Runcrate (@runcrate) header
Runcrate (@runcrate) profile picture

Train, deploy and scale AI.

GB300
Joined July 2025
Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)Photo by Runcrate (@runcrate)
GPUs and clusters don't have to break your bank. Spawn the cheapest on-demand instances or request a cluster. Choose Runcrate today.

1
12
43,874
We got loads of GB300 racks, double digit B300 nodes, but nobody to lock them yet 😞😪 ask prices in dm.
1
1
1
171
A GB300 NVL72 rack is open on Runcrate: 72 GPUs in Malaysia, ready around 25 Nov, 3 month minimum, paid upfront. For teams that train or serve models. DM for price.
1
195
or just reply here, or mail [email protected]
156
AMD MI300X (192 GB) is live on Runcrate, from $4.85 per GPU-hour on one 8-GPU box. 8 hour minimum. Launch it from the dashboard, or reply here if your team wants it.
1
2
202
PageLM has Runcrate as an AI provider on main now, so you can run it on open models. Thanks to recabasic for merging it.
145
Thanks @thehecticbyte for merging the Runcrate section in pr-agent. The docs in v0.50.0 now show how to point PR-Agent at an OpenAI-compatible Runcrate endpoint, so reviews can run on open models.
129
AMD MI300X (192 GB) is coming to Runcrate on demand. If your team wants it, reply or DM us how many GPUs and for how long.
2
96
Thanks to the ClaudeBar team for merging Runcrate. v0.5.26 shows your Runcrate credit balance in the macOS menu bar next to your other AI usage. It is off by default: turn it on in Settings and add your Runcrate API key. github.com/tddworks/ClaudeBa…
83
Thanks to the tgpt maintainers for merging Runcrate. v2.16.0 lists it under new providers, an OpenAI-compatible provider with tool calling, so you can chat with open models from your terminal using your Runcrate API key. github.com/aandrew-me/tgpt/r…
87
writing evals is boring and it is the only thing that tells you whether the cheaper model is good enough. do it before the migration, not after
123
Thanks @axolotl_ai for merging our Runcrate guide. Axolotl's docs now have a Runcrate page and the README lists it with the other clouds, so you can fine-tune on rented GPUs.
55
open models rarely fail on quality now. they fail on boring things: context limits, rate limits, a node that disappears mid-run. design for that first
98