Original · Unique Research · 2026-03-24 · Shanghai
Editor's note: This complete English edition preserves the original author's interpretation and Bu Liangyuan's March 2026 investment views, not independent investment guidance. The opening describes 90% or more of a portfolio as AI projects without specifying the measurement basis; the later 90% focus on applications concerns attention and future allocation and is not the same measure. The 12–18-month cash-flow horizon is an investment-screening expectation, not a forecast guarantee. The Robot MART case reports 30% higher store foot traffic—not revenue or profit—and provides no baseline, period or independent verification. Corporate descriptions, partnership details and English renderings of names remain subject to confirmation; Shidai Bole is a romanization of 时代伯乐. Predictions about AI markets and AGI are attributed opinions. All substantive questions, repeated arguments and Q&A are retained.
Unique Awards · Guest Interview
AI Accounts for 90% of His Portfolio, Yet He Says: When Investing in AI in 2026, First Ask Whether It Can Make Money
The Yardstick for AI Investment Is Already Changing
At a listed company's investment department, AI projects account for more than 90% of the portfolio.
You might expect his first questions to founders to be: What is your technical moat? Where is your training data? How far have your multimodal capabilities progressed?
But my first strong impression of Bu Liangyuan was not that he understood technology deeply.
It was that he no longer put technology first.
His judgment is simple:
"In 2026, what matters most to us is not whether you have AI technology, but whether you can make money with it."
That says a great deal about the present moment.
Over the past two years, what commanded the highest price in AI was often neither revenue nor profit, but technological imagination.
With strong enough team credentials, a sufficiently hot field and a big enough story, the market was willing to assign a valuation even when customers and revenue were still at an early stage.
But now the questions are changing.
The market no longer asks only how far the technology has progressed. It is pressing for more concrete answers:
Who is paying?
Why are they paying?
Is this an essential need?
Can it be replicated?
Can it generate stable revenue?
Can you achieve positive cash flow within 12 to 18 months?
Once those become the questions, projects quickly separate into tiers.
Some companies may not have the flashiest technology, but they have found a real need in an industry and customers willing to pay;
others look cutting-edge and talk constantly about trends and the future, yet become vague as soon as you ask about revenue, repeat purchases, delivery and cash flow.
So what makes this interview worth reading is not merely how an investor evaluates AI.
When an industrial investor heavily invested in AI begins putting making money explicitly on the table, something larger is being said:
The yardstick for AI investment is changing.
Previously, the first question was whether you could build something.
Now, it is whether you can sell what you have built.
That was my strongest impression after speaking with Bu Liangyuan:
In 2026, AI investment no longer treats the ability to build technology and the ability to run a business as the same thing.
Today's More Valuable AI Companies May Not Have the Strongest Technology, but They Are Closer to a Real Business
Many people interpret this as the market becoming shortsighted.
As if asking about revenue before technology in an AI discussion means capital has become conservative.
That is not really what is happening.
More precisely, the market is beginning to distinguish building a product from building a company.
From 2024 to 2025, AI investment was largely looking at one question:
Can you build it?
People were willing to pay for going from 0 to 1.
A technically strong team that could produce a product quickly and find a few pilot use cases was already attractive.
But in 2026, the market is asking more than whether it can be built:
Once built, who will buy it?
Why must customers buy it?
Is the pain point sufficiently acute?
Can it grow from a project into a business?
Can a handful of customers become a repeatable revenue model?
Bu Liangyuan's phrase for AI investment in 2026 is an end-to-end path to commercialization.
The meaning behind those six Chinese characters is clear:
The market will pay less and less for potentially enormous opportunities, and more and more for some traction already demonstrated.
A recurring problem in AI has been confusing the ability to build a demo with the ability to build a company, and customers' willingness to listen with their willingness to pay.
Those have never been the same things.
For companies that continue to win market recognition, technology will still matter, of course. But they must take another step:
From technical capability to business results.
