---
title: "True Localization for AI Going Global Is Not Registering a Company, but Accepting “No Replies on Weekends”"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2025-12-12T08:01:15+00:00"
canonical: "https://ffcap.cn/en/research/src-20251212-01html"
source: "https://uniqueresearch.substack.com/p/src-20251212-01html"
language: "en"
---

# True Localization for AI Going Global Is Not Registering a Company, but Accepting “No Replies on Weekends”

_Original · Unique Research · 2025-12-12_

_Historical edition note: This is the complete rendition of Unique Research's December 12, 2025 article. Company figures, cultural generalizations, pricing examples and regulatory interpretations are the speakers' claims at that time, not current measurements or universal legal requirements. The 76-year logistics experience is the source's description of accumulated business experience, not an independently verified founding date for Cuber AI. Beta Data is a source-based English rendering of 贝塔数据. The ordinary roundtable photograph identifies the venue as Developer Space at Google Singapore and the session as Roundtable 2 on Chinese companies expanding overseas; the panelists and roles are already identified below. It adds no unique operating data and is not reproduced._

If you plotted today's Chinese AI entrepreneurs on a world map, you would notice an interesting phenomenon: everyone talks about "North America," "Europe," and "globalization," yet when it is time to make a real move, a considerable number first press the "confirm" button in Singapore. At this roundtable on "How Chinese Companies Can Break Through Overseas," the moderator was Unique Research CEO Wu Wei. Opposite him sat three representative types of global operators: Gao Shouzhi, CEO of EntGroup, who supplies "data fuel"; CK Ng, Cuber AI Managing Director, who works deeply on Agentic AI in logistics and industry; and Li Shouguo, CEO of Beta Data, who uses AI to coach sales communication.

The three companies have entirely different businesses, yet they resemble three mirrors reflecting three fundamental questions for Chinese companies going global:

Where to go, how to establish themselves, and why they should win.

The true overseas barrier for a Chinese company may not be a feature or relationship, but its ability to understand people, markets, and time anew.

I. "Where to Go Global" Is Not a Geography Question but a Product-Time Choice

When many people plan global expansion, a world heat map first appears in their minds: where is the market largest, average contract value highest, and willingness to pay strongest?

All these questions sound right. After a full round of research, however, you discover a harsh reality: the reddest places on the map are often where you are most likely to fall into a trap.

When Gao Shouzhi described EntGroup's choice of its first destination, he did not begin by calculating "who has the most money." He worked backward from two dimensions:

First, how well does this market match my existing product?

Second, do my channels, talent, and cultural understanding here give me a starting-line advantage?

Japan is a classic counterexample.

Its market is substantial and individual customers are large, but building a real business may require "one year to land one order and two years just to validate the product." In the AI era, however, products iterate weekly: while you spend two years validating something overseas, the domestic market may already have iterated through three or four versions.

By the time you finish navigating the process, the need itself has changed shape, naturally distorting the validation result.

EntGroup's final choice was therefore to use Singapore as its first destination—a real-world testing ground for products going global.

Its location is central, its culture relatively neutral, its regulatory system clear, and it has a certain density of AI customers and large-model companies.

Simply put, it may not be "the fattest piece of meat," but it is the step Chinese teams can most easily cross when venturing abroad.

CK put it more directly: starting from Singapore, he would consider only Malaysia as the second destination and Indonesia as the third. The reason is not which country is more fashionable, but which is "more like China": a broad geography, complex logistics, pragmatic customers, and a willingness to pay for solutions that genuinely save money and improve efficiency.

Markets such as Thailand and Vietnam, meanwhile, are addressed later and more gradually because of higher language barriers.

This yields one conclusion that is easily overlooked: the first overseas destination is not "the sexiest one," but "the place where you have the best chance of beating time."

Does market size matter? Yes.

But during a window when both technology and demand are changing rapidly, whether you can complete in one year the validation that others need three years to obtain matters even more.

