtualization

Experience a disruptive
Managed Kubernetes.