At 69, still calling past investments a warm-up: Masayoshi Son talks about betting on the endgame, a 70% dream achievement rate, and rapid strategic adjustments.
Contents
Original Statement
Macroeconomic and Capital Environment: The AI Infrastructure Boom and Economic Barometer
- High corporate borrowing costs and interest rates:
- The yield on 10-year U.S. Treasury bonds once broke 5.3%, reaching the highest level since 2002; however, this phenomenon is mainly driven by economic fundamentals and growth expectations, rather than a severe recession.
- In the past few months, U.S. stocks have risen nearly 20% against the trend, with third-quarter GDP growth expectations maintained at around 3% (significantly higher than the long-term potential growth trend in the U.S.).
- Explosive capital expenditure (Capex) in super-scale computing power:
- Major global cloud providers and tech giants (Hyperscalers) are expected to spend nearly $800 billion on AI infrastructure this year, with projections to officially surpass $1 trillion next year.
- Infrastructure investments are rapidly transmitting to balance sheets and revenue: leading cloud computing companies have achieved quarterly revenue growth rates of 50% to 120%, directly driving the underlying growth of logistics real estate and digital energy networks under Blackstone.
- Blackstone's practical experience: the real bottleneck for companies implementing AI:
- Overcoming activation barriers: By 2025, the biggest beneficiaries of AI will mostly be "programmers who walk their dogs earlier after work"; achieving real net profit increments on the corporate side requires overcoming organizational friction and system reconstruction costs, which creates a high barrier to success but also builds a long-term barrier that is not easily eroded.
- Implementation rules: firmly avoid "full-scale rollout" and focus on 1-2 core business processes; prohibit the technology department from acting independently; it must be closely paired with the core decision-making team of the business to promote progress.
Masayoshi Son's core prediction: from AGI to ASI (Superintelligence)
- The fateful intuition from 50 years ago:
- At 17, when I first saw a photo of a microprocessor while flipping through an electronic magazine on the street, I cried; at that moment, I determined that this chip would eventually surpass the human brain in computation and intelligence; I predicted it would take 50 years, and now at 69, it has fully come to fruition.
- The grand narrative that "everything in the past was just a warm-up":
- Reflecting on investments in Alibaba, Yahoo, acquiring ARM, and founding the $100 billion Vision Fund, Masayoshi Son admits that in the face of the upcoming AI revolution, the past business scale is merely insignificant warm-up.
- Ten thousand times human wisdom (ASI) and autonomous evolution:
- It is predicted that within the next 10 years, the computing power and cognitive ability of superintelligence will reach 10,000 times that of humans.
- Once this threshold is crossed, discussing whether AGI has been achieved will be meaningless; AI will possess completely independent self-development, code iteration, and scientific paradigm discovery capabilities, thoroughly conquering difficult diseases like cancer and reconstructing climate and disaster predictions.
SoftBank's strategic matrix: full industry chain ecosystem and embodied robots
- Refusing to make single-point bets, building a full-stack ecosystem:
- SoftBank is not positioned as a single operating entity CEO but as a full ecosystem architect.
- Advancing comprehensively from underlying chip architecture (ARM), data center energy infrastructure, core large language models (holding significant equity in OpenAI) to the application layer; the investment strategy insists on becoming the largest shareholder with 20%, 30%, or even 80% equity to promote internal synergy through control and deep empowerment.
- Physical world takeover: the secret layout of 30 robotics companies:
- AI capabilities are accelerating from "virtual data centers" into "physical machines embodied."
- SoftBank's ecosystem has secretly laid out over 30 embodied robotics companies and launched a physical-level AI super factory that "manufactures robots that make robots."
- The goal is to completely liberate humans from tedious and repetitive physical labor (washing dishes, basic logistics, hazardous operations) and drive society towards high emotional connections, family companionship, and creative entertainment.
Governance bottom line and safety ethics: accelerating the fight against AI risks
- Opposing artificially slowing down development:
- In response to recent controversies regarding system data forgery, concealment of errors, and unauthorized operations disclosed by OpenAI, Masayoshi Son clearly advocates not to slow down the pace of technological evolution.
- Automobile civilization analogy: The popularization of cars led to traffic laws such as prohibiting drunk driving, driving on the right, and traffic lights; the essence of AI governance is also to build firewalls with more advanced technology, rather than prohibiting car manufacturing.
- The cultivation mechanism of "Good AI Agents":
- Technology itself is the best means to solve risks; while accelerating progress, human social order, empathy, and moral bottom lines must be embedded as core alignment functions in the system to eliminate falsehoods and loss of control.