II. Localization Is Not "Registering a Company" but Replacing the Organization's Operating System

"We have already established a company in Singapore." This sentence appears so frequently in project roadshows that people have begun assuming an overseas registration means global expansion is complete.

The guests' meaning of "localization," however, was entirely different.

For data-service provider EntGroup, the first wall in localization is compliance. Cross-border data is one of the easiest areas in which to trigger serious risks.

You must provide large models with fresh data at "submarine-cable scale" while simultaneously facing platform lawsuits, privacy regulation, cross-border storage, and a chain of other issues.

One careless mistake, and the business you painstakingly built in a market may be eliminated because you did not fully understand one regulation.

Gao Shouzhi's judgment is therefore that, when serving business customers—especially as a supplier of "data fuel"—you need local customer-service and compliance capabilities online from your first day entering the market.

The R&D team can remain in China, but interaction and service must be local.

ISO certification, privacy protection, and local data laws are all unavoidable.

CK described another hard reality of localization—culture and ways of working.

Chinese companies are accustomed to being "always online" and "voluntary weekend self-improvement." In many countries, however, a weekend is simply a weekend, and not replying to email is normal. When hiring, he can only earnestly describe weekend study as "self-improvement" and then add half-jokingly, "This is not overtime."

The room laughed, but behind the joke lies a serious question: will you forcibly pull overseas employees into a Chinese pace, or accept the reality that "the pace slows, but trust becomes steadier"?

At a deeper level, localization also means "simultaneous reconstruction" at both the product and institutional ends.

CK's company already had 76 years of logistics experience and builds a vertical supply-chain large model and an Agentic Digital Employee.

The same product may emphasize "maximum efficiency and dense functionality" in its Chinese version, but overseas it must redesign permissions, compliance rules, and even UI complexity while complying with international frameworks such as PDPA and GDPR.

This is not a matter of simply translating an English version and going abroad, but genuinely rewriting an operating manual "suitable for a multinational regulatory environment."

Li Shouguo simply divides his team's structure in two: local frontline sales and customer-success teams handle deals, scenario refinement, and relationships, while engineering and delivery remain in China and use the country's engineering advantage for rapid iteration.

Between them is a "partner mechanism": the company must serve key early customers itself; after the product and go-to-market method stabilize, contracts and delivery can be entrusted to local partners.

In short, localization has at least three layers:

The legal and compliance safety line

Reconstruction of organizational pace and ways of working

Redesign of product form and business model

Registering a company is only the prologue. The real difficulty lies in "installing the company inside the local society's system of rules."

III. What Is Truly Difficult to Copy Is Not Technology but the Combination of "People + Scenarios"

This discussion produced a remarkably consistent consensus: every guest works in AI, yet none believes that "AI will consume everything."

CK put it bluntly. In logistics, AI can automatically read orders, process tedious data, and take over 80% of mechanical labor.

But the final decision about "which shipping company to book" may come down to "I had drinks yesterday with the person in charge at Maersk, so I will give him a little more business over the next two weeks."

Behind that decision are relationships, trust, and emotion—variables that any model finds difficult to grasp completely.

He therefore positions the product as a "digital employee": machines connect the threads, linking data flows among systems and freeing people from endless repetitive labor to handle matters genuinely tied to money, judgment, and relationships.

Beta Data, Li Shouguo's company, goes further by writing "people are irreplaceable" into its product philosophy.

Sales ability resembles swimming or riding a bicycle: reading a hundred books is less useful than spending ten minutes in the water.

Traditional training for large-enterprise sales uses real customers for practice, but that is both expensive and dangerous. One poorly handled interaction can consume an opportunity cultivated over several years in a single "practice session."

Their product uses AI to simulate all kinds of difficult, demanding, and even emotional customer scenarios, allowing salespeople to be "educated through setbacks" in a safe environment and repeatedly practice communication, negotiation, and closing.

What he truly believes is not "AI replaces salespeople," but that with AI, "people training people" can become more efficient and reproducible.

One detail is particularly interesting: although the team could easily build a system in which "AI automatically sells for you," it deliberately chose the less glamorous direction of "practice partner."