The Biggest Bubble Is Not Fake Technology, but Fake Deployment
When people mention an AI bubble, they usually first think of inflated valuations, inflated concepts or model hype.
All of those certainly exist.
But one point Bu Liangyuan made felt closer to reality:
For many projects today, the problem is not that the technology is entirely unsound, but that its real-world implementation does not hold up.
On the surface, there is a product, an interface, conversations with customers, partnership cases, a clear story and polished packaging.
But keep asking questions and weak points begin to emerge:
Is the need painful enough?
Can it be solved without AI?
How much efficiency does AI actually add, how much cost does it remove, and how much does it improve the experience?
Is the customer paying for a one-off experiment, or will they keep buying?
If a major company builds the same thing tomorrow, what will you have left?
They use one internal question that I think is especially effective at screening projects:
"Can the problem your product addresses be solved without AI?"
Once that is asked, many projects cannot go much further.
Many so-called AI products today are simply old problems in new wrappers. The interface, narrative and fundraising vocabulary change, but no new value is created.
Such projects can all look respectable while the trend is hot.
But once the market starts seriously examining business fundamentals, their fragility becomes apparent.
For AI companies to pass the test in 2026, the label AI is no longer enough.
They need to answer one question clearly:
Was this inherently unviable without AI, and is AI precisely what makes it viable for the first time?
That distinction determines whether a project merely borrows the AI label or has genuinely been rewritten by AI.
Why an Investor Heavily Exposed to AI Ranks Industry Experience Above Technical Background
Bu Liangyuan ranks founders' capabilities as follows:
Industry background > business background > technical background.
A few years ago, many people might not have accepted that.
Today, it increasingly resembles the actual preferences of those working on the front lines of the market.
The reason is not complicated.
When AI first took off a few years ago, competition centered on who could capture the technology dividend earlier and build capabilities faster.
Technical founders had a clearer advantage at that stage.
But increasingly, projects are not stuck on whether the technology can be built. They are stuck on whether it can work once placed inside an industry.
Do you truly understand that industry?
Do you know where customers' real pain points lie?
Do you know who controls the budget, who makes the decision, who uses the product and who bears the cost of change?
Do you know whether the product will encounter bottlenecks in procurement, delivery, data permissions or organizational cooperation once it goes live?
Those questions determine not whether the product can be demonstrated, but whether it can close a sale.
That is why Bu Liangyuan particularly values three qualities in founders:
Depth of industry understanding, execution in commercialization, and the resolve to think long term.
The ordering itself is revealing.
For many AI founders today, the problem is not inadequate technology, but choosing the wrong direction from the outset.
They expend enormous effort optimizing a problem their customers are in no hurry to solve;
they build something that looks advanced, but for which customers have set aside no budget.
The project ultimately fails not because it cannot be built, but because no one buys it.
The market now wants more than people who understand technology:
It wants people who can place AI in a real business workflow and turn it into revenue.
What Gets Rejected Immediately Is Not Weak Technology, but a Disconnect Between Understanding and Commercialization
When asked which projects they would immediately pass on, Bu Liangyuan named two conditions.
First: a founder whose understanding is misaligned and who lacks a commercial mindset.
Second: a field with extreme competition and established leaders that have effectively monopolized it, where the team has no differentiated advantage.
Both are very practical.
Many founders are excited to discuss technology and can speak convincingly about future market potential.
But once the conversation turns to customers, revenue, delivery, repeat purchases, channels and gross margins, they become vague.
These projects have something in common:
They are passionate about making products but lack sufficient respect for what it takes to do business.
In the past, such projects could sometimes buy time with the prospect of high growth.
That route is now narrowing.
Capital markets are returning to a simpler question:
Can you sustain yourself?
That is why Bu Liangyuan places such emphasis on a viable commercial cycle.
He wants to see not whether you can describe a grander future, but whether you can make several milestones real over the next 12 to 18 months:
A clear model for profitability;
paying customers already using the product;
the prospect of revenue at scale;
and the possibility of turning cash flow positive.