Investment philosophy and organizational decision-making mechanism
- "Dreams are proportional law" and long-termism:
- 70% achievement rate rule: Whether setting small wishes or grand visions to change the world, the average execution achievement rate is usually around 70%; since the same energy and cost are expended, one must conceive an extremely grand ending from the beginning.
- Extreme long-cycle strategy and high-frequency 24-hour correction:
- The core topic of SoftBank executive meetings always revolves around "strategic simulations for the next 10 years," but every day is dynamically reconstructed based on new variables and technological breakthroughs captured in the past 24 hours.
- The less clear the world will be in 10 years, the more one cannot give up on simulations; one must never retreat to merely thinking about the next year under the excuse of "the future is unknown."
- Personal AI symbiotic habits:
- Masayoshi Son revealed that he deeply uses AI for thought alignment and debate for 3 hours every weekday, and up to 7 hours on weekends; he sees it as a business thinking partner (Thought Partner) that challenges his strategic blind spots around the clock.
Steve Schwarzman's closed-door review
- The foresight success rate of a very few people:
- Masayoshi Son's core trait is to see early the endgame of the combination of technology and commercialization, and after identifying key leaders, he dares to push all chips to the table (Bet the House).
- He possesses an extraordinary resilience and steel nerves; because he firmly believes that the direction of major trends will inevitably be established, short-term valuation fluctuations and cyclical setbacks cannot shake his underlying will.
- Consistency of Blackstone's philosophy:
- In the business world, the effort expended on doing a mediocre small thing is exactly the same as doing a world-changing big thing; the most precious asset in life is irreversible time, and one must invest all energy into significant opportunities that have a long slope and thick snow for 20-30 years.
Implementation verification checklist (for strategic planning and technology investment review)
- Macroeconomic goal scale testing: Review the team's current annual and three-year plans, and based on the "70% dream achievement rate" principle, check whether the goal is a conservative incremental improvement or has enough commercial ceiling for a generational leap.
- AI application business pairing audit: Check whether the AI efficiency improvement pilot projects being promoted by the company have a disconnection phenomenon of "IT/technical department self-indulgence," ensuring that the business's top position and R&D backbone establish a landing group with weekly indicator linkage.
- Embodiment and physical scenario simulation: Sort out the links in the business chain that involve physical delivery, offline customer service, or heavy labor, and assess the maturity of scenarios for integrating embodied robots or multimodal agents in the next 3-5 years.
- Decision-making correction agility verification: While maintaining the long-term strategic direction unchanged for 5-10 years, establish a rapid calibration mechanism for absorbing new technological variables in short cycles (weekly/monthly) to eliminate strategic rigidity.
ABAB AI Insight
AI Deep Interpretation | Masayoshi Son's Tenfold Intelligent Gamble: In the Next Decade, Who Will Truly Control Wealth, Productivity, and the Lifeblood of the Global Economy?
- Masayoshi Son is betting not on artificial intelligence, but on the control of the next generation of the global economy.
SoftBank's continuous layout from Arm chip architecture, OpenAI, data centers to robotics is essentially a competition for the key production factors of the future intelligent economy. This has a similar business logic to Rockefeller's integration of the oil industry chain and the establishment of transportation networks through railway capital: not satisfied with selling a single product, but attempting to gain benefits at multiple key links.
However, owning equity in multiple links of the industrial chain does not equate to controlling the entire ecosystem. What truly determines long-term returns are scarce resources, bargaining power, investment costs, and cash flow. The scale of the AI industry may grow tenfold in the future, but most companies in the industry may not achieve tenfold profits.
- The biggest contradiction in AI infrastructure: the more advanced the technology, the more fragile the capital may be.
Traditional software companies can serve more customers at a lower marginal cost, but the AI industry must continuously invest in chips, electricity, data centers, and computing networks. This makes the AI revolution simultaneously a massive capital financing competition.
In October 2026, SoftBank issued approximately $11 billion in high-yield bonds, with some bond yields reaching as high as 9.75%. This reveals an important risk: if companies borrow high-cost funds to invest in long-term assets, future cash flows must cover interest, operations, equipment depreciation, and ongoing capital expenditures.
Assuming a long-term capital return rate of only 8% for an asset, while the financing cost reaches 10%, expanding the investment scale may not necessarily create value.
The historical American railroad bubble has already proven: infrastructure can change the world, but shareholders building infrastructure may go bankrupt due to excessive debt.
