SoftBank’s Strategic AI Investments
Masayoshi Son, the Chairman and CEO of SoftBank Group, has been vocal about his vision for ASI (Artificial Super Intelligence), projecting that ‘AI will eventually achieve a level of intelligence ten thousand times greater than humans within the next decade.’ This declaration, made in various public forums in 2024, underscores SoftBank’s accelerating focus and strategic maneuvers in the AI sector.
Around this period, SoftBank has significantly ramped up its investments and strategic initiatives in the AI domain.
In 2024, SoftBank Group made a series of notable investments in AI-driven companies. These included investing in AI startup Perplexity AI, leading an investment round in humanoid robot startup Skild AI, forming a healthcare joint venture with Tempus AI in the United States, and acquiring Graphcore, a British AI chip unicorn.
By 2025, SoftBank intensified its collaboration with OpenAI. In late March, SoftBank further expanded its footprint in the AI chip sector by announcing the acquisition of Ampere, an American chip design company, for $6.5 billion (approximately RMB 47 billion).
Coupled with its existing significant stake in Arm, these moves indicate SoftBank’s strategic ambition to bolster its investments in AI chip infrastructure. SoftBank’s approach is not just about investing in individual companies; it is about building an ecosystem where these companies can collaborate and synergize, creating a comprehensive AI infrastructure. This involves creating a vertical stack of capabilities that spans from chip design to deployment of AI applications.
A Missed Opportunity with Nvidia
Six years prior, SoftBank divested its entire stake in Nvidia, missing out on the company’s subsequent explosive growth, which saw it reach a trillion-dollar market capitalization. Now, amid the current AI surge, SoftBank appears to be making a comeback, signaling its ambition to potentially challenge Nvidia’s dominance. This episode is a constant reminder of the potential gains from a long-term view of investing, especially in a rapidly evolving technological landscape.
In November 2024, at an AI summit in Japan, Nvidia’s founder and CEO, Jensen Huang, remarked to the audience, ‘You may not know that at one point, Masa (Masayoshi Son) was Nvidia’s largest shareholder.’ He then humorously shared a moment of mock ‘crying’ with Son, adding, ‘It’s okay, we can cry together.’
This episode is viewed as a significant missed opportunity for SoftBank, a sentiment Son has publicly acknowledged with regret. The decision to sell Nvidia shares before its monumental rise is often cited as a case study in the challenges of predicting market movements and the long-term potential of disruptive technologies.
In 2017, SoftBank acquired Nvidia shares on the open market, eventually holding nearly 5% of the company, making it one of Nvidia’s largest shareholders. However, SoftBank sold its stake in 2019, missing out on Nvidia’s ascent to the peak of its growth trajectory.
Son’s enthusiasm for investing in AI chips is becoming increasingly fervent. In a public interview in October 2024, he asserted that Nvidia was ‘undervalued.’ This statement, considering Nvidia’s current market position, reflects Son’s aggressive stance on AI and his belief that the sector’s potential is far from being fully realized.
Over the past two years, SoftBank Group has been actively forging alliances and investing in AI chips and related infrastructure industries to realize its ASI vision, possibly aiming to rectify past oversights.
Son has even articulated a rationale: to advance human evolution by promoting the development of super artificial intelligence. He predicts that super artificial intelligence (ASI) will be achieved by 2035. The timeline set by Son is ambitious, reflecting his desire to rapidly push the boundaries of what is possible in the field of AI.
Son emphasizes that ASI differs from the more commonly discussed AGI (Artificial General Intelligence). AGI refers to general intelligence capable of handling multiple tasks and exhibiting human-like flexibility, which is unlikely to significantly alter existing rules in human society. ASI, on the other hand, will far surpass human intelligence, marking a turning point in human history, with ASI-driven intelligent robots performing various physical tasks on behalf of humans. This highlights the transformative potential that Son sees in ASI, a level of AI that could fundamentally reshape human society.
SoftBank’s ASI Deployment Strategy
According to SoftBank Group’s plan, deploying ASI involves four key dimensions:
- AI chips
- AI data centers
- AI robots
- Energy
Among these, AI chips are the core infrastructure. The emphasis on AI chips underscores their critical role in enabling advanced AI capabilities. Without robust chip infrastructure, the deployment of advanced AI models and applications would be severely constrained.
