While AMD, Cerebras and Intel are ambitiously working towards taking on chip leader NVIDIA in their own ways, Tenstorrent has emerged as the brand new player to join the opposing force. And they are doing it by partnering with automobile and electronics companies.
Canada-headquartered Tenstorrent, a computing company that develops processors and AI-based deep learning processing units, received a $100 million investment from Hyundai and Samsung. Prior to this funding, Tenstorrent had secured $234.5 million in investments with a valuation of $1 billion, and stands as one of the many companies aiming to contest NVIDIA. It secured $30 million from Hyundai, $20 million from Kia, and the remaining $50 million originated from Samsung’s Catalyst Fund along with additional contributions from investors such as Fidelity Ventures, Eclipse Ventures, Epiq Capital, and Maverick Capital, among various others.
Tenstorrent is led by Jim Keller, a well-known figure in the semiconductor realm. Playing a pivotal role as a lead architect for AMD K8 microarchitecture and also participating in the design of processors such as Athlon and Apple A4/A5, Keller’s experience and vision puts him in the forefront for building something that might give NVIDIA a run for the money.
Keller said that for individuals seeking to construct a high-performance solution integrated with AI, NVIDIA will dominate a significant portion (60%) of the product’s profit margin. “The problem with the winner-take-all strategy is it generates an economic environment where people really want an alternative.” Tenstorrent not only manufactures its own AI chips but also offers its intellectual property and other technologies to clients interested in creating their own AI chips.
The Golden Automotive Industry
In 2021, the worldwide market size for automotive chips amounted to $49.8 billion, with a projected growth of $121.3 billion by 2031. This expansion is expected to occur at a Compound Annual Growth Rate (CAGR) of 9.6% from 2022 to 2031. With a positive market on the horizon, chipmakers can reap benefits from the same.
Hyundai formed a semiconductor development division in the previous year and announced intentions to integrate Tenstorrent’s technology into forthcoming vehicles under the Hyundai, Kia, and Genesis brands. With this investment, the parties look to develop “optimised but differentiated semiconductor technology” which will aid future AI technology development.
Being one of the globe’s major semiconductor contract manufacturers, Samsung’s decision to invest is understandable. The company has said that the funding will be directed towards expediting the company’s product development, advancing the design and creation of AI chiplets, and enhancing its roadmap for ML software.
In May, Tenstorrent collaborated with LG Electronics. The partnership aims to develop chips to fuel consumer electronics such as TVs, automotive solutions and data centres. LG will initially adopt Tenstorrent’s AI chip blueprint for its own chip design and Tenstorrent said that they would look at some of the technology that LG has developed.
A Long Way before the Catch Up
Racing through the computing world, NVIDIA has also branched out to cater to the automobile industry. In collaboration with NVIDIA, Mercedes Benz will work on creating intelligent cockpits and architectures to accommodate AI-driven capabilities for driving. The company had also partnered with other automobile giants such as Jaguar Land Rover and Volvo.
NVIDIA, is currently the undisputed leader with an 80 to 95% of market share in the AI computing market. The company even touched a market capitalization of $1 trillion and is continuing to surpass its competitors.
Recently, NVIDIA’s leading H100 chip achieved its highest performance to date on a series of MLPerf training benchmarks. (MLPerf benchmarks assess the hardware capabilities of chips by measuring the time required to complete specific workloads). Partnering with CoreWeave and Inflection AI, the GPU established new records across various parameters in a recent test. The trial employed a cluster of 3,584 H100 GPUs hosted on CoreWeave’s platform, interconnected using InfiniBand technology, enabling exceptional performance at both individual and scalable levels.
Tenstorrent and Keller’s ambitious plan to take on NVIDIA may be far-fetched, but setting a path through partnerships with strategic players may get them close to its competitor. However, NVIDIA’s smooth trajectory route puts them in another league altogether.