There is only one reason: complex sales is fundamentally about emotional connection and solutions, not a click or an order.

You can use AI to lower training costs and help a newcomer grow in one year into a salesperson who previously required three years of development. But using AI to replace face-to-face trust building, subtle perception of emotion, and the moment someone finally relaxes because "you understand my difficulty" is far harder.

This also explains an apparent contradiction: at conferences we discuss "agents," "Agentic AI," and "digital employees," yet still insist on holding in-person forums, meeting face to face, and drinking coffee, even though every piece of content could be translated into a hundred languages and uploaded to the cloud.

It is because we all know that the invisible line between people—not lines of code—is what truly crosses languages, cultures, and borders.

IV. A Partner Is Not a "Channel" but Your "Second Brain" in the Local Market

At the more detailed operational level, global expansion contains another frequently underestimated hurdle: should you "win deals yourself + deliver yourself," "win deals yourself + entrust delivery to partners," or "entrust everything to partners" from the outset?

Li Shouguo offered a practical answer: overseas, he wins early customers himself, then hands delivery and long-term operations to partners after signing.

The reason is simple: for the first few orders, refining the product and validating demand matter most, and partners are far less sensitive to both than the founding team.

If you entrust the most critical firsthand feedback to a partner, by the time information passes through several WeChat groups and rounds of meetings before returning to the product team, it is often distorted and far too slow.

Only after the product logic runs smoothly is it time to entrust delivery and expansion to local partners.

How, then, should partners be selected? He uses only two very simple but effective indicators:

First, does the customer have a sales team of more than 50 people?

Second, does it have at least "half a person" dedicated to sales training and development?

Meeting both conditions shows the company genuinely builds "sales capability" as an organizational capability rather than treating people as disposable consumables.

Only with such customers and partners can the value of a sales-practice system be amplified.

This approach can, in fact, extend to most companies going global:

For the first 3–5 orders, the founding team must participate personally. The purpose is not merely to "win the order," but to understand the relationship among the market, customer, and product.

Beginning with order 6, partners can start applying leverage. Only then do you know what should be "standardized and handed to them" and what must remain firmly in your own hands rather than outsourced—such as core algorithms, critical data, and pricing power.

When you view a partner as a "second sales department," it will always be a cost.

When you view a partner as a "second brain," it can become a genuine asset for your global expansion.

V. Chinese Companies' Global Advantage Is Not "Cheaper" but "Faster + Harder"

When the conversation turned to competition, it inevitably reached "China-US confrontation" and "how to compete with mature Western players."

Chinese companies can win opportunities in global AI applications today not because they are cheap, but because of two things—speed and scenarios.

Li Shouguo's memorable "provocative claim" was: when building application products in the AI era, speed surpasses everything; there is no second place. One minor release a week and one major release every two months is the baseline configuration.

This pace rests on China's engineering advantage and the extraordinarily brutal competitive environment of China's market.

Surviving domestically already shows that you have passed the threefold test of massive user volumes, complex scenarios, and frequent feedback.

CK offered another intuitive analogy from logistics: "However complex Southeast Asia may be, it cannot be more complex than China."

China's logistics system combines extremely low prices per shipment with extremely fast fulfillment and exceptionally diverse business models.

A system trained in this "pressure cooker" often looks overpowered when its language and cultural layer is replaced and it is applied in another country.

A subtler advantage is that China's market is itself a combination of "multiple tiers, multiple cultures, and multiple stages of development."

Small-business lending in Jiangsu and Zhejiang is already highly automated, while some northwestern regions still use traditional relationship-driven selling by gathering people and shouting. A single product serving both customer types must accommodate needs at different levels of maturity simultaneously.

What does this mean? When you go abroad, many supposed "market differences" have similar versions you have already encountered in China.

China is not a single market. It is more like a miniature "world," except that everyone uses the same language and currency.

The pricing logic of Chinese companies is also quietly changing.

CK said he does not play the price-cutting game in Singapore: "Whatever numerical price we charge in renminbi domestically, I charge the same number in US dollars overseas."