‘Arm will provide the foundational technology for ASI,’ Son stated. He added that while Arm is significant, no single company can achieve ASI alone. All members of the SoftBank Group will work together to achieve this goal. This highlights the collaborative approach that SoftBank is taking, leveraging its diverse portfolio of companies to achieve its ambitious goals.
This explains SoftBank’s increasing acquisition of companies in the AI chip sector: starting with its investment in Arm, followed by the acquisitions of Graphcore and Ampere, SoftBank’s AI chip strategy is becoming increasingly evident. The acquisitions reflect a deliberate effort to build a comprehensive AI chip ecosystem, with each company playing a distinct role in the overall strategy.
Anand Joshi, AI Technology Director at TechInsights, told 21st Century Business Herald that SoftBank aims to become a global leader in Artificial General Intelligence (AGI), and its recent investment activities reflect this ambition.
‘To fully realize the potential of AGI applications, a complete infrastructure is needed, covering chips, IP, servers, CPUs, AI accelerators, and more,’ he further explained. When SoftBank invests in AI semiconductors, it always focuses on a broader vision, with the three forming a perfect complement in this blueprint: Arm provides processor IP for data centers; Ampere builds data center-specific chips based on these IPs; and Graphcore focuses on the research and development of data center AI accelerator chips. This breakdown illustrates the interconnectedness of SoftBank’s investments and their strategic alignment towards building a comprehensive AI infrastructure.
Regarding how the three will form business synergies, Anand Joshi noted, ‘It is not yet clear whether the three companies plan to integrate existing products or launch new solutions, but the combination of these three has the potential to build a complete AI application infrastructure.’
Through this vertical integration, OpenAI can provide models optimized to run on this exclusive architecture, thereby achieving leading model performance worldwide. ‘Enterprise customers will purchase these AI server capabilities through API calls, and the pay-per-use model is very likely to create huge profits for them,’ he added.
Because SoftBank is building an AI core chip ecosystem through investment and acquisition, some believe that SoftBank is planning to create a potential competitor to Nvidia.
Challenges and Competition
However, at this stage, this is just a vision. On the one hand, Nvidia has built a strong moat based on more than a decade of continuous investment in software ecosystems such as CUDA. To this day, Nvidia GPU chips are still the industry’s first choice for AI training. This ecological advantage gives it a certain competitive barrier on the AI inference side; on the other hand, the ‘anti-Nvidia alliance’ that the market jokes about is accelerating its growth. A typical example is that cloud service vendors are rapidly iterating self-developed AI inference chips through cooperation with ASIC chip design companies, and Broadcom and Marvell (Marvell Electronics) are important beneficiaries. The established position of Nvidia in the AI chip market presents a significant challenge for any new entrants, including SoftBank.
Faced with the existing competitive environment, it is not easy for new entrants to make breakthroughs quickly, especially since Graphcore and Ampere both faced major financial difficulties when they were acquired by SoftBank, which means that the commercialization capabilities of the two companies remain to be improved. This raises concerns about the ability of these companies to effectively compete against established players with more mature business models and larger customer bases.
According to SoftBank’s disclosure, Ampere’s operating income narrowed from US$152 million to US$16 million between 2022 and 2024, a reduction of nearly ten times. The company seems to be trying to restore profitability, but it still lost US$581 million as of 2024. Net assets and total assets are also continuing to decline significantly. The financial challenges faced by Ampere underscore the competitive pressures in the AI chip market and the difficulties of achieving profitability in this capital-intensive industry.
According to public information, Ampere initially focused on cloud-native computing and has since expanded into the field of artificial intelligence computing (AI compute). The company’s products cover a range of cloud workloads from the edge to the cloud data center.
Graphcore’s previously submitted documents show that its sales in 2022 were US$2.7 million, with a loss of US$204.6 million. The financial performance of Graphcore further highlights the challenges of commercializing innovative AI chip technologies and achieving sustainable growth.
Regarding operating conditions, Anand Joshi told 21st Century Business Herald that although Arm and Ampere performed well, Graphcore’s development was not satisfactory. The varying performance of the different companies within SoftBank’s AI portfolio illustrates the diverse challenges and opportunities present in the AI chip market.
