ARM's Value Multiplied 5 Times by Son Jeong-ui's Magic... Will It Work in AI? [Tech Talk]

SoftBank Challenges AI Business Investment This Time
Foresight That Increased ARM's Value Fivefold in the Past
Interest in Whether It Can Succeed in the AI Chip Sector

Masayoshi Son, chairman of Japan's SoftBank Group, is once again embarking on a "once-in-a-lifetime challenge." He plans to mobilize a staggering $100 billion (approximately 138 trillion won) to invest in AI ventures. This initiative, known as the "Izanagi Project," aims to develop the ultimate AI.

Son's investment portfolio has always been eventful. Just a few years ago, he faced a critical crisis, even having to mortgage a U.S. mansion due to financial difficulties. The reason he has always been able to rebound like a "phoenix" is that, no matter how many trillions of won in losses he suffered, his most core investments succeeded. ARM is a prime example.

Masayoshi Son Quintupled ARM's Valuation

Son Jeong-ui, Chairman of SoftBank, speaking at the regular general meeting of shareholders
Photo by AFP Yonhap News

In 2016, SoftBank acquired the UK-based ARM Holdings for $32 billion (about 44 trillion won). Eight years later, ARM is now listed on the U.S. Nasdaq, with a market capitalization ranging between $160 billion and $180 billion (approximately 220 to 250 trillion won). SoftBank still holds 90% of ARM's shares. This means they succeeded in a "jackpot" by increasing the company's value fivefold.

Riding on ARM's success, Son has embarked on another once-in-a-lifetime bet. Recently, SoftBank acquired the UK AI semiconductor company Graphcore. Although the exact acquisition price has not been disclosed, various media sources estimate it to be between $400 million and $600 million (approximately 550 billion to 830 billion won).

This company once attracted $700 million in investment and was considered a rival to Nvidia. Its valuation exceeded $2 billion (about 2.772 trillion won). However, the AI semiconductor market is effectively monopolized by Nvidia, and most AI semiconductor startups, including Graphcore, are experiencing severe financial difficulties. Son essentially swallowed the company and its intellectual property (IP) at a bargain price when Graphcore was at its most vulnerable moment.

Core to Artificial General Intelligence, Acquisition of AI Semiconductor Company

On the 12th (local time), SoftBank officially announced the acquisition of Graphcore. [Image source=Captured from Graphcore homepage]

So why Graphcore? SoftBank could have invested in the already successful Nvidia or other GPU manufacturers. SoftBank has not disclosed the specific reasons for acquiring Graphcore.

However, SoftBank stated in a press release, "Next-generation semiconductors and computing systems are essential on the journey to Artificial General Intelligence (AGI)," adding, "Graphcore is a leading employer in the tech economy and will continue to invest in creating highly skilled jobs across various fields."

Graphcore develops Intelligence Processing Units (IPUs), which are considered rivals to Nvidia's GPUs. IPUs are known for their design optimized for AI training and inference. They arrange over 1,000 cores on a chip, each equipped with SRAM memory, increasing total memory capacity while enhancing parallel computing capabilities. Both features are essential for AI operation.

However, at present, GPU rivals including IPUs have failed to attract significant interest from big tech companies. Why is that? Nigel Toon, co-founder and CEO of Graphcore, cited "capital issues" in interviews with foreign media. In other words, developing advanced semiconductors requires enormous initial capital, and the roughly $700 million Graphcore raised was insufficient.

In fact, Graphcore spent over $200 million (about 270 billion won) annually on employee salaries and research and development (R&D). Yet, the practical gap with Nvidia continued to widen. Nvidia not only spends trillions of won annually on its own R&D but also has the capital to acquire key technology companies outright?an option unavailable to startups like Graphcore.

Can Masayoshi Son's Magic Happen Again?

SoftBank has successfully completed the initial public offering (IPO) of ARM Holdings.

SoftBank's acquisition of Graphcore likely values the "team" that Graphcore has built over the potential of the IPU product itself. The company's co-founders, Nigel Toon and Simon Knowles, have been designing AI chips since 2011, around the same time Nvidia CEO Jensen Huang decided to apply GPU functions to scientific fields such as physics and simulation.

Since its founding, Graphcore has built an elite chip design team of about 400 people, whose accumulated know-how and expertise are likely among the best in startups. If given sufficient "ammunition," they might create something that could cause a seismic shift in Nvidia's current dominance.

In fact, SoftBank used a similar strategy when acquiring ARM. At that time, ARM, listed on the London Stock Exchange, was already the best company in mobile chip design but struggled to secure R&D funds to expand into other fields.

Thanks to joining SoftBank at a timely moment, ARM was able to focus on technology investment despite temporary losses and has since reorganized into a high-growth company covering AI, server chips, automotive chips, and more.

Ultimately, Son and SoftBank's survival secret has been to accurately select companies with high potential and support them materially and financially until they reach their true value. Will "Masayoshi Son's magic" shine in the AI chip business as well?

This content was produced with the assistance of AI translation services.