That sounds exaggerated but makes sense on closer inspection:

AI that genuinely solves problems is never a "cheap tool," but a "restructurer of costs";

Once the efficiency forged through domestic competition works, it can be converted into a reasonable premium overseas.

A low price is only a bargaining chip when you begin; it cannot become your identity.

What truly supports a durable overseas position is whether you can iterate at a speed others find difficult to copy, train models on complex scenarios others lack, and within a reasonable period translate those capabilities into value the local market understands, trusts, and will pay for.

VI. From "Going Out" to "Surviving, Then Thriving"

If this conversation were compressed into one sentence, it would be: global expansion does not merely add another market; it forces you to rewrite "how a company operates."

You must answer many seemingly simple questions again:

Where to go first is determined not by maps and trends, but by "product and time";

Localization is not changing the language and establishing a company, but truly placing your organization, institutions, and product within the local system of rules;

The advantage AI creates is not "eliminating everyone," but freeing people from low-value labor to do work more worthy of their time;

Partners are not your "outsourced sales team," but a second brain grown in another country;

China's advantage is not "more aggressively competitive pricing," but "faster iteration + validation through harder scenarios."

Seen this way, "breaking through" in global expansion also changes meaning. It is no longer merely "how to defeat a competitor and win a market," but more like how to make capabilities forged in China understood and recognized anew around the world.

Perhaps when we look back several years from now, we will find that the product versions now being tested on a small scale in Singapore, Malaysia, and Indonesia were the early drafts of the next generation of "Chinese global companies."

That is also why, even though everyone was working on the most advanced AI and the smartest agents, Wu Wei still closed the roundtable with a slightly playful line: "Believe in China."

The belief is not only in a country, but in an entire generation: people who dare to take products, engineers, and speed tempered in China into an unfamiliar world, stumbling forward while learning to rewrite "how to do business with the world."

The stories you and I are reading at this moment may be the "mental exercises" most worth rereading before you take that future step.

Selected Roundtable Q&A

Q1: First, please introduce yourselves and your companies in one or two sentences. What is each of you doing in relation to taking AI global?

Gao Shouzhi: I have worked with data for twelve years, from the small data of the early internet, through big data, to today's work "feeding data" to large models. In my view, computing power, algorithms, and data are the three engines of the large-model era, and we supply part of the "fuel."

To use a film analogy, we give customers fresh data at "submarine-cable scale"—like someone who travels into the future in a science-fiction film, plugs in a network cable, and instantly reads the present-day world. For large models, our work resembles the "feeding" in Lucy: continually using various human data to expand their cognitive boundaries, bringing them closer to a general-purpose OS or even an "ubiquitous" system.

CK: I have worked in this industry for 28 years, always in data and technology. Cuber AI has dual headquarters: its international headquarters is in Singapore and its domestic headquarters in Shanghai. We build Agentic AI for industry and logistics, also understood as "digital employees," helping enterprises entrust large volumes of tedious and difficult-to-organize work to agents.

Simply put, we do not seek to replace people. We want machines to eliminate 80% of repetitive labor and leave the remaining 20% for human decision-making and relationship management.

Li Shouguo: Beta Data is building an AI practice-partner system for sales-communication skills. Sales resembles swimming or riding a bicycle: listening to any amount of theory is less useful than practicing for real. Traditional training for large-enterprise sales uses real customers for practice, which is both expensive and dangerous. We use AI to simulate complex and demanding customers, allowing salespeople to be "put through the wringer" and improve in a safe environment. We do not believe "AI consumes everything"; we believe in "people developing people," with AI making the process more efficient and reproducible.

Q2: Mr. Gao, you provide the "data fuel" of the large-model era. What are the opportunity and danger in this business?

Gao Shouzhi: The opportunity is that as large models become stronger, their need for high-quality data grows and becomes increasingly refined and scenario-specific. Traditional data annotation was understood more as a human step in the "pre-large-model era." We have now entered the era of the "AI trainer"—someone must understand the business and the model, then connect the two through data.

