Tenstorrent Galaxy
Rack-mount AI compute server built from 32 Tenstorrent Tensix processors with built-in Ethernet scale-out, sold in Blackhole and Wormhole generations and in multi-server supercluster configurations.
Find alternatives to Tenstorrent Galaxy
What it does
- Server systems
- Model inference
Used for
- Video generation
Features 6
- SuperclustersGalaxy Blackhole comes in a supercluster of four, extending into multi-server clusters.
- Fast video generationRuns state-of-the-art video models, generating 720p, 81-frame video in 2.4 seconds.
- Large-context LLM inferenceOptimized for latency-sensitive workloads with the same systems used for decode and prefill.
- Broad model coverage90% of Hugging Face models run, with growing coverage across LLMs, image, speech and vision.
- Simple ScaleThe Tensix Neo architecture scales from one chip to thousands under one programming model.
- Open source stack end to endFull software stack from compiler to kernel is open source.
Sources
- Pricing
-
Page title 'Tenstorrent Galaxy Scalable AI-Compute Server'. This figure supersedes the stale $440,000 four-system supercluster figure carried on the tenstorrent-blackhole record.
- Used for
-
Official documentation · 4 Nov 2026
'are optimized for workloads including AI video generation, large-scale LLM inference, and private AI infrastructure' (https://tenstorrent.com/en/hardware/galaxy)
- Sold within
-
Official documentation · 6 Sep 2026
Own catalogue page under /en/hardware with its own list prices and its own Contact Sales path; not bundled inside any larger Tenstorrent offering.
- URL
-
Official documentation · 6 Sep 2026
Self-canonical product page, distinct from the /en/hardware/cards catalogue page the Blackhole record points at.
- Intended for
-
only purchase path is sales-led — "Deploy"/"Create Configuration"/"Contact Sales" CTAs all route to /contact/enterprise, no self-serve buy
Also from Tenstorrent
-
Blackhole Hardware
RISC-V-based AI accelerator sold as standalone PCIe developer cards (Blackhole p100a, p150a, p150b) alongside the previous-generation Wormhole boards on Tenstorrent's card catalogue page.
-
TT-QuietBox Hardware
Liquid-cooled AI workstation sold in three generations: TT-QuietBox 2 (two Blackhole p300c), TT-QuietBox (four Blackhole p150c) and a Wormhole scale-out version (four Wormhole n300s), each with an open-source software stack for running and developing AI models locally.
-
TT-LoudBox Hardware
Air-cooled desktop AI workstation built around four Tenstorrent Wormhole n300s cards (eight Tensix processors) with a 96GB expandable memory pool, sold for local AI model development and HPC library porting.
-
Tenstorrent Cloud Infrastructure service
Hosted remote access to Tenstorrent accelerator instances of 1-128 Wormhole Tensix processors, reached through a browser or remote session via the TT-Console portal.
-
TT-Ascalon Hardware
High-performance RISC-V CPU core licensed as silicon IP to chip designers for use in AI accelerators, chiplets and custom SoCs.
-
Tensix Neo Hardware
AI accelerator core licensed as silicon IP to chip designers for embedding Tenstorrent's Tensix compute into their own SoCs and chiplets.
-
TT Studio Application
TT Studio provides a graphical interface for deploying and chatting with AI models on Tenstorrent hardware; also generates images and video entirely on-device.
-
tt-local-generator Application
tt-local-generator generates video, images, and animated characters locally on Tenstorrent hardware.
-
Wormhole Hardware
Wormhole is a family of Tensix Processor PCIe add-in boards for AI computing.
Open Tenstorrent in the directory
Something wrong here? Send a correction — quote this product id: tenstorrent-galaxy.