- Even if superintelligence is realized, the economic benefits gained by human society do not equate to the profits obtained by investors.
Masayoshi Son predicts that in the next decade, there may be superintelligence that is a million times smarter than humans. However, "millionfold intelligence" currently lacks a unified, verifiable measurement standard, and its realization time and capability scope are highly uncertain.
From an economic perspective, what truly matters is not how many times AI is smarter than humans, but how much the cost of completing the same economic task can decrease.
If a professional service that originally costs $1,000 can be completed by AI for only $10 in the future, the theoretical cost decreases by 99%.
However, the remaining $990 will not automatically enter the profit statement of AI companies. Customers may save costs, competitors may lower prices, and new businesses may enter the market.
This means that AI can create enormous social wealth, but it does not necessarily allow all AI companies to achieve excess profits.
Technology creates value, competition determines price, and business barriers determine who can retain profits.
- The most easily overlooked winners in the AI era may be traditional companies that have customers and real business scenarios.
The market tends to focus on Nvidia, OpenAI, and robotics companies, but easily overlooks how AI changes the profit structure of traditional industries.
Assuming an insurance company spends $1 billion annually on handling claims. If a rigorously validated AI workflow helps it reduce related costs by 20%, it could potentially free up $200 million in annual cost space.
This economic benefit does not necessarily all belong to AI technology suppliers.
Insurance companies that have customers, licenses, business processes, and industry data may also become major beneficiaries.
Therefore, the future value capture of AI may emerge in two paths: one is selling intelligent services to the entire market; the other is traditional companies using intelligent technology to improve profit margins.
The latter, although lacking the appeal of a tech narrative, may be closer to real cash flow.
- The key to the robotics revolution is not to imitate humans, but to give machines a first-time economic advantage.
Masayoshi Son revealed that SoftBank's portfolio includes over 30 robotics companies, with the goal of pushing AI from the digital world into the physical world.
However, the robotics industry cannot simply replicate the growth logic of the software industry.
Software can be copied without the need to manufacture new physical products, while robotics involves materials, energy, maintenance, safety, depreciation, and on-site deployment.
The core metric for a robotics company should not just be how many robots are sold, but rather the effective working cost per hour.
For example, if a robot costs $100,000, assuming it can be used for 5 years and operates effectively for 4,000 hours each year, the amortized purchase cost would be about $5 per hour. Adding maintenance, energy, software, insurance, and downtime costs will yield the real economic cost.
If its reliability meets commercial requirements and it can complete tasks at a significantly lower total cost than human labor, then scaling up may be possible.
Thus, standardized environments like warehousing, manufacturing, and logistics may see mature robotic business models emerge earlier than complex home services.
- Masayoshi Son's successes and failures prove: betting on the right era does not equate to betting on the right company.
SoftBank's investment in Alibaba is one of the classic cases in the history of global tech investment, while WeWork has become an important warning of capital allocation errors.
There is a common point between the two: investors believe the companies have the opportunity to change large traditional markets.
But the business outcomes are starkly different.
Alibaba established a vast trading network and business ecosystem. WeWork, on the other hand, has long faced high leasing obligations, cash consumption, and corporate governance issues.
Therefore, investments must simultaneously assess three questions:
Is the industry direction correct?
Does the company have a sustainable competitive advantage?
Is the current purchase price reasonable?
If the industry judgment is correct but the company lacks a moat, or if the investment price is too high, significant losses may still occur.
Excellent long-term investment is not about perpetual optimism, but about ensuring that the returns from correct judgments are sufficient to cover losses from incorrect judgments.
- Blackstone's AI experience reveals a counterintuitive phenomenon: the more efficient employees are, the less profitable the company may be.
If employees use AI to reduce report preparation time by 80%, it does not mean the company's profits increase by 80%.
Because a company's costs also include customer acquisition, approvals, management, rework, and actual delivery.
Assuming a business process takes 10 hours, with report writing only taking 2 hours. Even if AI completely eliminates those 2 hours, the entire process may save at most 20% of the original working hours.
If this leads to additional costs for review, system maintenance, and data security, the final savings may be even less.
This indicates that the core of AI commercialization is not to have everyone install AI tools, but to redesign end-to-end workflows.
Companies should prioritize high-cost, high-frequency, verifiable outcome businesses, implemented by business leaders and technical teams together.
Truly valuable AI transformation must be verifiable in profit statements, cash flow, or operational quality, rather than merely reflected in the frequency of employees using tools.
