Hyperstack
On-demand NVIDIA GPUs by the hour from a London-registered company, with published spot prices, managed Kubernetes and data centres in Norway, Canada and Texas.
Who owns it
Hyperstack is the self-serve GPU platform of NexGen Cloud Limited, a private company registered in England and Wales (number 12556681, incorporated 15 April 2020) with its registered office at 99 Gresham Street in London. That matters for a UK buyer, because the contract is with a UK company under UK law in a category where almost every rival contracts from the United States. Control changed hands recently: Bure Valley Group Limited held 75% or more of the shares from 2020 until 14 May 2026, when Artemis 2 Limited (company 17174874) was recorded as holding between 50% and 75%. Co-founder Chris Starkey has been a director since incorporation, and co-founder Youlian Tzanev was appointed a director on 27 January 2026, the same day founding director Antony Wade resigned. No funding round or investor is disclosed publicly. The company owns its hardware and places it in other operators' data centres: NexGen Cloud says its Infrahub Compute arm "finances, sources, and deploys the physical compute hardware", and that "the majority of NexGen Cloud owned infrastructure is hosted in data centres powered by 100% renewable energy". Accounts to 31 December 2024 are on file at Companies House, and accounts to 31 December 2025 fell due on 30 September 2026.
## What it costs
Prices are in US dollars, VAT not stated, read 22 September 2026. One H100 SXM 80 GB with 24 vCPU and 240 GB of RAM costs $3.20 an hour on demand, which is $2,336 a month if you run it for all 730 hours. A reserved rate is advertised "from $2.72 an hour", about 15% less, but the term you have to sign for is not published anywhere, so the commitment behind that discount cannot be worked out in advance. The cheapest GPU is an RTX A4000 at $0.15 an hour, or $110 a month. Hyperstack also publishes spot prices, which most boutique GPU clouds do not: H100 PCIe at $2.00, A100 at $1.08, L40 at $0.80 and RTX A6000 at $0.40 an hour. No eight-GPU node is sold as its own product with its own price. Storage is quoted as roughly $0.10 per TB per hour, which works out at about $0.073 per GB a month, or $73 per TB. Egress and ingress are free. A public IPv4 address costs $0.00672 an hour, about $4.91 a month.
Billing runs per minute according to the pricing and billing-policy pages, although a second documentation page describes it as hourly. Two things will catch a small buyer out. A stopped virtual machine still bills in full: Hyperstack's own documentation says "billing continues for all resources, as they remain reserved exclusively for that VM". Only hibernation removes the GPU, CPU and RAM charge, and even then the root disk keeps billing at about $0.07 per GB a month and attached volumes keep billing at the standard rate. Running out of credit does not stop the meter either, and the documentation is blunt about it: "Reaching a zero balance does not stop billing. Resources on your account continue to accrue charges, which can take your balance negative until you recharge or the resources are removed." Accounts are prepaid by default and paid by card through Stripe with 3D Secure, and postpaid invoicing needs sales approval. The smallest realistic monthly bill for someone leaving one machine running is an RTX A4000 at $109.50, a public IP at $4.91 and a 100 GB volume at about $7.30, so roughly $122 a month. Swap in an H100 SXM and the same sum comes to about $2,341.
## What you get and what to watch
Signup is self-serve with a card, and no identity or business verification is documented. No published GPU quota applies to a new account, though the documentation points you at sales for increases or specialised access, so whether a brand-new account can start an eight-GPU H100 job on day one is not something the vendor states. Your balance has to cover everything already running plus one hour of the new machine before it will launch. The real constraint looks like stock rather than paperwork: a one-star review from July 2026 describes GPU inventory as "basically non-existent".
There are three self-serve regions: Norway (Vestland), Canada (Quebec) and the United States (Texas). There is no UK data centre, and the nearest option for a UK or EU buyer is Norway, which sits in the EEA but not the EU. The parent company also lists a Swedish deployment, but it is not a region you can pick in Hyperstack. Hyperstack does not publish which GPUs sit in which region, so you cannot confirm before signing up that H100 SXM or B200 capacity exists in Norway. The platform covers one to eight GPU flavours, InfiniBand on the HGX B200 and H100 SXM lines with the bandwidth unpublished, managed Kubernetes with a CSI driver and ingress, shared and bootable storage volumes, firewalls and snapshots, a full REST API with documented rate limits, an official Terraform provider, and Python, JavaScript and Go SDKs. No command line tool and no MCP server were found.
Compliance is where the published record thins out. NexGen Cloud's security page still presents SOC 2 Type 1 as a Q2 2025 target with Type 2 to follow, while a support article claims Type 2 has been achieved, and no report date or auditor is published either way. ISO 27001 was promised for early 2026 and there was still no certificate on the site in September 2026. The CLOUD Act is not addressed. Support runs through a Zendesk portal with tickets and live chat plus email, with no published hours or response targets, and prioritised support is sold as a feature of the Premium Postpaid tier, which tells you what the standard tier is not. The service level agreement link in the site footer returned a 404 on 22 September 2026, so there is no uptime figure or service credit schedule to quote. On Trustpilot the score is 2.5 from eight unprompted reviews, split 37% five-star and 63% one-star with nothing in between, and the company has not replied to any of them. The one-star themes are dated and specific: undisclosed storage charges leaving a negative balance of $29.26 (August 2026), no GPU stock (July 2026), support waits of 30 minutes and chat sessions disconnecting instantly (June and July 2026), and broken VM ports after a restart with a refund refused (June 2025). The August 2026 billing complaint matches Hyperstack's own documented policy, which makes it corroborated rather than anecdotal. No outage, breach, regulator action or lawsuit was found on the public record.