Commercially, the wealth-creation effect in this industry is extremely visible. Consider OpenAI and the AI data and training companies created by newly wealthy young ethnic-Chinese entrepreneurs: much of their value rests on the chain of "turning data into model capability."

The danger is also real. Particularly in law and compliance, data is an industry that "constantly dances along the edge of a gray zone."

A typical example is a company sued for scraping platform data even though those same platforms are its customers. That illustrates the industry's current complexity: legal disputes persist, yet the entire value chain cannot function without such data services.

Q3: CK, you build Agentic AI in industrial and logistics scenarios. How specifically do you create value for customers?

CK: Logistics has one highly typical pain point: many systems, long processes, and highly fragmented information. You may already have purchased the best TMS and FMS, yet still need substantial human labor to "thread the needle" between them, copying data from one system to another.

Our Agentic Digital Employee is the digital employee that replaces this entire stretch of "manual threading."

For example, suppose you receive an order containing dozens of pieces of information. Traditionally, each must be entered into the system individually. At hundreds or thousands of orders a day, people inevitably become tired and make mistakes. Our digital employee can automatically identify the order's contents, call our proprietary vertical large model (QLM), combine it with more than 70 years of know-how accumulated in supply chains and logistics, and automatically provide a complete plan spanning quotation, booking, and customs clearance.

We currently have more than 2000 customers, and the product has evolved from its original RPA form through to today's fourth-generation Agentic AI product—called "Xiaoda AI" in China and Agentic X overseas.

I always emphasize that AI has not come to eliminate people, but to let them do less low-value work and spend time on activities that create more revenue, instead.

The final decision about "which shipping company to book" may still return to a person: who is your long-term partner? With whom did you just share drinks and whom do you want to support a little more? These are the boundaries of large models and the value of people.

Q4: Mr. Li, in sales, marketing, and customer service, you chose to "build a sales practice partner" rather than "use AI to replace salespeople directly." Why?

Li Shouguo: Simply put, the simpler the sales scenario, the more easily automation replaces it; the more complex the sale, the more it depends on people.

In complex sales, customers are not buying a feature, but whether their problem can be solved, whether the solution has been understood seriously, and whether the service is trustworthy. At a deeper level, they are buying an emotional connection.

Consider medical aesthetics, which is fundamentally an emotionally driven business. What customers truly need is the feeling of being understood and respected behind the desire to "become more beautiful," not the skin-booster injection itself.

Current large models cannot fully replace these elements. Why are we sitting in person today for a Panel instead of holding an online livestream and letting a large model automatically translate it into more than a hundred languages? Because face-to-face meetings, eye contact, and tone form the foundation of trust and emotion.

Therefore, in what we define as "complex sales," people cannot be replaced in the short term. Since people will continue to exist over the long term and create extremely high value, helping them improve their skills is a long-term business.

Q5: For global expansion, the first question is "where to go." How did you select your first destination and priority markets?

Gao Shouzhi: If I had to put the factors in order, first I would examine which market matches your product best; only second would I consider channel resources and the team's inherent strengths.

Japan is a highly typical market that "looks beautiful" but severely tests patience.

Validating a product from the first customer until it genuinely runs smoothly may require one to two years. In the AI era, product iteration happens "by the week." By the time you complete validation two years later, the market already contains several new product generations. This time cost is fatal for AI applications that require rapid iteration.

From a global perspective, the United States is certainly a market we must win, both because of its scale of investment and because AI has been elevated to the strategic level of a "contest over national destiny."

CK: We begin in Singapore and then look at Southeast Asia. Malaysia must be the first destination.

The reasons are straightforward:

Geographically, it is closest to Singapore;

Its industry structure contains both international logistics and a large domestic logistics market;

Its customer base resembles China's: pragmatic and willing to pay for visible efficiency.

Malaysia has another characteristic: industry associations are highly influential. Serve one association well, and an entire customer group may stand behind it.

