In mild of Donald Trump’s comment about Taiwan, Canadian AI device, software, and hardware architecture firm Tenstorrent stands out as a viable alternative to TSMC and works with Circuit factories in South Korea and Japan.
In an interview with Bloomberg Businessweek published on July 16, Trump said,” I know the people]of Taiwan ] very well, respect them greatly. They did get about 100 % of our device company. I think Taiwan may compensate us for defence…. Taiwan does n’t give us anything”.
” Taiwan took our device business from us”, he repeated,” I mean, how ridiculous are we? They took all of our device business. They’re exceedingly wealthy. And I do n’t think we’re any different from an insurance policy. Why? Why are we doing this”?
By the week’s close, TSMC and Nvidia’s share costs had both fallen more than 8 %.
Tenstorrent is never publicly traded. Unlike Nvidia, it does not count on TSMC. Founded in 2016, Tenstorrent is headquartered in Toronto with offices in Santa Clara and three other locations in the US plus Bengaluru ( Bangalore ), Beijing, Gyeonggi-do ( Seoul Capital Area ), Tokyo and Beograd ( Belgrade ).
The company now has about 460 employees and is growing rapidly, with 45 available positions – CPU designer, CAD expert, systems program expert, account executive, etc. – listed on its site.
CEO Jim Keller, who joined the company in 2020, is a famous Circuit architect who built his standing at AMD, Apple, Tesla and other important businesses. His approach is to create Artificial options that are much more cost-effective and affordable than those from Nvidia.
Consumers “need the most cost-effective cars that run as quickly,” according to his partner, Chief CPU Architect Wei-han Lien, who also worked for Tenstorrent through AMD and Apple.
Tenstorrent models AI Graph Computers, RISC-V CPUs, and customizable chiplets to manage its software to build laptops for artificial intelligence. They can be cheaper than Nvidia’s because they are designed for lower price developing, make extensive use of open-source RISC-V systems, and do not need expensive high-bandwidth storage.
Wei-han Lien,” Some multinational firms have come to beg us to design cost-effective choices for them because they find popular products to be too cheap when they start running inferences.”
Last February, when OpenAI CEO Sam Altman said an outrageous$ 5 trillion to$ 7 trillion, or even$ 8 trillion, would be needed in order to make enough AI chips to meet anticipated demand, Jim Keller wrote on X that he could do it for$ 1 trillion.
In addition, Tenstorrent models AI machines, PCIe ( peripheral component interconnect show ) planks, and desktops for AI type builders, which it also provides with cloud computing services. The company claims that the Galaxy AI compute platform, which costs a third less than Nvidia’s DGX platform, is three times more cost-effective for businesses and research institutions.
The company’s TT-Buda and TT-Metalium high- and low-level software development kits, which run AI models on Tenstorrent hardware and enable custom development, compete with Nvidia’s CUDA software.
Tenstorrent has received funding from several American institutional investors, including Fidelity, Eclipse Ventures, Real Ventures, Maverick, and Moore Capital, as well as financing from the Samsung Catalyst Fund and the Hyundai Motor Group of South Korea.
In October, 2023, the company announced that it had selected Samsung Foundry to fabricate its RISC-V CPU and AI acceleration chiplets using 4nm process technology. Shipments of 4nm devices from Samsung’s fab in Taylor, Texas, are expected to begin this year. It was reported earlier this month that TSMC is preparing to use its 4nm process to produce Nvidia’s new Blackwell GPU in Taiwan.
CEO Jim Keller met with executives and engineers at Hyundai Motor’s Asan factory in South Korea in February of this year to talk about developing AI for Hyundai’s autonomous vehicles, robots, and air-mobility products ( flying cars ) and schedule meetings.
Keith Witek, the COO of Tenstorrent, was elected to the board of directors of Hyundai Mobis, the auto parts manufacturer for the Hyundai Motor Group, in March. With both hardware and software capabilities, Hyundai Mobis is leading the charge toward connected and autonomous driving. Witek previously held senior positions at AMD, Tesla, and Google.
Tenstorrent is also working with LG Electronics in South Korea on RISC-V AI technology being used in consumer electronics, consumer electronics, and automotive electronics.
An agreement to jointly develop AI semiconductor design capabilities was signed by Tenstorrent and Rapidus, Japan’s advanced logic IC foundry. Rapidus, which was established in 2022, intends to begin mass production in 2027 at its new factory in Hokkaido and pilot production at the 2nm node in 2025.
Rapidus is backed by Sony, Toyota and Toyota’s auto parts affiliate Denso, national telecom carrier NTT, telecom equipment maker NEC, investment firm Softbank and Mitsubishi UFJ, Japan’s largest bank.  , Its process technology, which will start at 2nm and advance from there, is the result of a partnership with IBM. Its objectives are to revive Japan’s semiconductor sector and lessen its dependence on Taiwan.
According to David Bennett, the company’s chief customer officer,” Japan is very important to both Tenstorrent and me personally.” Bennett was the CEO of NEC Personal Computers and the president of Lenovo Japan before joining Tenstorrent in 2022. He is a member of the board of Sanrio, the Hello Kitty company, and serves as a visiting lecturer at Yamagata University.
In February 2024, Japan’s Leading-edge Semiconductor Technology Center ( LSTC ) announced a project commissioned by NEDO ( Japan’s New Energy and Industrial Technology Development Organization ) to develop advanced semiconductor manufacturing technology for post-5G telecom infrastructure. LSTC is affiliated with Rapidus.
The project, which is called” Development of edge-AI accelerator implemented in 2-nm logic technology,” combines Tenstorrent’s RISC-V CPU and chiplet design with an accelerator chip created by Japan’s AI Chip Design Center, a collaborative research project between the National Institute of Advanced Industrial Science and Technology and the University of Tokyo. Rapidus is creating a Rapid and Unified Manufacturing Service, which it calls, to quickly produce the new AI devices.
In order to support this and other projects, Tenstorrent intends to establish a high-performance computing design center in Japan, and LSTC intends to train some 200 engineers there over the next five years.
This will lower the cost of Tenstorrent’s expansion and strengthen its ties with Japan, thanks to financial support from Japan’s Ministry of Economy, Trade and Industry. Many of the engineers are expected to return home after a year or two of service with Tenstorrent to address Japan’s shortage of IC design talent.
Follow this writer on , X: @ScottFo83517667