Without these, even the strongest technology and biggest story will find it increasingly difficult to raise money.
The Three Most Common Founder Mistakes Are Mostly About Feeling Too Good About Oneself
Bu Liangyuan named three common mistakes by AI founders that I found particularly representative.
First, becoming absorbed in technology while losing touch with real industry needs.
Many technically trained founders become fixated on parameters, performance and capability limits. But customers do not make decisions that way. They may not care how powerful your model is; they care whether it solves a problem, saves money, increases revenue or reduces errors.
Second, lacking respect for commercialization and relying on fundraising to finance spending.
A few years ago, many assumed AI companies should first acquire users, build scale and talk about the future, with making money coming later. The financing environment no longer supports that approach. Projects that cannot sustain themselves and depend solely on external funding will find survival increasingly difficult.
Third, blindly benchmarking against major companies and falling into undifferentiated competition.
They build whatever the major companies build and repeat whatever those companies say. Products become more alike, prices fall, and when a major company makes its offering free, the startup loses its room to operate.
These mistakes look different, but share an underlying problem:
Failing to find one's own position.
Building in AI today requires asking more than whether a direction is hot:
Am I solving a real problem?
Do I have my own moat?
If a major company arrives, what can I rely on to survive?
Without clear answers, the road ahead will only get harder.
Major Companies Do Not Eliminate Startups' Chances, but They Make Positioning More Important
When founders see ByteDance, Alibaba or OpenAI enter a field, their first reaction is often that the opportunity is gone.
Bu Liangyuan believes opportunities remain—and substantial ones.
But the condition is finding a niche, not confronting the giants head-on.
The reasons are straightforward.
First, major companies cannot cover every specialized use case.
China's industrial chains are too long, detailed and fragmented. Major companies must prioritize broad applications and sufficiently large markets; they cannot deeply adapt their offerings to every vertical industry.
Second, they struggle to obtain the most critical data and know-how in vertical industries.
Many industry customers are unwilling to give major companies their key data. They understand that a company providing tools today might enter their own industry tomorrow.
Third, major companies struggle to offer extensive customization and rapid responses.
Many industrial customers do not want a standardized product for everyone. They want a solution that fits into their own workflows.
That is why many startups still have an opportunity.
Not because they can outspend major companies,
but because they can be stronger in industry depth, customization, response speed and customer relationships.
The more sensible approach today is therefore neither simply avoiding major companies nor insisting on challenging them.
It is finding areas they do not want to serve, cannot serve well, or cannot serve economically.
Many companies are not ultimately killed by a giant.
They choose the wrong battlefield first.
Why He Prefers AI Native Founders Learning an Industry to Industry Veterans Learning AI
This judgment is also revealing.
Given a choice between industry veterans learning AI and AI native founders learning an industry, Bu Liangyuan favors the latter.
The reason is practical.
AI changes too quickly.
Technical approaches, model capabilities and product forms are evolving rapidly.
For many traditional entrepreneurs, the challenge is not merely learning a new technology. It is replacing ways of thinking, managing and setting a pace that have formed over decades.
That is difficult.
Many people from traditional industries are not lacking in effort. They instinctively bring their previous expectations to AI entrepreneurship:
Be cautious with R&D spending,
keep experimentation costs low,
show results sooner,
and preferably advance along a linear path.
But that is often not how AI startups develop.
They require frequent experimentation and rapid iteration, as well as accepting uncertainty for a time.
By contrast, AI native founders may not initially know the industry, but they can acquire that understanding if they are willing to immerse themselves in it and spend sustained time with customers, experts and partners.
Of course, the ideal combination remains:
An AI native founder and an experienced industry co-founder.
One understands technology; the other, the industry.
One takes responsibility for rebuilding capabilities; the other, for the path to implementation.
Such teams are more complete.
