From OpenAI to Malaysia, Masayoshi Sun launches the “SoftBank ASI Blueprint”! $50 billion investment plan ignites bullish sentiment in semiconductors

The Zhitong Finance App learned that SoftBank Group Corp. (SoftBank Group Corp.), headed by legendary investor Sun Zhengyi, is further extending the AI investment layout around OpenAI, the AI application leader, and ARM, a leader in chip design, to the Southeast Asian AI computing power cluster infrastructure and application market. Sun Zheng is accelerating his “AI ambition” to expand and promote a “full-stack AI ecosystem” including underlying chip architectures, AI computing infrastructure, energy, and AI applications.

According to recent media reports, Malaysian Prime Minister Anwar said that Sun Zhengyi plans to invest 50 billion US dollars in the country, involving artificial intelligence, new technology and data centers. The selected locations include Kuching in Sarawak and the Malaysian peninsula. This plan is in line with Malaysia's goal of becoming an “influential AI country” by 2030, and also provides new demand clues for the semiconductor sector of the global stock market, which has recently been heavily pressured: large investment institutions and cloud computing giants are still expanding the geographical layout of AI computing power infrastructure. Whether subsequent purchases of AI infrastructure core equipment, including AI chips, memory chips, and data center CPU systems, can be realized beyond expectations depends on the project financing process, power supply guarantee, and all-round implementation of customer computing power resource requirements.

In 2000, Sun Zhengyi first invested in Alibaba with about 20 million US dollars. Later, through long-term shareholding, he shared the record huge value growth of Chinese e-commerce platforms from the stage of huge losses in the early stages of entrepreneurship to the spotlight of the global capital market, which can be called an “investment myth.” Today, OpenAI, which Sun Zhengyi supports with huge sums of money, and many leading AI computing power forces, also has unparalleled growth potential to become a large-scale user portal and enterprise technology platform — that is, betting that cutting-edge AI chip components, AI big model capabilities, developer ecology, payment services, and enterprise workflows promote each other. In the future, the B-side and C-side commercialization space will expand fully as the scope of AI applications expands.

SoftBank plans to invest $50 billion in Malaysia striving to become an “AI nation”

Malaysian Prime Minister Anwar Ibrahim said that SoftBank Group plans to invest 50 billion US dollars in Malaysia.

According to reports, Anwar said on Saturday that after the two held a video conference earlier this week, SoftBank Group founder and helmsman Sun Zhengyi has selected locations in Kuching, Sarawak, and Peninsular Malaysia to carry out the proposed investment.

According to reports, the Malaysian prime minister said that the proposed investment will focus on artificial intelligence, new cutting-edge technology, and large-scale AI data center construction projects.

The investment plan comes at a time when Malaysia is seeking to become an “influential AI nation” by 2030. The government launched a national artificial intelligence action plan earlier this year to strengthen the country's artificial intelligence capabilities and infrastructure.

Earlier this month, SoftBank signed a framework agreement with Singapore-based Grab and Malaysia's Sarawak Petroleum to explore the development of artificial intelligence and digital infrastructure in Sarawak, including AI computing power services and advanced technology.

SoftBank isn't the only company looking to invest in technology in Malaysia. Google's parent company Alphabet previously announced an investment of 2 billion US dollars in 2024 to build its first large-scale AI data center and large-scale AI cloud computing region in the country; AWS Cloud Technology, the world's largest cloud computing service provider under e-commerce and cloud computing giant Amazon, said it plans to invest 6.2 billion US dollars by 2038. Recently, Microsoft also announced an investment of about 2.2 billion US dollars to build infrastructure for traditional cloud computing capacity and artificial intelligence computing power, and other large-scale digital projects.

According to data from the Malaysian Investment Development Authority, from 2021 to the first half of 2026, the country recorded an investment scale of around 385.7 billion ringgit, or about US$94 billion, associated with large-scale AI data centers. SoftBank representatives, on the other hand, did not immediately respond to the media's first request for comment.

Sun Zhengyi paves AI ambitions to AI infrastructure, energy and applications

Sun Zhengyi's layout has formed a multi-layer structure connecting flagship AI applications such as large AI models, chip-based AI computing power infrastructure, energy, and AI agents.

On October 1, SoftBank announced the completion of the final additional investment of 10 billion US dollars in OpenAI, with a cumulative investment of 64.6 billion US dollars and a shareholding of about 13%. On the chip side, ARM central processor architecture and ARM's self-developed data center CPU, jointly expand the computing capacity layout with the acquired Ampere and Graphcore; on the infrastructure side, Stargate (the “Stargate” large-scale AI data center project in the US), SB Energy, and SB Neo, which plans to carry out computing power cloud service leasing after new cloud providers such as CoreWeave, undertake data center construction, energy support and computing power supply; AI application side for Japanese market enterprises, SB OpenAI Japan Layout customized agent applications.

Judging from the AI industry structure, this comprehensive layout for super artificial intelligence (ASI) aims to simultaneously obtain superplatform equity appreciation, AI chip-based AI infrastructure business growth, and AI cloud computing resource leasing and sales revenue.

