Hyperstack vs Cerebrium
Both are GPU clouds, but they hand you different things. Hyperstack rents you a whole virtual machine with root and SSH that runs until you stop it, while Cerebrium runs your model or app as containers behind endpoints that start on a request and scale back to zero.
The companies and products
Hyperstack
Hyperstack is the GPU cloud of NexGen Cloud, a UK company founded in 2020: you rent GPU and CPU virtual machines, Kubernetes clusters and storage by the minute.
A Hyperstack VM is a full virtual machine with root and SSH, booted from an Ubuntu image (with or without Docker) and sized from a fixed list of flavors, from one RTX A4000 to 8-GPU B300 nodes, plus CPU-only flavors from 4 to 32 vCPUs. It also runs managed Kubernetes, volumes, object storage and an AI Studio inference API. It is built for teams that train, fine-tune and serve models, not as a sandbox API for agents, so this page compares the parts that overlap: VMs as machines for agent work.
Cerebrium
Cerebrium is a serverless GPU cloud for real-time AI: voice agents, video models, LLMs and other ML apps.
You describe an app in a cerebrium.toml file, with your Python code or your own Dockerfile, and Cerebrium builds it and serves it as HTTP or WebSocket endpoints. It starts containers when requests arrive, scales them across GPUs and regions, and scales them back to zero when traffic stops. It is built for teams that serve models in production, not for agents that need a machine of their own, so this page compares the places where the two overlap.
Which use cases each one fits
For each use case, Battleships prices every sandbox provider on the same workload and sorts them from the cheapest monthly bill up. “122nd cheapest of 377” means 121 of the other 376 providers run that workload, with every feature it needs, for less. “With limits” means the provider runs it but misses something, so it is listed after every full match. Open a use case for its whole ranking and what each provider misses.
| Use case and its workload | Hyperstack: price rank | Cerebrium: price rank | bochka: price rank |
|---|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month A coding agent needs root, Docker and a disk that lasts: Hyperstack has all three but bills a stopped VM in full, and Cerebrium keeps only one folder and documents no Docker inside. | 343rd of 367, with limits $11,458 a month | 182nd of 367, with limits $2,008 a month | 3rd cheapest of 367 $634 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB | 337th of 366, with limits $12,775 a month | 221st of 366, with limits $2,534 a month | 5th cheapest of 366 $657 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Short code runs suit Cerebrium's model of a container that starts on a request and bills per second, while a Hyperstack VM bills by the minute until you stop it. | 172nd of 345, with limits $1,167 a month | 225th of 345, with limits $131 a month | 186th of 345, with limits $2,000 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month Cerebrium can start many user containers at once (500 on Hobby), while Hyperstack does not publish how many VMs can run at once; bochka gives each user a full VM. | 173rd of 361, with limits $4,207 a month | 230th of 361, with limits $821 a month | Cheapest of 361 $209 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month Neither fully fits RL rollouts here; Cerebrium's per-second containers come out lower than Hyperstack's per-minute VMs. | 173rd of 360, with limits $17,500 a month | 72nd of 360, with limits $4,068 a month | 34th of 360, with limits $2,000 a month |
| Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month | 204th of 364, with limits $3,250 a month | 32nd cheapest of 364 $639 a month | 10th cheapest of 364 $200 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month Both fit CI jobs, Hyperstack as full VMs billed by the minute and Cerebrium as containers, while bochka adds Docker and /dev/kvm on every machine. | 122nd cheapest of 377 $1,560 a month | 81st cheapest of 377 $766 a month | 4th cheapest of 377 $86 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A 24/7 dev box fits Hyperstack, a full VM with SSH and a whole disk; Cerebrium has no SSH and resets everything outside /persistent-storage, and bochka keeps the whole disk for $26 a month. | 115th cheapest of 368 $475 a month | 259th of 368, with limits $102 a month | 18th cheapest of 368 $26 a month |
| Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Neither has a desktop or browser for computer-use agents; bochka streams a 60fps desktop with Chromium in every sandbox. | 332nd of 369, with limits $3,250 a month | 179th of 369, with limits $575 a month | 6th cheapest of 369 $200 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month | 335th of 367, with limits $1,750 a month | 221st of 367, with limits $287 a month | 7th cheapest of 367 $200 a month |
| Scraping & bots (startup scale) 50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month | 327th of 346, with limits $1,750 a month | 223rd of 346, with limits $195 a month | 6th cheapest of 346 $90 a month |
| Mass scraping & bots (undetectable) 1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month | 310th of 338, with limits $35,000 a month | 135th of 338, with limits $3,991 a month | 71st of 338, with limits $1,800 a month |
| Parallel red teaming 200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month | 309th of 360, with limits $876 a month | 197th of 360, with limits $191 a month | 2nd cheapest of 360 $200 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month Both fit GPU eval jobs: Hyperstack ranks 2nd of 98 at $625 a month and Cerebrium 28th at $1,975, and bochka has no GPUs. | 2nd cheapest of 98 $625 a month | 28th cheapest of 98 $1,975 a month | Can't run it |
| macOS agents 5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month | Can't run it | Can't run it | Can't run it |
Key features
| Feature | Hyperstack | Cerebrium | bochka |
|---|---|---|---|
| What you get | A full VM with root and SSH, sized from a fixed list of flavors | An app you deploy as containers built from your code or Dockerfile, served as endpoints | A full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts, which you or an agent work inside |
| Shell access | SSH and root | No SSH; you reach the app through HTTP or WebSocket endpoints | SSH, root and systemd, like a normal Ubuntu server |
| Docker inside | Ubuntu images with or without Docker | The app is the container; Docker inside and /dev/kvm are not documented | Docker, Docker Compose and /dev/kvm on every machine |
| Files after a stop | The whole disk stays, but a stopped (SHUTOFF) VM bills at the full rate; hibernation bills only disk and a kept IP | Only the shared /persistent-storage volume survives; packages and files elsewhere reset | The whole disk is saved every minute and on stop, and a stopped sandbox costs nothing |
| Snapshots and fork | Snapshots by hand with the VM shut off; no fork | No fork; an early-beta memory checkpoint speeds up new containers of the same build | Disk snapshot every minute while running, and fork of a running sandbox |
| Scale to zero | No; a VM runs until you stop, hibernate or delete it | Yes; apps scale to zero and start again on the next request | No; a stopped sandbox is started by API |
| Run time | No maximum runtime documented | No container cap, but a request times out at 15 minutes by default, 12 hours at most for async runs | No session cap on paid plans; you set the time to live |
| GPUs | Nvidia RTX A4000 to H100, H200, B200 and B300, single cards or 8-GPU nodes, on demand, spot or reserved | T4 to B200 and RTX PRO 6000, up to 8 per container, plus Inferentia2 and Trainium | None, CPU only |
| Inbound access | A public IPv4 you can keep static; open ports with firewall rules and run your own TLS | One endpoint per app on one port; custom domains on Standard | HTTPS URLs for up to 50 ports, token-protected; raw TCP; dedicated IPv4 or IPv6 |
| Network rules | Per-VM firewall: inbound closed by default, outbound rules by IP, protocol and port | No outbound allowlist documented | Open internet out; inbound through HTTPS URLs and TCP ports |
| Desktop and browser | SSH and text console logs; no graphical desktop | No desktop or browser product | Linux desktop streamed at 60fps with Chromium and computer use |
| Coding agents and keys | Nothing runs an agent inside a VM; keys go in by cloud-init, files or environment variables | No agent to run inside; secrets reach the app as environment variables | Claude Code, Codex and more preinstalled with a prompt, steer and events API; a credentials proxy keeps keys outside |