The second destination is Indonesia. Its market is large and opportunities abundant, but it also contains some "under-the-table" complexity that companies must evaluate themselves. I previously worked at a US company and later at a Singaporean state-owned enterprise, so I am relatively sensitive to anti-corruption issues. That also affects how we choose to operate and at what scale.

For markets such as Thailand and Vietnam, the largest problem is language. Without a strong command of Thai or Vietnamese, it is difficult to penetrate the market genuinely. By comparison, operations can begin in Malaysia and Indonesia using English, making them better second and third destinations.

Q6: Localization involves more than "registering a company"; it includes systemic issues such as law, employment, and channels. What pitfalls have you encountered, and what lessons can you share?

Gao Shouzhi: For a company like ours providing data services, especially business clients, virtually every pitfall comes down to three words: lack of familiarity.

Lack of familiarity with local regulation and legal boundaries makes it easy to cross red lines in data compliance;

Lack of familiarity with local business scenarios and decision mechanisms makes it difficult to understand what customers truly need;

Lack of familiarity with local transaction practices makes it difficult to design a form of collaboration the other party will accept over the long term.

In data businesses, for example, many countries and regions have complex rules governing privacy protection, cross-border data, and storage locations. You cannot wait to learn after receiving your first penalty. ISO systems, privacy certifications, and data-governance principles must be built to the target market's standards from the beginning.

My current view on team configuration is clear: a local customer-service team should exist from the first day you enter an overseas market.

R&D and delivery can rely on Chinese engineers, but the customer-facing layer must be local. Otherwise, you may think you understand the need when you have only understood its translated version.

CK: What impressed me most was "cultural differences in the pace of work."

Chinese companies are accustomed to remaining online on weekends, with many colleagues "voluntarily working overtime" on Sundays. A friend says, "Let's optimize it a little more," and everyone keeps going into the evening.

Outside China, however, Saturday and Sunday are weekends. Customers may not answer the phone, and employees do not treat "responding at all times" as self-evident.

When hiring in Singapore, we therefore earnestly explain to candidates that we do not call it "overtime," but "self-improvement" and "learning time." People laugh, but in practice this realigns mutual expectations about pace.

From the product and compliance perspectives, the greatest pitfall is that you must reshape your product according to international rules.

Our domestic Xiaoda AI version has its own permissions and user rules. In Singapore, it must fully align with PDPA and GDPR requirements alongside the globally recognized ISO system.

The implementation team must also be localized, because language and culture directly affect how smoothly a project advances.

Q7: Mr. Li, how do you combine frontline teams, local partners, and the business model in global expansion? How do you find the first group of reliable partners?

Li Shouguo: Our approach is:

Place the frontline customer team locally whenever possible, keeping it close to customer scenarios while refining the product;

Continue relying on China's engineering advantage for engineers and technical delivery;

Use local partners whenever possible for overseas contracts and long-term service.

I enforce one principle strictly: we must serve key early customers ourselves.

There are two reasons:

First, you need to experience overseas customers' real needs, usage habits, and attitudes toward budgets firsthand. This is essential to refining the product.

Second, partners will never be as sensitive as you are to "details of demand." If every piece of firsthand feedback is relayed through them, information loss will be substantial.

After you and several key customers have stabilized the product, pricing, and service model, it is safer to hand contracting authority and large-scale promotion to partners.

How do we find partners?

Our screening criteria are very straightforward:

The customer must be a medium-sized or large enterprise with a sales team of at least 50 people;

The company must have a dedicated employee, or "half an employee," responsible for sales training and capability development.

These two conditions demonstrate:

First, the customer has sufficient scale for trained capabilities to create leverage;

Second, it genuinely treats "sales capability" as an organizational asset rather than a part that can be replaced at any time.

We are currently seeking partners in Singapore, Malaysia, Thailand, and Indonesia that value sales training and have stable sales teams.

Q8: How do your overseas pricing and business models differ from those in China?

Li Shouguo: The main model today charges by license, but in China we have already begun experimenting with usage-based billing and will later bring this model overseas.