‘The latter’s chips are difficult to reach the performance level of the same generation of products released at the same time, which has become its main challenge. However, Graphcore has realized the importance of supporting software and has begun to invest in compilers and other technical fields. This link is precisely the core challenge of building artificial intelligence infrastructure and must be overcome,’ he continued. This underscores the critical importance of software development in the AI chip market, where the performance of the chip is heavily dependent on the software that runs on it.
In Anand Joshi’s view, in comparison, server chips based on the Arm architecture have entered the market and have a relatively mature software ecosystem. However, these products still lack the horizontal scaling ability (scaling ability) that x86 architecture possesses. ‘To succeed, these three companies need to work together to develop a unified software roadmap.’ The need for a unified software roadmap highlights the importance of collaboration within SoftBank’s AI portfolio and the potential benefits of leveraging the strengths of each company to create a more competitive overall offering.
Among them, Arm is undoubtedly a relatively mature manufacturer in terms of development. Although in the public eye, chip products based on the Arm architecture cover more than 99% of smartphones on the market, in recent years, it has also been developing rapidly for data centers, PCs and other fields. The established position of Arm in the mobile chip market provides a solid foundation for its expansion into other areas, such as data centers and PCs.
Arm Senior Vice President and General Manager of Infrastructure Business Unit Mohamed Awad recently published an article pointing out that more than six years ago, Arm launched the Arm Neoverse platform for the next generation of cloud infrastructure. Today, the deployment of Neoverse technology has reached a new height: 2025 Nearly 50% of the computing power shipped to leading hyperscale cloud service providers will be based on the Arm architecture. Hyperscale cloud service providers such as Amazon Web Services (AWS), Google Cloud and Microsoft Azure have all adopted the Arm computing platform to build their own general-purpose custom chips. The increasing adoption of Arm-based chips by hyperscale cloud service providers is a testament to their performance and cost-effectiveness, further solidifying Arm’s position in the data center market.
Anand Joshi told reporters that Arm has become an important player in the data center market. For example, Amazon is promoting its self-developed chip Graviton as a low-cost alternative to X86, and its market performance is currently good. Similarly, Amazon’s ‘Graviton+Inferential’ series of self-developed chip products are positioned as a low-cost alternative to the ‘x86+Nvidia’ solution. Nvidia has also adapted the Arm architecture to its Grace CPU chips in the Blackwell series of products. The use of Arm architecture by both Amazon and Nvidia further validates its potential in the data center market and its ability to compete with traditional x86-based solutions.
‘Therefore, if SoftBank, Arm, and Ampere can successfully implement this strategy, Arm is expected to become an unignorable force in the data center market,’ he continued.
SoftBank’s Broader AI Investment Strategy
Due to excessive investment in AI-related industries, SoftBank Corporation was required to explain its overall investment strategy in the AI industry at the investor conference in February this year. The need to explain the overall investment strategy highlights the scale and ambition of SoftBank’s AI investments and the importance of providing clarity to investors about the company’s long-term vision.
Company President and CEO Junichi Miyakawa analyzed that this includes 8 levels: deploying enterprise-level artificial intelligence project ‘Cristal intelligence’ through a joint venture with OpenAI; developing a native large language model (LLM) specifically for Japanese; working with Microsoft Japan as part of a strategic alliance in the field of generative artificial intelligence; providing enterprise-level customers with Google Workspace’s Gemini model; establishing a top Japanese artificial intelligence computing platform; establishing AI data centers in Hokkaido and Osaka; developing AI-RAN and deploying AITRAS to promote AI-RAN from concept into life; building a super distributed computing infrastructure. This breakdown illustrates the breadth and depth of SoftBank’s AI investments, covering a wide range of areas from AI models to infrastructure and applications.
This means that, facing the ASI vision, SoftBank’s layout covers a comprehensive dimension from hardware to software, from computing power to communication, and from infrastructure to solutions. The comprehensive nature of SoftBank’s AI strategy underscores its commitment to building a complete ecosystem that can support the development and deployment of advanced AI technologies.
Objectively speaking, this is also expected to help AI chip companies, which currently appear to be relatively weak in the game, further consolidate their capabilities. The investments in AI infrastructure and applications are expected to create a demand for AI chips, which will benefit SoftBank’s AI chip companies and help them to compete more effectively against established players.