What Makes Shidai Bole's Model Distinctive Is Not Just Investing, but Making Collaboration Part of Growth
To view Shidai Bole as an ordinary investment institution would be to underestimate it.
It is neither a typical purely financial VC nor the CVC of a single listed company.
It looks more like a shared industrial-investment platform connecting the resources of multiple listed companies.
That means it can offer projects more than funding.
It can provide use cases, supply chains, customers, channels, brand endorsement, management experience and potential future capital-market paths.
Why does this matter?
Many AI startups are not stuck because they cannot build a product.
They are stuck because, after building it, they lack real-world validation, reference customers, industry resources and supply-chain capabilities, leaving commercialization perpetually at an early stage.
Those are precisely the areas where an ecosystem of listed companies can help most.
The support Bu Liangyuan describes falls into several layers:
First, support for deployment in real scenarios.
Help companies access genuine business settings and complete commercial validation from 0 to 1.
Many AI companies have good technology but cannot secure their first reference customers. Without crossing that threshold, progress becomes difficult.
Second, supply-chain and industry-chain support.
Help companies move from a working example to delivery at scale.
There is a long chain of work between being able to build a product and being able to deliver it reliably.
Third, support with capital and exits.
Backing from a listed company's industrial investment can itself improve a company's standing in subsequent fundraising.
At a more mature stage, a listed company may also provide a route to an acquisition exit.
Fourth, support with strategy and management.
Established companies have learned from many mistakes in organizational governance, operations and risk control. That experience is valuable to early-stage startups.
Taken together, these explain his repeated emphasis on collaborative innovation.
For most application-layer AI companies, technology is only the starting point.
The real difficulty is connecting it to an industrial system.
One Example Makes the Point: Many Application-Layer AI Companies Are Better Suited to Growing Through Collaboration
Bu Liangyuan cited a case involving Stardust Intelligence and Jinma Amusement.
On one side was Jinma Amusement.
The first A-share-listed company in China's amusement-equipment industry, it understands theme parks, cultural tourism, supply chains and safety standards deeply.
On the other was Stardust Intelligence.
An AI native team with strong robotics technology, but almost no experience in cultural tourism and no existing industry customer base.
Each team had obvious weaknesses if it worked alone.
Jinma Amusement lacked the ability to reconstruct its offering through AI.
Stardust Intelligence lacked use cases, customer access and industry know-how.
Once the two were connected, the path became clear:
Jinma Amusement would provide deployment settings, industry experience, product-requirement definition and supply-chain support;
Stardust Intelligence would handle technical R&D, product design and project delivery;
and an industrial fund and joint R&D center would align their interests in practice.
The result was not a superficial partnership, but a functioning end-to-end system.
For example, "Robot MART" combined the operating capabilities of cable-driven robots with theme-park consumption scenarios, increasing foot traffic at an individual store by 30%.
The most valuable point of the case is not its novelty, but what it demonstrates:
For many application-layer AI startups, growing independently may not be the most efficient path.
They often need technology, use cases, supply chains and commercialization capabilities to come together.
That is where cross-generational collaboration truly becomes meaningful.
The Biggest Opportunities in 2026 Still Lie Where AI Reaches Deep into Industry
On opportunities, Bu Liangyuan's view was clear:
The largest opportunities still lie in deep integration between AI and industry.
Behind that statement is one of the Chinese market's most distinctive characteristics.
China has the world's most complete industrial system.
Every vertical industry contains many processes that have never been fully digitized or fundamentally rebuilt.
Those processes leave extensive room for AI-driven change.
That is why he will continue to devote most of his attention to the application layer in 2026.
He will examine infrastructure, but will not place his heaviest bets there.
Broad, general-purpose narratives are becoming more expensive; those actually delivering growth are putting AI into specific settings.
Beyond applications themselves, he is optimistic about multimodal generation and interaction, physical AI, robotics models, world models and new forms that combine AI with smart hardware.
On text-focused fields, his view was blunt:
There are not many opportunities left.