- The biggest organizational change brought by AI may be the decoupling of company size and production capacity.
In the industrial age, expanding capacity typically required more factories, machines, and workers.
In the internet age, software companies can serve millions of users with relatively few employees.
In the AI era, some knowledge work may be further automated, with a few individuals managing multiple specialized agents to accomplish tasks that previously required large teams.
However, having production capacity with a small team does not automatically equate to commercial success.
When all competitors can cheaply generate code, designs, videos, and business reports, general production capacity itself will depreciate.
The truly scarce resources will gradually shift to customer trust, unique brands, expertise, original intellectual property, and distribution channels.
This is strikingly similar to the internet revolution: web page creation is becoming cheaper, but platforms that have user relationships and strong network effects can still achieve significant commercial value.
In the future, what individuals should pursue is not just learning to use AI, but leveraging AI to build production systems that can independently generate income, accumulate customers, and sustain operations.
- Masayoshi Son's ten-year strategy truly worth learning is the combination of long-term direction and short-term corrections.
Many people understand long-termism as never changing decisions.
But Masayoshi Son emphasizes that on one hand, one should continuously think about the next decade, while on the other hand, constantly adjust strategies based on the latest information.
These two are not contradictory.
Long-term strategy answers: which structural changes are worth investing resources in?
Short-term execution answers: are current assumptions being validated by facts?
For example, a startup believes there is a huge market for AI education in the next decade, but cannot continuously invest funds for five years without verifying whether customers are willing to pay.
A more reasonable approach is to choose a direction with a ten-year perspective, allocate resources with a one-year goal, use quarterly data to assess the business model, and adjust based on real results in a timely manner.
Sticking to long-term goals does not mean sticking to the wrong path.
- The biggest wealth change brought by AI may not be the creation of new jobs, but the redistribution of labor, capital, and intellectual property returns.
Assuming AI allows a company to achieve the same output with half the manpower.
How will the economic value saved be distributed?
It may convert into corporate profits, benefit consumers through lower prices, or be used to increase employee salaries or expand the business.
The final outcome depends on market competition, workers' bargaining power, asset ownership, and corporate governance.
This means that merely improving individual work efficiency does not guarantee a simultaneous increase in personal income.
When a skill is rapidly popularized by AI, its market price may even decline.
Therefore, ordinary people need to gradually shift from accumulating a single labor skill to accumulating assets and capabilities that can create long-term value, including professional brands, customer relationships, intellectual property, corporate equity, and sustainable operational business systems.
However, asset ownership also means bearing corresponding risks, and one should not view entrepreneurship or investment as a guaranteed shortcut to wealth.
- After the widespread adoption of AI, the most expensive resource may no longer be intelligence, but reliability and accountability.
When a large number of models can generate fluent answers, providing answers may become increasingly cheap.
But who can guarantee the correctness of the answers?
Who can take responsibility for errors?
Who can meet the safety requirements of finance, healthcare, law, and national critical infrastructure?
For example, if a piece of AI-generated ordinary advertising copy contains an error, the loss may be limited; but if a system that automatically executes financial transactions, medical judgments, or industrial control tasks makes a mistake, it could lead to serious consequences.
Therefore, in the future, an important source of profit in the AI industry may not only be model performance but also model evaluation, security auditing, enterprise-level reliability, permission management, and professional accountability systems.
The more critical the decision, the more customers will purchase not just intelligence, but verifiable results and accountable responsibility.
- Masayoshi Son's greatest insight: do not mistake the ability to predict the future for the ability to master the future.
From microprocessors to the internet, and then to superintelligence, Masayoshi Son has always been willing to bet on technological changes in advance.
This foresight is worth studying.
However, truly successful capital allocation must also meet four conditions: correct direction, quality assets, reasonable purchase price, and a financing structure that can withstand cycles.
Even if superintelligence truly emerges in the next decade, companies participating in today's investments may still experience technological obsolescence, financing crises, and fierce competition.
Therefore, for entrepreneurs, one should learn from Masayoshi Son's strategic vision in seeking huge markets; for investors, it is even more important to combine risk control and capital discipline emphasized by long-term capital management institutions like Blackstone.
Final Judgment: In the next decade, AI may become one of the most widely influential productivity technologies in human history. However, technology that changes the world does not necessarily create successful investments. The true long-term winners are those companies that can acquire scarce assets at bearable capital costs, establish competitive barriers, and continuously convert technological advancements into free cash flow.
Understanding trends determines which era you enter, while understanding capital determines whether you can leave wealth in this era.