What changed, and when
Launches, hardware, prices and forge updates. Price changes are announced here and in the heat ledger a month before they apply.
Blackforge is taking work orders. Forge F1 on the Reykjanes peninsula runs sixteen NVIDIA B200 GPUs across two HGX nodes, cooled by outside air and powered by geothermal and hydro.
Access is by invitation while F1 is in preview. Send a work order with your workload, GPU count and hours, and a person replies within one business day with start dates and a quote.
Each node is an HGX B200: eight Blackwell GPUs with 180 GB of HBM3e each on one NVLink 5 fabric at 1.8 TB/s per GPU, two Xeon Platinum 8570 CPUs, 2 TB of memory and 30 TB of local NVMe.
The host runs NVIDIA driver 580 and CUDA 13.0. Bring any image built with CUDA 12.8 or later, the first release that targets Blackwell.
Strike at 3.79 USD per GPU hour, billed to the second. Hour bank at 3.09, prepaid in blocks of 1,000 hours. Anvil, a whole node for a month, at 16,640 USD, which is 2.89 per GPU hour. Ember at 1.39 for preemptible work on idle GPUs.
CPU, memory, scratch and traffic inside the forge are included, with 2 TB of shared storage and 5 TB of egress a month. Prices change at most once a month, announced a month ahead in the heat ledger.
Twenty four pages covering the quickstart, jobs, storage, interactive sessions, every plan, scheduling, telemetry, the full API and the forge CLI.
The API answers 503 until your key is activated; the request and response shapes in the docs are final.
F2 will be the same B200 build, beside a hydro dam on the St. Lawrence. It is listed when its nodes pass a 96 hour burn in, not on a date.
Teams that need capacity in North America can say so in a work order and we will keep you posted as F2 moves.
That is everything so far. Want capacity on F1?
Write a work order