Masayoshi Son's AI Ambition: Arm, which has tripled in value in a year, aims directly at NVIDIA?

What does the future AI landscape look like?

Recently, SoftBank founder Masayoshi Son stated in an interview that by 2035, ASI (Artificial Super Intelligence) will be achieved, which will be 10,000 times smarter than the human brain, but it will require at least 200 million chips and $900 billion to build! NVIDIA is underestimated, and the future is broader!

This view highlights Masayoshi Son's optimistic expectations for AI, while his ambitions for AI are further expanding, aiming Arm directly at the "king of chips," NVIDIA.

Currently, Arm almost monopolizes all mobile chip designs, with a market size of $500 billion, more than twice that of the PC industry. After its initial public offering, SoftBank owns about 90% of Arm, and now Masayoshi Son hopes that Arm's share of the chip industry chain will far exceed its current share in chip design.

Arm's CEO also openly revealed the AI ambitions in an interview, stating, "In the future, all AI will run on Arm in some form."

It is worth mentioning that ARM has already benefited from the current wave of AI; since Arm went public last year, its stock price has nearly doubled, with a current market value of about $157 billion, surpassing the once Silicon Valley idol, Intel.

Arm's "AI Ambition" — More than Just IP Design

Currently, Arm's role in the AI investment boom is mainly as an "ally of NVIDIA." For example, in the large data centers of Microsoft and OpenAI, the main applications are based on Arm's CPU and NVIDIA's new GPU architecture, Blackwell.

According to a report by the Financial Times on Wednesday, Masayoshi Son hopes that Arm's share of the chip industry chain will far exceed its current share in chip design, and Arm may soon develop its own AI chips to compete directly with NVIDIA.

Sources familiar with Masayoshi Son's thoughts revealed that despite the risks, SoftBank is determined to play a core role in what Son believes to be the next stage of human evolution.

Earlier this month, after SoftBank invested $500 million in OpenAI, Masayoshi Son hopes to make Arm's technology the core of data centers, specifically for training and running AI systems. This will not only lead to conflicts with NVIDIA but also mark a complete departure from its traditional business model of merely selling IP.

Stanford Business School professor Robert Burgelman stated that Arm's emergence is indeed a disruptive innovation; although its performance in computing is not outstanding, its low power consumption makes it more suitable for non-PC devices.

Arm's Competition with Qualcomm, Reaching Downstream

Masayoshi Son's AI expansion plan, coupled with legal disputes with Qualcomm, has raised concerns, as Arm attempts to enhance its position in the semiconductor industry chain while maintaining good relationships with its partners.

Recently, the conflict between Arm and Qualcomm has escalated, starting from Qualcomm's acquisition of a startup called Nuvia in 2021, which has been developing PC chips based on Arm designs. Arm claims that Qualcomm failed to obtain the necessary licenses to continue developing Nuvia chips according to its intellectual property licensing terms Earlier this month, Arm issued a formal notice announcing the complete cancellation of Qualcomm's architecture license, thereby preventing Qualcomm from releasing its latest generation of products. Qualcomm accused Arm of "taking a hardline approach" to increase patent fees.

Analysts stated that Arm took "nuclear action" against one of its largest customers, causing panic on Wall Street and leading to a 9% drop in its stock price in a single day.

The case exposes the complex relationship between Arm and its licensing parties, with some analysts concerned that this could undermine Arm's position as the "neutral Switzerland" of the chip industry.

"From Friends to Rivals," Targeting NVIDIA

In fact, Arm entered the data center field a decade ago, with Amazon being the first cloud computing company to adopt Arm. However, it wasn't until 2018 that AWS launched its first Arm-based CPU, Graviton, which initially had poor sales. Analysts noted that data center chips still account for just over 10% of Arm's patent fee revenue today.

What truly drove Arm into the core of the data center was its collaboration with NVIDIA, whose latest GB200 "super chip" combines two Blackwell GPUs with an Arm-based Grace CPU.

It is worth mentioning that NVIDIA attempted to acquire Arm for $40 billion in 2020 but subsequently failed during antitrust reviews. Today, the two companies maintain a close cooperative relationship.

Nevertheless, Masayoshi Son still hopes to compete directly with NVIDIA, which currently dominates the AI chip supply market. According to insiders, Son plans to invest billions of dollars to create an AI data center powered by Arm.

Son's ambition is to transform Arm from a knowledge property licensing company into a firm capable of producing data center chips that can be used for both "training" or building AI models and "inference." According to insiders, SoftBank's acquisition of British AI chip manufacturer Graphcore in July was also motivated by its expertise in chip production.

In addition to chip production and software, Son's plans also include providing power for data centers. SoftBank is actively engaging with potential customers about its proposed systems and promoting the advantages of creating a counterbalance to NVIDIA's ecosystem to large tech companies like Google and Meta.

Furthermore, after failing in early negotiations with Intel, SoftBank is negotiating with TSMC, the world's largest chip foundry, to ensure its foundry can produce AI chips.

Challenging the "King of Chips," What Are the Odds?

How far is it from designing IP to producing chips?

Last week, Arm CEO Rene Hass stated in an interview with the Financial Times:

Replicating NVIDIA's entire technology "stack" (which includes not only chips but also networking technology and software) would involve "a lot of heavy lifting."

While Arm has indeed manufactured so-called NPUs optimized for the neural networks required for artificial intelligence, Arm's NPUs "do not perform at the level of NVIDIA" because they are primarily designed for edge devices (smartphones, IoT, and other products outside the cloud), rather than data centers

But I can assure you that we are doing a lot of work on how to expand this space beyond just the edge.

Analysts are skeptical:

Even with SoftBank's resources backing it, can Arm create a competitor that rivals NVIDIA? Because NVIDIA's success depends not only on its hardware expertise but also on its widely used CUDA software platform, and replicating this software would take years.

Edward Wilford, a semiconductor analyst at technology consulting firm Omdia, stated:

I don't know what tools they have to compete against NVIDIA.

Ben Bajarin, an analyst at another consulting firm Creative Strategies, said:

Arm lacks data center accelerator IP, and the most likely challengers to NVIDIA's dominance are large tech companies like Microsoft, Google, and Amazon, whose AI accelerator chips do not heavily rely on Arm's IP.

However, Arm is confident, as Hass has publicly stated:

Arm can benefit from the AI boom without needing to compete with NVIDIA; chips based on Arm technology are already handling AI tasks, such as photo editing on smartphones or image processing on security cameras. The company recently collaborated with Meta to create a streamlined version of its Llama AI system that can run on its mobile CPUs.

70% of the global population is interacting with Arm in some way, and in the future, AI will run on Arm, that's just how it is.