The reason is simple:

Customers with many SKU items and complex scenarios have high usage and receive high value, making usage-based billing fairer;

Customers that train new employees only during specific stages and have low total usage may feel an annual fee is a "loss," while usage-based billing is easier for them to accept.

Simply put, we want our growth to "breathe together" with the customer's: the more they use and depend on the product, the more we earn; if they use less, our costs are also lower.

CK: In Singapore I use a relatively "anti-rat-race" approach: whatever numerical price we charge in renminbi domestically, we charge the same number in US dollars overseas. It sounds like a joke, but the logic is serious. An AI solution that genuinely changes cost structures and efficiency should never be sold as a "cheap tool."

China's environment is excessively competitive and everyone is accustomed to price wars, but overseas customers care more about:

Can you show me visible ROI?

Can you make a complex scenario genuinely run smoothly?

Given those conditions, a reasonable premium is accepted. We would rather turn the "efficiency and scenario experience forged through domestic competition" into overseas price and value advantages than continue playing the game of "who is cheaper."

Q9: As US companies also expand globally and may already be mature in your fields, what fundamental advantages do Chinese companies have overseas? How can those advantages become real competitiveness?

Li Shouguo: I have a "provocative claim": when building application products in the AI era, speed surpasses everything; there is no second place.

More specifically, a minor release every week and a major release every two months is the baseline pace.

At this pace, you do not need a "perfect design" but a feedback chain that can test and correct errors rapidly. China's engineering advantage is unique worldwide in this respect.

The second advantage is the complexity of China's market itself.

We often say that "China's market is large," but more importantly, its internal differences are enormous:

Small-business lending in Shanghai, Jiangsu, and Zhejiang is already highly automated;

Some northwestern regions still conduct business by gathering people and shouting, driven by personal relationships.

A single product serving both types of customers is forced to possess an extremely broad range of adaptability.

After your product has been refined for two or three years in this "ultracomplex single market," you will encounter many apparently novel scenarios in Southeast Asia that already have "shadow versions" in China.

Gao Shouzhi: I agree that "speed" and the "engineering advantage" form the first layer.

In data services, we have another advantage: experience providing services deeply integrated with leading large-model companies.

The earlier and more deeply you participate in these companies' training and alignment, the sooner you see subtle changes in technical directions and sense a new generation of needs, allowing your product to move "half a step ahead."

US companies currently lead, but no company can capture the entire market. We can see this in OpenAI surpassing Scale AI, which had existed for more than a decade, within five years: once a technological inflection point arrives, market positions are not necessarily assigned by seniority.

_Editorial clarification: The preceding comparison is Gao Shouzhi’s original claim, retained in full; the source does not identify the metric meant by “surpassing.” The company-age description conflicts with [Scale AI’s official history](https://scale.com/about), which gives its founding year as 2016—about nine years before this December 2025 article, not more than a decade._

For Chinese companies, the key is to seize the window of the next several years and package "engineering advantage + scenario depth + iteration speed" into genuine global competitiveness.

CK: I would add that China is an excellent "incubator for products."

In logistics, for example, China's complexity and competitive intensity are difficult for many countries to replicate in the short term or even the long term.

Customers in East China and North China may place more value on "complete solutions" and "city-level applications";

Many customers in South China are highly pragmatic: "If it does not make money, do not do it; if it does not save money, do not do it either";

The extreme fulfillment efficiency of completing delivery for five yuan is entirely unimaginable in many overseas markets.

After refining a product in this environment, you can take mature Chinese cases to Southeast Asian customers, who will strongly buy in.

Many overseas customers ask: "Your product is so new—has it been validated in complex scenarios?"

My usual answer is: "Do not worry. Nothing can be more complex than China."

In pricing, excessive domestic competition instead gives us an opportunity: when you move efficiency and capability compressed to extremely low prices in China into a market willing to pay for genuine efficiency, you discover that your product has ample confidence to discuss value and profit.

---

Original publication: https://uniqueresearch.substack.com/p/src-20251212-01html
On-site reading page: https://ffcap.cn/en/research/src-20251212-01html