Clockwork: Keep GPU jobs running through failures
Clockwork is a software layer that keeps AI training and inference jobs running through hardware and network faults on GPU clusters.
What is Clockwork?
Clockwork is software for large GPU clusters. It reroutes traffic when a network link fails, moves work from a failed GPU or node to a replacement, captures training state quickly, and checks links, NICs and GPUs before a job starts. Its monitoring tool points to the faulty link, switch or NIC when jobs slow down. It is used by cloud providers, platform teams and model builders running training or inference at scale. It is software only and works with frameworks such as TorchTitan, Megatron-LM and DeepSpeed.
Who is Clockwork for?
- GPU cloud providers
- Platform teams
- Model builders
Clockwork funding
- Funding round: $31M, announced 5 October 2026. Source
Data: LaunchRounds, verified funding rounds. Each round is backed by the linked announcement, filing or report. Data as of 17 November 2026. Funding says nothing about whether the tool suits you.
Is Clockwork worth it?
Useful for teams losing time to cluster failures; not relevant for small setups.
Where Clockwork is strong
- Works without model code changes at platform level
- Covers both failure recovery and diagnosis
Where it falls short
- Pricing not published
- Aimed at large clusters
How do you get started with Clockwork?
- 01Book the free consultation
- 02Run a failure test on your cluster
- 03Install the plugin or patches
Clockwork: frequently asked questions
How much does Clockwork cost?
Pricing is not stated on the site.
Does it need changes to model code?
Platform-level protection needs no model changes; some features use light code integration or patches.
Clockwork alternatives
The closest tools in our directory, matched on category, focus areas and audience.
| Tool | Best for | Free plan | Starting price | Focus areas |
|---|---|---|---|---|
| Clockwork | Operators of large GPU clusters | No | — | — |
| NBULA | MENA enterprises and agencies needing Arabic-first, in-region AI deployment | No | — | — |
| Applied Compute | Enterprises building their own models | No | — | — |
| Unify | Production agent teams wanting models that improve from real usage | No | — | — |
| Imagga | Developers adding image tagging, search or moderation to apps | Yes | $79/mo | — |
| BentoML | Teams deploying and scaling model inference on their own infrastructure | Yes | — | — |
| Kubeflow | Kubernetes teams running ML training, pipelines and notebooks | Yes | — | — |
| vLLM | Self-hosting open-source LLMs with high throughput and low latency | Yes | — | — |
| Atero AI | AI teams seeking better GPU memory efficiency | Unknown | — | — |
| Heimdall | Teams turning tables or text into a callable prediction API | Unknown | — | — |
| UnionAI | Teams running AI, ML and data pipelines inside their own cloud | Unknown | — | — |
