Tenstorrent Sets New Performance Records, Launches TT- Ascalon S, and Expands Across Japan

At TT-Deploy JP, Tenstorrent set new records on language and video models, launched TT-Ascalon S RISC-V CPU IP for agentic AI, and joins ai&'s sovereign heterogenous inference platform with Tenstorrent Galaxy™ superclusters, a general-purpose system that can drop in beside GPUs or stands alone.

TOKYO, JP / ACCESS Newswire / June 30, 2026 / Tenstorrent, the AI compute company led by CEO Jim Keller, today at TT-Deploy JP set new performance records across language and video, launched TT-Ascalon S RISC-V CPU IP for agentic AI, and detailed its largest deployment to date, a general-purpose, heterogenous AI build in Japan. Each rests on the same foundation: a single architecture that runs major AI workloads faster than GPUs and scales from a licensable core to a Tenstorrent Galaxy^™ supercluster over standard Ethernet.

That makes Tenstorrent's Networked AI architecture a different kind of solution - open, general-purpose, flexible for heterogeneous or stand alone deployments, and backed by AI experts - that can withstand the constant change in the AI industry.

New industry records from language to video

Continuing to build on previous performance, Tenstorrent shared new LLM and video with lip-sync and audio benchmarks. On the latest models enterprises are deploying right now, Tenstorrent Galaxy Blackhole superclusters post:

• Kimi K2.6: 900 tokens/second/user, 3x faster than GPUs

• DeepSeek-R1-0528 671B: 400+ tokens/second/user, up from 350+ at TT-Deploy SF

• LTX 2.3 Fast: roughly 6-second video generation at 144 frames, 1080p, with audio and lip-sync, 4x faster than GPUs

Different model families, one architecture, with capacity that grows near-linearly as Galaxies are added. Tenstorrent's performance enables enterprises to scale premium inference workloads efficiently.

TT-Ascalon S: Tenstorrent expands the TT-Ascalon portfolio with TT-Ascalon S suited to emerging agentic AI workloads

Launching today at TT-Deploy JP, Tenstorrent announced TT-Ascalon S, a compute-dense RISC-V CPU for agentic AI. Agentic AI leans on the CPU in a new way, gated less by raw compute than by orchestration, I/O, and latency, and TT-Ascalon S is built for it:

• Density: Built on the foundation of TT-Ascalon X, TT-Ascalon S is purpose built for a footprint ~50% the size, delivering ~140% performance per mm².

• Efficiency: A compact, power-efficient design for high-throughput execution layers.

• Latency: Ascalon S is tuned for the mixed, branch-heavy, tool-connected execution patterns typical of agent runtimes, helping enable more predictable execution.

In addition to agentic AI, TT-Ascalon S is applicable to high-efficiency servers, networking and storage SoCs, and data-center edge deployments. As licensable RISC-V IP, it enables customers to extend the same architectural foundation into custom silicon designs.

Built to drop in, or stand alone

Connecting the accelerators and the CPU alike is Tenstorrent's Networked AI. The architecture unifies compute, memory, and networking over standard Ethernet, with an open-source software stack. Tenstorrent Galaxies and superclusters stand alone or drop into an existing GPU fleet without replacing existing infrastructure. Customers can add capacity without betting on a single model, workload, or vendor; the systems keep performing as models change; and the infrastructure stays in the customer's hands. For the enterprises and nations building next-gen private AI infrastructure, that independence and flexibility is key.

"We built one architecture that runs everything, drops in next to whatever you already own, and scales from a licensable core to a supercluster. That's what lets a company, or a country, own its AI and adapt to the only constant in AI - change. We're excited to partner so widely in Japan with key partners like ai&, Rapidus, Preferred Networks, Socionext and Turing." said Jim Keller, CEO of Tenstorrent.

Running across Japan, at national scale

Across Japan, Tenstorrent enables cloud service providers, data centers, and private infrastructure with everything from licensable IP to large-scale supercluster deployments.

Its largest deployment to date runs with ai&, Japan's vertically integrated frontier AI platform. ai& launched its heterogeneous inference platform to expand sovereign AI in Japan. Tenstorrent Galaxies and superclusters support key inference and agentic workloads, and enables customer use cases such as: chat, RAG, vision, and post training, served entirely from within the country. They've deployed more than 120 Tenstorrent Galaxy systems representing the largest scale of sovereign AI compute in the region.

"ai& is built on the conviction that the best inference infrastructure routes every workload to the silicon best suited to it. Tenstorrent's Galaxy superclusters have proven themselves on exactly the workloads our enterprise customers care about, and this deployment represents the largest sovereign AI compute footprint in Japan. We're excited for what the partnership unlocks," said David Bennett, CEO & co-founder at ai&

That reach rests on a deep base in the country. Turing, the Tokyo-based autonomous-driving company, has demonstrated Tenstorrent Blackhole running inside an autonomous vehicle as a proof-of-concept. And through the national 2nm program with Rapidus, adopted by NEDO and led by LSTC, Tenstorrent contributes the RISC-V CPU chiplet. Tenstorrent Japan has operated in Tokyo since 2023, runs an AI data center in Osaka today, and is bringing up to 200 Japanese silicon engineers into its design teams.

TT-Deploy JP brings the ecosystem on stage, with partners including ai&, Rapidus, Preferred Networks, Socionext and Turing, plus live demos from ai&, BOS and Turing.

About Tenstorrent

Tenstorrent is an AI compute company led by CEO Jim Keller - architect of Apple A4/A5, AMD Zen, and Tesla's Full Self-Driving chip. The company builds RISC-V-based AI processors and systems for developers, enterprises, and sovereign infrastructure worldwide. In addition to servers and workstations, Tenstorrent licenses its TT-Ascalon RISC-V CPU and Tensix AI cores to chip designers including Samsung and LG. Backed by Bezos Expeditions, Samsung, LG Electronics, Hyundai Motor Group, Fidelity, and others, Tenstorrent has raised over $1B+ and operates from Santa Clara, Austin, Toronto, Belgrade, Tokyo, and Bangalore. Learn more at tenstorrent.com.

Media Contact

Justin Mauldin
Salient PR
achievemore@salientpr.com
737.234.0936

SOURCE: Tenstorrent