| Regions | Norway, Canada (Quebec) and the US (Texas) | US East, US Central, Stockholm and Finland; 8 more, including Tokyo and Sao Paulo, on request | EU only: Germany, Finland, France |
Every feature Battleships tracks (43)
| Feature | Hyperstack | Cerebrium | bochka |
|---|---|---|---|
| Full VM (own kernel) | Yes | No | Yes |
| Docker inside | Yes | No | Yes |
| Nested virtualization | Not published | No | Yes |
| Root | Yes | Not published | Yes |
| Systemd | Not published | Not published | Yes |
| Desktop GUI | No | No | Yes |
| Browser | No | No | Yes |
| Computer-use API | No | No | Yes |
| SSH | Yes | No | Yes |
| Persistent disk | Yes | Yes | Yes |
| Pause/resume | Yes | No | Yes |
| Automatic snapshots | No | No | Yes |
| Fork/clone | No | No | Yes |
| Live fork (no pause) | No | No | Yes |
| Memory fork | No | No | No |
| Custom image (Docker or snapshot) | Yes | Yes | Yes |
| ≥24 h sessions | Yes | Yes | Yes |
| Idle auto-stop | No | Yes | No |
| Wake on request | No | Yes | No |
| HTTPS preview URLs | No | Yes | Yes |
| Raw TCP inbound | No | Not published | Yes |
| Public IPv4 | Yes | No | No |
| Egress allowlist | Yes | No | No |
| Secret proxy | No | No | Yes |
| Secret proxy for any API | No | No | Yes |
| Preinstalled agents | No | No | Yes |
| Agent harness API | No | No | Yes |
| MCP server | Yes | No | No |
| GPU | Yes | Yes | No |
| Windows | No | No | No |
| Arm64 | No | No | No |
| Self-hosting | No | No | No |
| BYOC | No | No | No |
| Open source | No | No | No |
| Scoped API keys | Not published | Not published | Yes |
| Spend limits | Not published | Not published | Yes |
| Usage per sandbox (API) | No | Yes | Yes |
| Audit logs | Not published | Yes | No |
| SSO | No | No | No |
| SOC 2 | Yes | Yes | No |
| HIPAA | No | Yes | No |
| EU data residency | Yes | Yes | Yes |
| Zero data retention | Not published | Not published | Yes |
Performance
There is no comparable CPU measurement for Hyperstack or Cerebrium: neither is in the StarSling hpc-sandbox-benchmarks runs we use, so we cannot say which is faster. bochka did 26.69 Node.js build-and-test runs per second at 4 vCPU / 8 GB in the run of 2026-09-29, on dedicated AMD Ryzen 9 9950X hosts. Both providers' own speed claims are about GPUs, which this test does not cover.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Hyperstack and Cerebrium have no published run in this suite yet, so there is no like-for-like number to compare.
Pricing in five scenarios
Each bill is priced from the providers' public price lists by the open Battleships engine, with the cheapest plan that runs the scenario.
| Scenario | Hyperstack, a month | Cerebrium, a month | bochka, a month |
|---|---|---|---|
| One coding agent, workdays 4 vCPU / 8 GB, 50 GB disk, 8 hours a day, 22 days a month, files kept between days | $114 · Prepaid on-demand | $53 · Hobby · with limits | $20 · $20/mo |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $475 · Prepaid on-demand | $210 · Hobby · with limits | $26 · $20/mo |
| 50 agents in parallel 50 machines of 4 vCPU / 8 GB at once, 8-hour sessions, 1,100 sessions a month, Docker inside | $5,720 · Prepaid on-demand | $2,557 · Hobby · with limits | $317 · $20/mo × 3 seats |
| Sandboxes for your users 2 vCPU / 4 GB, 30,000 sessions of 10 minutes a month, up to 300 at once | $1,750 · Prepaid on-demand · with limits | $715 · Hobby | $200 · $100/mo × 2 seats |
| 1,000 short jobs at once 2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month | $3,500 · Prepaid on-demand · with limits | $1,681 · Standard | $2,000 · $2000/mo · with limits |
Hyperstack: Hyperstack bills a running VM per minute from a prepaid balance, at a fixed hourly price per flavor, with no charge for bandwidth. A stopped VM keeps billing in full, so the scenarios assume you hibernate it between runs.
Cerebrium: Cerebrium bills per second for each vCPU, GB of memory and GPU while a container runs. Hobby has no monthly fee; Standard is $100 a month on top of compute, and the protected tier costs 2x on all compute.
Cerebrium: You pick a monthly plan from $20, and all of it becomes usage credit spent per second on running sandboxes. A stopped sandbox costs nothing and keeps its whole disk.
