Tenstorrent is a privately held semiconductor company that designs and sells AI accelerators and RISC-V processors. Founded in 2016 and headquartered in Toronto, Canada, the company focuses on developing hardware and software for Artificial intelligence workloads, including Deep learning and Large language model inference and training. Tenstorrent is recognized as a unicorn startup, with a valuation exceeding US$1 billion as of 2024.
The company was co-founded by Ljubisa Bajic, a former chip architect at AMD and Intel, and later joined by Jim Keller, a prominent microprocessor engineer known for his work at AMD, Apple, and Tesla. Keller became the chief executive officer in 2021, bringing significant industry attention and technical leadership.
History
Tenstorrent was established in 2016 by Ljubisa Bajic, who previously worked on AMD's Zen architecture and Intel's graphics processors. The company initially aimed to create a novel AI processor architecture that could scale across a wide range of devices, from edge to data center. In 2018, Tenstorrent introduced its first architecture, called Grayskull, which was designed for inference tasks. The Grayskull processor was released in 2020, targeting Machine learning inference in data centers.
In 2021, Jim Keller joined Tenstorrent as president and chief technology officer, and later became CEO. Under Keller's leadership, the company shifted its focus to include both AI accelerators and RISC-V-based CPUs, aiming to provide a complete solution for AI computing. In 2022, Tenstorrent announced its second-generation architecture, Wormhole, which added support for training and larger models. The company also began offering development boards and IP licensing, making its technology accessible to a broader range of customers.
Technology and Products
Tenstorrent's core technology is a mesh-based architecture that connects multiple processor cores on a single chip, enabling efficient scaling of compute resources. The company's AI accelerators are designed to handle both inference and training for Neural network models, including Transformer (architecture)-based architectures used in Generative AI applications. The processors use a custom instruction set and a software stack that includes a compiler and runtime, supporting popular frameworks like PyTorch and TensorFlow.
Key products include the Grayskull and Wormhole accelerators, as well as the Blackhole series, which was announced in 2023. Tenstorrent also develops RISC-V CPU cores, such as the Ascalon, which are integrated into its AI systems. The company offers a range of form factors, from PCIe cards to full server systems, and provides IP licensing for customers who wish to integrate its cores into their own chips.
Market Position and Partnerships
Tenstorrent competes in the AI chip market with companies like NVIDIA (not in list, but implied), Cerebras, Groq, and SambaNova. Unlike many competitors that focus solely on accelerators, Tenstorrent differentiates by offering open-source RISC-V cores and a flexible software stack. The company has partnered with TSMC for manufacturing its chips, using advanced process nodes. In 2023, Tenstorrent announced a collaboration with Samsung to develop AI solutions for edge devices, and it has also worked with LG (not in list) on automotive applications.
As of 2024, Tenstorrent has raised over $200 million in funding from investors including Samsung, Fidelity, and Eclipse Ventures. The company's valuation reached $1 billion in 2023, making it a unicorn. Tenstorrent's revenue model includes hardware sales, IP licensing, and software subscriptions, with a focus on data center and edge AI markets.
Future Directions
Tenstorrent aims to expand its presence in the AI accelerator market by offering cost-effective and energy-efficient solutions. The company is working on next-generation architectures that integrate AI accelerators with RISC-V CPUs on a single chip, enabling more efficient processing of Large language model workloads. Tenstorrent also plans to grow its software ecosystem, making it easier for developers to deploy AI models on its hardware. With the increasing demand for AI computing, Tenstorrent is positioned to be a significant player in the industry, though it faces intense competition from established companies and other startups.