JVPhoto placeholder
Co-founder · CEO
Founder name
Plans which racks we lease next and signs every reserved deal.
- Based
- EU-AMS-1
- Owns
- Capacity, contracts
About Nexpod
The GPU cloud we wanted to rent.
Nexpod rents whole 8-GPU servers by the hour, from our own racks in Amsterdam, Frankfurt and New York. Billed per second, invoiced in euros.
EU-AMS-1 · Hall B · Rack 07 live
Spine uplink
Leaf · compute fabric
GIGABYTE G894-ZD3-AAX7
8 x NVIDIA B300 SXM · 8U
B300 · 8 GPUs1 tenant
NVIDIA DGX H200
8 x NVIDIA H200 · 8U
H200 · 8 GPUs1 tenant
ASUS ESC N8A-E12
8 x NVIDIA H100 · 7U
H100 · 8 GPUsready soon
PDU A+B · 3-phase2 × 32 A
Servers
3 × 8 GPU
Inlet
22 °C
Fabric
400G
Hall B · cold aisle · 22 °C inlet
Mission
Whole servers, honest meters, euro invoices.
Most clouds sell a slice of a GPU and a bill you need a spreadsheet to read. We went the other way.
Every Nexpod server is a full 8-GPU machine that belongs to one customer for as long as it runs. No slices, no neighbours.
We rack the hardware ourselves, meter it by the second and invoice from the EU. Nothing clever on top, just the basics done with care.
8GPUs
in every node
20
server models
from 4 vendors
3
live regions
1 more in build
What we hold to
Five rules. They break every tie.
They show up in the product, on the invoice and on the status page. When a decision is close, these settle it.
Shared host · 5 tenantsnoisy neighbours
ABACDBEC
Nexpod node · 1 tenant8 / 8 yours
youyouyouyouyouyouyouyou
01
Whole machines only
No GPU slices, no shared hosts. A node is eight GPUs and one tenant.
run · train-b300-07 stopped
- started
- 09:14:07
- stopped
- 10:52:51
- billed
- 6,524 s × 8 GPU
- rate
- €0.00167 /GPU·s
total€86.99
02
The meter is the contract
Billing starts when the server boots and stops when it does. To the second.
Invoice · OctEUR
H200 server · 212 h€6,784.00
H100 server · 96 h€2,304.00
VAT · reverse charge€0.00
Total€9,088.00
03
Priced in euros
EU entity, EU invoice, reverse-charge VAT. No exchange-rate surprises.
- 14:02DegradedEU-FRA-1 · boot times up
- 14:09IdentifiedImage cache node full
- 14:31ResolvedCache rebalanced
- +2 dPostmortemPublished, in full
04
Say what broke
Incidents go on the status page while they happen. Postmortems go up in full.
your appEU-AMS-1
Data at restEU
BackupsEU
Support accesslogged
Billing entityEU
05
Data stays where you put it
Pick a region. Data, backups and support access stay inside it.
Since 2024
Sample datesFrom one cage to three regions.
We grew one rack at a time, and only into places where we run the hardware ourselves.
EU-AMS-1
EU-FRA-1
EU-HEL-1soon
US-NYC-1
Live region Planned3 live · 1 planned
- 2024
Founded in Amsterdam
Three engineers, one rented cage and a plan to rent GPU servers whole, never in slices.
- Q2 2025
EU-AMS-1 goes live
The first H100 and H200 servers, billed per second from day one.
- Q4 2025
Serverless endpoints
Autoscaling GPU endpoints for inference, with workers that scale to zero.
- Q1 2026
EU-FRA-1 and US-NYC-1
A second EU region for redundancy and a first US region for east-coast teams.
- Q2 2026
Instant Clusters
Multi-node clusters of 8-GPU servers on one 400G fabric, ready in minutes.
- Q3 2026
B300 arrives
NVIDIA DGX B300 and GIGABYTE B300 servers join the fleet, next to B200.
- Now
EU-HEL-1 in build
A fourth region in Helsinki, with free-air cooling for most of the year.
Careers
Work on the machines, not around them.
If you like hardware you can touch and software that has to be right, we should talk.
nexctl — careers
How we work
- 01
Engineers answer tickets.
Support is not a separate floor. The people who write the scheduler read the tickets about it, so the fix and the reply usually come from the same person.
- 02
Write it down first.
Every change bigger than a day starts as a short doc that anyone can comment on. Decisions live next to the reasoning, so nobody has to ask twice why a thing is the way it is.
- 03
Small teams, whole problems.
A team owns a problem from the rack to the invoice line. Fewer handoffs means fewer gaps, and the person on call understands the system they are paged for.
- 04
Measure, then argue.
Boot times, queue depth, failure rates per GPU family. We put the number on the table before the opinion, and we change our mind when the number says so.
- 05
Leave the rack cleaner.
Label the cable, update the runbook, delete the dead flag. Small acts of care compound across every 8-GPU node we run, and so does the lack of them.
Working hours · UTC core 15–17 CET
Amsterdam
CET
Frankfurt
CET
Helsinki
EET
New York
ET
0006121824
Remote and hybrid
Amsterdam first. Remote across Europe.
Most of us work near a data center: Amsterdam, Frankfurt or New York. Everyone else works remotely from Europe. Two shared core hours keep the time zones honest.
Team
PlaceholderThe people behind the racks.
Photos and names follow once the team signs off. Until then: who does what, and where.
MKPhoto placeholder
Co-founder · CTO
Founder name
Owns the placement engine, the 400G fabric and the on-call rota.
- Based
- EU-AMS-1
- Owns
- Scheduler, fabric
Roster · 8 people
- AMS 5
- FRA 1
- NYC 1
- HEL 1
NameRoleFocusHub
- SBTeam memberHead of Infrastructure · Racks, power, coolingHead of InfrastructureRacks, power, coolingFRA
- LDTeam memberPlatform Engineering · nexctl and the APIPlatform Engineeringnexctl and the APIAMS
- TATeam memberSite Reliability · On-call and incidentsSite ReliabilityOn-call and incidentsHEL
- RNTeam memberData Center Operations · Hands in the racksData Center OperationsHands in the racksAMS
- ECTeam memberSolutions Engineering · US customersSolutions EngineeringUS customersNYC
- IHTeam memberFinance & Billing · Invoices in eurosFinance & BillingInvoices in eurosAMS
Start here
Rack 08 has room. Take a node.
Pick a GPU family and a region. Your server boots in about a minute, and the meter stops the second you stop it.
nexctl — first node