That does not mean text is unimportant.
It means this layer is already crowded.
The space remaining for new companies increasingly lies where AI enters the real world, complex workflows, perception and execution.
The barriers in these areas are no longer just models.
They also include adaptation to use cases, engineered delivery, hardware-software coordination, data feedback loops and industry resources.
Those are precisely the areas where many startups still have a chance.
Which AI Companies Will Capital Reprice in 2026?
By the end of the conversation, I felt the answer was clear.
The companies repriced next will not simply be those with the strongest technology.
They are more likely to be companies of this kind:
They have identified a real problem;
they serve a real customer;
they deliver a real incremental improvement;
they have begun generating real revenue;
and they live on business activity, not market sentiment.
That is also what Bu Liangyuan meant in his closing message to AI founders:
"
"Return to business fundamentals, deepen your contribution to industry, and navigate AI's cycles with real products, real demand and real revenue."
Put more plainly:
From 2026 onward, the market will reward companies that merely look like the future less often, and companies that have already turned a little of that future into revenue more often.
This is not the market becoming conservative.
It is a sign of an industry beginning to mature.
A genuinely mature industry will not price imagination alone forever.
Sooner or later, it returns to harder questions:
Is anyone actually buying what you have built?
Can you deliver even a small part of the future you describe?
That may be the most concrete change in AI investment in 2026.
Selected Q&A
Q1: What is the central change in AI investment in 2026?
A shift from technology first to deployment first.
Previously, attention centered on the team's technical strength and whether it could build a product. Now, it centers on paying customers, real revenue, a viable commercial cycle and whether positive cash flow can be achieved within 12 to 18 months.
Q2: What matters most when you evaluate AI projects?
Four dimensions:
Genuine demand in the use case,
a genuine technical moat,
a genuinely well-matched team,
and a viable commercial cycle.
We will not invest in projects that simply wrap a general-purpose model without a moat.
Q3: What is your most important internal question?
One question is particularly important:
"Can the problem this product addresses be solved without AI?"
If it can easily be solved without AI, and adding AI delivers no clear improvement in value, the project's foundation is unstable.
Q4: Which founder capabilities matter most?
The ranking is:
Industry background > business background > technical background.
AI has entered the stage of industrial deployment. Whether a founder understands industry pain points, customer logic and the route to commercialization matters more than a purely technical background.
The ideal combination remains an AI native founder and an experienced industry co-founder.
Q5: Which projects are rejected quickly?
Two types:
Those whose founders lack a commercial mindset and are disconnected from the market;
and those in already crowded fields with established leaders, where the team cannot demonstrate a differentiated advantage.
Q6: Do startups still have opportunities after major companies enter?
Yes.
The key is finding your position, not confronting them head-on:
Serve specialized scenarios that major companies do not want to address, cannot address well, or cannot address cost-effectively.
Startups still have room in vertical-industry know-how, data, customization and response speed.
Q7: Do you prefer industry veterans learning AI or AI native founders learning an industry?
AI native founders learning an industry.
AI technology changes so quickly that traditional entrepreneurs often struggle to keep up. Founders who understand AI have a better chance of acquiring industry knowledge if they are willing to immerse themselves in the sector and spend sustained time with customers.
But combining the two remains the best solution.
Q8: Will you bet more on the application layer or infrastructure in 2026?
The application layer.
Bu Liangyuan's position is clear: 90% of his attention and investment allocation will continue to go to applications.
He is particularly optimistic about AI combined with vertical industries, multimodal generation and interaction, physical AI, robotics models, world models and new forms of AI plus smart hardware.
Q9: If AGI is truly achieved, will investors' value be replaced?
It will not be replaced, but it will be reconfigured.
AI will certainly handle more repetitive, standardized information organization and analysis.
But investors' real value remains in industry judgment, assessing founders, integrating resources and making decisions under uncertainty.
Those displaced will be investors who merely relay information, not those with genuine judgment and resources.