Not yet tested by Saaskly. Assessed on published pricing, vendor documentation, company filings, the vendor's own status and trust pages, and third-party reviews.
Pros
- +The contract is with NexGen Cloud Limited, registered in England and Wales as company 12556681, so a UK buyer gets a UK counterparty rather than a US one.
- +Spot prices are published rather than quoted: H100 PCIe at $2.00 an hour against $2.50 on demand, A100 at $1.08, L40 at $0.80 and RTX A6000 at $0.40.
- +Egress and ingress are free, and the entry GPU is an RTX A4000 at $0.15 an hour, about $110 a month.
Cons
- −A stopped machine bills in full. Hyperstack documents that billing continues for all resources in the SHUTOFF state, and hibernation still charges for the root disk at about $0.07 per GB a month plus any attached volumes.
- −Running out of credit does not stop charges. The documentation states that reaching a zero balance takes the account negative until you recharge or delete the resources.
- −No uptime figure or service credit schedule is published: the SLA link in the site footer returned a 404 on 22 September 2026, and no support hours or response targets are published either.
- −ISO 27001 was promised for early 2026 and is still unevidenced, and the security page contradicts a support article on whether SOC 2 Type 2 has been achieved.
Pricing
Minimum commitment: $2,336 ($2,336/mo × 1 seat × 1 month (no commitment)). The least you can spend before walking away; excludes numbers, minutes and add-ons.
H100 SXM 80 GB, one GPU
$2,336/mo · month-to-month
$1,986/mo on an annual commitment
Month-to-month costs 18% more
- ✓One H100 SXM 80 GB (24 vCPU / 240 GB RAM) at $3.20 an hour on demand, $2,336 a month at 730 hours; reserved from $2.72 an hour ($1,986 a month), term not published. USD, VAT not stated, read 22 Sep 2026.
RTX A4000, one GPU
$110/mo · no commitment
Same price on any term
- ✓One RTX A4000 (6 vCPU / 24 GB RAM) at $0.15 an hour, $110 a month at 730 hours. Public IPv4 about $4.91 a month, storage about $73 per TB-month, egress free. Read 22 Sep 2026.
Specifications
| Rating | 3.4 |
| Pricing and plans | |
|---|---|
| Committed H100 | $2.72/h, 15% off |
| Cheapest dedicated server | None |
| Billing | Per minute |
| Egress | Included |
| Committed pricing | Published |
| Free credit | None found |
| Features | |
| H100 per GPU-hour* | $3.20/h (NVIDIA H100 SXM 80 GB) |
| Cheapest GPU | $0.15/h (NVIDIA RTX A4000) |
| 8xH100 node | Not on demand |
| Spot / interruptible | ✓ |
| Storage (per GB/mo) | $0.073 |
| Self-serve signup | ✓ |
| GPU quota | None |
| UK location | No |
| H100 in the UK | No |
| EU locations | 0 |
| Locations | 3 |
| Cluster networking | ✓ |
| Kubernetes | ✓ |
| Slurm | No |
| Bare metal | No |
| Stopped instance billed | ✓ |
| Sales-only above | Postpaid billing, reserved terms and the single-tenant AI Supercloud |
| NVIDIA GPUs | ✓ |
| AMD GPUs | No |
| Multi-GPU nodes | ✓ |
| Dedicated servers | No |
| Object storage | No |
| Block storage | ✓ |
| Private networking | ✓ |
| Integrations and developers | |
| API | ✓ |
| Terraform | ✓ |
| MCP server | No |
| CLI | No |
| Ownership and trust | |
| Hardware** | Own hardware, colocated |
| Owned in | UK-owned |
| Owner | Artemis 2 Limited |
| Trustpilot (reviews) | 2.5 (8) |
| HQ | UK |
| Publicly listed | No |
| Sovereign offering | Local contracting entity |
| ISO 27001 | No |
| SOC 2 | No |
| 2FA | No |
| Support | |
| Phone support | No |
| Support channels | Ticket, Live chat, Email |
* One NVIDIA H100 80 GB per GPU-hour, on demand, in the currency the vendor shows a UK visitor; where the vendor sells no H100 the nearest GPU is named. Committed prices are the cheapest published rate and its minimum term.
** Who owns and runs the GPUs. "Third-party hosts" means the vendor is a marketplace for other people's machines and the SLA is theirs, not the vendor's.
Researched and drafted with AI assistance; reviewed and approved by Tim Meredith. How we use AI
Work at Hyperstack? Something not look right? Register in the vendor portal to see what we publish and send corrections.