The AI superplatform now being built by SoftBank is no longer a single point of betting, but an increasingly complete set of “ASI integrated stacks” — that is, the bottom layer includes Arm architecture and Armself-developed data center CPUs, the middle layer includes the British AI chip design company GraphCore Ltd., which was fully acquired by SoftBank Group, and Ampere, which was officially acquired recently; the upper layer includes OpenAI, Stargate (or “Stargate”), and the enterprise-level cloud AI computing power platform layout in partnership with Oracle, plus SB Neo, a new cloud power ( Neocloud); Meanwhile, Sun Zhengyi himself has already publicly stated that SoftBank's goal is to become the largest AI computing power and application-level basic platform provider under the “Super Artificial Intelligence” (ASI) era in the next ten years, and has adopted an almost “all in” aggressive investment attitude for ChatGPT developer and global AI model leader OpenAI. Oracle is mainly Stargate's cloud and infrastructure partner, while SoftBank's joint venture platform for enterprises is SB OpenAI Japan, which was co-founded with OpenAI.

The Malaysian cooperation further reflects this industrial collaboration: SoftBank provides AI and semiconductor ecosystems, Grab provides regional customer networks and application scenarios, and Sarawak oil company PETROS participates in energy supply and supporting infrastructure. Its business logic is to combine computing power, power resources, and actual usage requirements to increase the possibility of data centers generating continuous revenue. Currently, SoftBank officially announced an exploratory framework agreement signed on October 8; the investment period, funding structure, and equipment procurement arrangements for the $50 billion plan have not been disclosed, so this is a potential increase in demand worth tracking, and cannot yet be fully included in SoftBank's own capital expenditure or chip manufacturer orders.

The financing window is still open, but the price is more expensive: a major semiconductor counterattack requires a relay between orders and cash flow

According to the September corporate junk credit bond subscription data, the initial subscription of more than 20 billion US dollars indicates that credit capital is still willing to support Sun Zhengyi's AI expansion, but the required returns have increased markedly.

The financing has reached the final result: SoftBank decided to issue 10 billion US dollars and 1 billion euro bonds in September, totaling about US$11.1 billion; the three-tier US dollar bond dividends are 8.625%, 9.25%, and 9.75%, respectively. The highest tier is close to 10% of the previous test. Based on the scale of issuance and coupon interest, the annual interest on the US dollar portion alone is approximately US$941 million. The company later confirmed that the proceeds from the bond issue will be used to complete the additional investment in OpenAI; the Malaysian project still needs to implement its own funding arrangements. The initial popularity of subscriptions reflected the availability of financing channels, and the final coupon revealed the true cost of using these funds.

The stock and corporate bond markets are already re-evaluating the risk-benefit ratio of this expansion. Based on the closing price from September 30 to October 23, SoftBank Group's Tokyo stock price fell from 6,430 yen to 4,276 yen, a cumulative decrease of about 33.5%; CoreWeave fell from $87.12 to $54.24, a decrease of about 37.7%. On October 22, the Philadelphia Semiconductor Index fell 6.7% in a single day, indicating that concerns about financing and returns have spread along the AI industry chain. SoftBank's 5-year credit default swap (CDS) rose to 432 basis points on October 20, reflecting investors' demand for more credit risk compensation.

Due to the recent rise in 10-year US bond yields and continuing to be near the highest level in more than 20 years, the pressure on AI financing costs faced by CoreWeave (CRWV.US), which is the core stock sell-off in the “AI computing power sell-off storm” and the “AI bubble bursting argument”, is more direct: the October 22 report showed that the yield on its 2031 bond maturity was about 15.76%, and the five-year CDS was about 1,115 basis points; previously reported by the media on October 15, the five-year CDS was priced at a time close to 63% of the cumulative five-year period Implied probability of default — However, this probability includes recovery rate assumptions and risk premiums; it is not an actual default prediction.

The mechanism behind this pessimistic pricing in the bond market is that large-scale AI data center construction and equipment procurement require cash payment in advance, while computing power revenue is gradually formed with delivery, customer use, and contract execution; the combined rise of benchmark interest rates and credit spreads will increase the cost of additional financing and reduce the project's cash return. SoftBank is also facing another level of timing mismatch: the increase in OpenAI's equity value will not be simultaneously converted into cash that can be used to pay interest.

Looking at the underlying architecture of AI computing power infrastructure, Muse, Astra, and Meta's latest open source Muse Spark Open are expected to jointly further expand the scope of use of AI agents. The massive expansion of AI inference workloads brought by them will also increase the demand for model inference, tool execution, and state management, while deepening the competitive pressure on some software companies. Accelerators such as GPUs and TPUs undertake model computation; high-performance server CPUs run browsers, virtual machines, product search, database queries, and transaction orchestration; HBM and server DRAM carry model data, context, and concurrent working environments; enterprise-grade SSDs store product indexes, task records, and persistent states, and high-speed networks and optical interconnections support distributed data exchange.

Judging from this, the positive significance of large-scale AI infrastructure plans led by the Malaysian government for semiconductors is to further expand the potential semiconductor procurement scale and demand sources closely linked to AI infrastructure. Continuous operation of intelligent devices requires AI accelerators such as AI GPU/TPU/AI ASIC to complete model calculation, data center CPU execution tools and task orchestration, HBM and server DRAM carrying models and operating states, and enterprise-grade high-performance SSDs to store data, and exchange information through high-speed networks. After the project is implemented, these 10 billion dollar semiconductor demand is expected to be transmitted along the lowest core AI computing power industry chain chain, such as accelerated computing, memory, wafer manufacturing, 2.5D/3D advanced packaging, high-performance network equipment, and data center optical interconnection; the specific beneficiary manufacturers and shares will still depend on technology selection. The most investment-worthy development is the gradual transformation of the $50 billion plan into AI infrastructure-related data center equipment orders with capital, electricity, and customer support. This not only enhances the visibility of suppliers' revenue, but also allows the market to more accurately determine whether the high financing costs that construction parties continue to bear can eventually be covered by cash flow from computing power operations.