Tenstorrent prices on-prem voice AI at $160,000

Tenstorrent and Smallest.ai partnered to run Lightning V2 voice inference on-premises, with Tenstorrent’s Galaxy Blackhole server starting at $160,000. Smallest.ai’s technical comparison claims a 4× lower accelerator cost than an NVIDIA L40S for a modeled 550 simultaneous five-second requests, but that is a vendor benchmark, not a fleet guarantee. Voice builders should compare that capex with usage-priced deployment and reproduce concurrency, latency, and audio-quality results before treating on-prem as cheaper; the archive’s voice-agent billing analysis sets the adjacent cost boundary.