Tenstorrent has officially released its highly anticipated Galaxy Blackhole AI compute systems. These cutting-edge systems boast 32 Blackhole accelerators within a compact 6U chassis, interconnected by 100 Tbps of Ethernet bandwidth. Each unit delivers impressive specifications: 1 TB of GDDR6 memory, 16 TB/s of memory bandwidth, and 23 petaFLOPS of FP8 performance, making it a formidable contender in the AI space. Notably, at $110,000, it costs significantly less than comparable systems. Drawing parallels with heavyweight competitors, the Galaxy system can be expanded beyond a single node, much like Google’s TPU or Amazon’s Trainium2 setups. Starting at $440,000, a Galaxy Supercluster integrates four Blackhole systems, scaling up to 144 nodes supporting over four thousand chips. Recent developments have enhanced the software stack, enabling smoother integration of models and better performance, a stark improvement from initial reviews. A remarkable showcase involves processing 100,000 token prompts in under four seconds, indicating high efficiency. Looking forward, Tenstorrent aims to expand its influence by partnering with major data center providers, with further updates expected at the upcoming TT-Deploy event.