Anand Joshi told 21st Century Business Herald that Nvidia’s excellent software stack has far surpassed its competitors in performance. Ampere and Graphcore are currently unable to outperform Nvidia in terms of performance. ‘They must focus on the total cost of ownership (Total Cost of Ownership) advantage, or use price/inference capabilities, performance/power consumption ratio as a breakthrough to achieve breakthroughs in market competition.’ The focus on total cost of ownership and performance/power consumption ratio highlights the importance of finding niche markets and developing differentiated products that can compete effectively against Nvidia’s dominant position.
He further pointed out that since SoftBank is a shareholder of OpenAI, they may optimize some of OpenAI’s models on the Arm and Graphcore platforms. These models may represent the most advanced AGI technology and adopt an exclusive sales strategy. This will give them a unique advantage relative to their competitors. The potential to optimize OpenAI’s models on SoftBank’s AI chip platforms could provide a significant competitive advantage, allowing them to offer unique solutions that are not available from other vendors.
‘In addition, I believe that SoftBank will promote adjustments to Arm’s technology roadmap to help the development of Ampere and Graphcore. Therefore, we will see that Arm’s IP roadmap will closely fit the AI large model needs proposed by OpenAI,’ Anand Joshi continued. The close alignment of Arm’s IP roadmap with the needs of OpenAI’s AI models highlights the potential for synergy within SoftBank’s AI portfolio and the benefits of a coordinated approach to technology development.
SoftBank is indeed strengthening its business association with OpenAI. The deepening collaboration with OpenAI is a key element of SoftBank’s AI strategy, providing access to cutting-edge AI models and technologies that can be integrated into its own AI solutions.
In February of this year, SoftBank announced its cooperation with OpenAI to build ‘Crystal Intelligence,’ and Arm is also an important member. SoftBank pointed out that as part of the agreement with OpenAI, SoftBank Group companies, including Arm and SoftBank Corporation, will be given priority in Japan to obtain the latest and most advanced models developed by OpenAI. This preferential access to OpenAI’s models will allow SoftBank to develop advanced AI applications and solutions tailored to the Japanese market.
On April 1, SoftBank announced further investment in OpenAI. SoftBank pointed out that OpenAI is an important partner in its efforts to advance towards ASI. Since September 2024, the company has invested a total of US$2.2 billion in OpenAI through SoftBank Vision Fund 2. On January 21, SoftBank and OpenAI jointly announced the ‘Stargate’ plan, which aims to build dedicated AI infrastructure for OpenAI. This time, SoftBank plans to invest up to US$30 billion in it, with another US$10 billion allocated to joint investors. The significant investment in dedicated AI infrastructure for OpenAI highlights the strategic importance of this partnership and the commitment to supporting the development of advanced AI technologies.
Of course, SoftBank’s attitude towards Nvidia is not entirely the ‘competitive/hostile’ sentiment that the outside world believes. In November 2024, that is, before and after the dialogue between Jensen Huang and Masayoshi Son, Nvidia and SoftBank announced that they would conduct business cooperation. On the one hand, SoftBank currently needs to use Nvidia GPU chips to build computing infrastructure; on the other hand, Nvidia also has deployments in communication acceleration, which will help improve the technical capabilities of AI-RAN in SoftBank’s ASI route. The recognition of Nvidia’s strengths and the willingness to collaborate reflects a pragmatic approach to achieving its AI goals, leveraging the best technologies available regardless of the source.
At the aforementioned summit, Huang Renxun said with emotion, ‘I have been involved in the technology field for many years, starting with the PC wave. The entire computing industry started with PCs, and then developed to the Internet, cloud computing, mobile cloud, and artificial intelligence. Masayoshi Son is the only person in the world who has (accurately) selected (potential) winners in each round and developed alongside them.’ Huang’s statement highlights Son’s track record of identifying and investing in disruptive technologies, a reputation that he is now seeking to reinforce with his focus on AI.
The current AI wave is surging, and the AI chip field is also surging, and giants are showing signs of accelerating competition and cooperation, seeking richer industrial chain capabilities. No matter what the outcome of Masayoshi Son’s ‘ten-year agreement’ is, it is laying the groundwork for an important footnote in the new round of technology transformation. The outcome of Son’s ambitious AI investments remains to be seen, but his efforts are undoubtedly shaping the future of the AI industry and contributing to the ongoing technological revolution. His vision for ASI and the investments he is making to achieve it are setting the stage for a significant transformation in the years to come.