Limits in those scenarios
Cerebrium does not fit the coding agent, always-on and 50-agent scenarios, because only /persistent-storage survives a stop and the fleet needs Docker inside, which is not documented. Hyperstack fits those, but does not publish how many VMs can run at once, so 300 user sandboxes and 1,000 short jobs at once are not confirmed there. Cerebrium fits both of those, on Hobby and Standard. bochka's start limits queue the 1,000-job burst on its $2,000 plan.
| Scenario | Hyperstack | Cerebrium | bochka |
|---|---|---|---|
| One coding agent, workdays | None that bite |
| None that bite |
| One agent that never stops | None that bite |
| None that bite |
| 50 agents in parallel |
|
| None that bite |
| Sandboxes for your users |
|
| None that bite |
| 1,000 short jobs at once |
|
|
|
Which one should you use?
Use Hyperstack when
you need a full VM with root and GPUs that runs as long as you want, large multi-GPU nodes for training, firewall rules, or regions in Canada or the US.
- A whole VM you control. Useful if you want root, SSH and Docker on a machine that runs as long as you like, with GPUs attached.
- Large GPU nodes and spot prices. Useful for training and fine-tuning: Hyperstack rents up to 8-GPU B300 nodes by the minute, with spot and reserved prices too.
- Firewall rules and Kubernetes. Useful if you must limit outbound traffic by IP and port, or run managed Kubernetes clusters next to your VMs.
Use Cerebrium when
you serve a model, voice or video app as an endpoint that scales with traffic and back to zero, billed per second.
- Endpoints that scale to zero. Useful for apps that wait for traffic: containers start on a request, bill per second and scale back to zero.
- Many containers at once. Useful for bursts: Hobby allows 500 CPU containers at once and Standard 1,000, while Hyperstack does not publish a limit.
- More regions on request. Useful if users are spread out: it runs in the US and EU, with Asia, South America and more on request.
Where bochka fits
Hyperstack and Cerebrium are GPU clouds, and if your work is training or serving models, one of them is the right pick. If what you are really running is agents that write and run code for hours, bochka gives each one a full VM built for that, with no GPUs.
A disk that lasts, and free stops
On Hyperstack a stopped VM keeps billing in full unless you hibernate it, and restoring needs the flavor in stock. On Cerebrium only /persistent-storage survives, so installed packages reset. bochka saves the whole disk every minute and stops for free.
Docker, /dev/kvm and fork
Cerebrium documents no Docker inside, and neither can fork a running machine; on Hyperstack you snapshot with the VM shut off and deploy a new one. bochka has Docker, systemd and /dev/kvm, and forks a running sandbox.
Agents, desktop and keys built in
Neither runs a coding agent inside or gives it a screen, and both hand keys to the machine as files or environment variables. bochka ships Claude Code, Codex and more, a 60fps desktop, and a credentials proxy for any HTTPS API.
Use bochka when your agents need a machine of their own for hours: a shell, root, Docker, /dev/kvm and a browser, on a full VM whose whole disk is saved every minute and costs nothing while stopped. Compare it one to one: bochka vs Hyperstack and bochka vs Cerebrium.
Questions people ask
Is Hyperstack cheaper than Cerebrium?
For CPU work in our scenarios, Cerebrium comes out lower, such as $210 against $475 a month for one agent that never stops, though Cerebrium does not fit that scenario because files reset. For GPU eval jobs Hyperstack is lower, at $625 against $1,975 a month.
Which is faster, Hyperstack or Cerebrium?
We have no comparable CPU measurement for either, so we cannot say. bochka did 26.69 runs per second on the StarSling CPU test at 4 vCPU / 8 GB.
Can Hyperstack and Cerebrium run Docker?
Hyperstack VMs are full virtual machines with root, and its Ubuntu images come with or without Docker. Cerebrium builds your app from a Dockerfile, but running Docker inside the app is not documented. bochka sandboxes run Docker, Docker Compose, systemd and /dev/kvm.
Do stopped machines cost money?
On Hyperstack a stopped VM bills at the full rate; a hibernated one bills only its disk and any kept IP. On Cerebrium a container that scales to zero stops billing, but only /persistent-storage keeps its files. On bochka a stopped sandbox costs nothing and keeps its whole disk.
Does bochka have GPUs?
No. bochka runs Linux VMs on dedicated AMD Ryzen 9 9950X hosts and is CPU only. For training or serving models, Hyperstack or Cerebrium is the right place for that part of the work.
Try bochka
Install the CLI, then run bochka new.
Product preview