Monday,7 August 2023, Bengaluru, India
By 2025, AI chips will account for up to 20% of the $450 billion global semiconductor market. In addition, sales of AI chips will rise 35% yearly from $5.7 billion in 2018 to $83.3 billion in 2027, according to The Insight Partners. That is nearly ten times the anticipated growth rate for non-AI chips.
As an illustration, Tenstorrent, the AI hardware firm led by engineering legend Jim Keller, reported this week that it had raised $100 million in a financing round sponsored by Hyundai Motor Group and Samsung Catalyst Fund in the form of convertible notes.
$50 million of the total comes from Hyundai’s two auto manufacturing subsidiaries, Hyundai Motor ($30 million) and Kia ($20 million), who intend to collaborate with Tenstorrent to build chips, notably CPUs and AI co-processors, for upcoming mobility vehicles and robots. The remaining $50 million was provided by Samsung Catalyst and various VC firms, including Fidelity Ventures, Eclipse Ventures, Epiq Capital, and Maverick Capital.
In contrast to equity, a convertible note is a short-term loan that becomes equity in the event of a specific occurrence. The reason Tenstorrent chose debt over equity and the company’s post-money valuation must be clarified. (In a release, Tenstorrent referred to it as an “up-round”). When Tenstorrent last raised $200 million, it was valued at almost $2 billion.
Tenstorrent’s total funding now stands at $334.5 million thanks to the convertible note tranche, which included investors such as Fidelity Ventures, Eclipse Ventures, Epiq Capital, Maverick Capital, and more. According to Keller, it will go toward product development, the creation and advancement of AI chipsets, and Tenstorrent’s machine learning software roadmap.
Toronto-based Tenstorrent offers RISC-V, the open-source instruction set architecture used to create bespoke processors for various uses by selling AI processors, AI software solutions, and intellectual property (IP).
Ivan Hamer, a former embedded engineer at AMD; Ljubisa Bajic, a former director of integrated circuit design at AMD. Milos Trajkovic, a former AMD firmware design engineer, founded Tenstorrent in 2016. From the start, the company focused heavily on building its internal infrastructure. Grayskull, an all-in-one solution powered by Tenstorrent’s exclusive Tensix cores, was unveiled in 2020 to accelerate the training of AI models in data centers, public and private clouds, on-premises servers, and edge servers.
But in the following years, Tenstorrent changed its focus to licensing and services, possibly feeling the pressure from market leaders like Nvidia, and Bajic, who had been in charge, gradually shifted to an advising role.
To train massive generative AI systems like PaLM-2 and Imagen, Google developed a processor known as the TPU (short for “tensor processing unit”). Customers of AWS can use Amazon’s chips for training (Trainium) and inferencing (Inferentia). And according to reports, Microsoft is collaborating with AMD to create the Athena internal AI chip.
Meanwhile, Nvidia, riding high on demand for its GPUs for AI training, momentarily reached the $1 trillion business milestone this year. (As of Q2 2022, Nvidia still held an 80% market share for discrete GPUs.) While only sometimes as powerful as specially created AI chips, GPUs can carry out several computations in parallel, which makes them ideal for training the most advanced models available today.
Unsurprisingly, it has been difficult for startups and established software companies. The “extremely challenging” financial environment, according to AI chipmaker Graphcore, which reportedly had its valuation dropped by $1 billion when a deal with Microsoft fell through, prompted the company to announce job losses last year. In the meantime, 10% of the employees at Intel-owned AI chip maker Habana Labs were let go.
[Source of Information: Techcrunch.com]
