Original · Unique Research · 2026-06-18
Editor's note: The first-person report and its judgments belong to the original Chinese author (hosted by Zhu He). This English rendition retains the roundtable narrative on fake vs. real globalization, reasons to leave, market selection, four entry playbooks, and closing, plus the full verbatim Q&A. All panelists, companies, and figures are preserved. Founder statements and cited numbers are source attributions, not independently verified findings.
AI Industry Observation
The besieged fortress of going global: those inside want out, those outside want in
Real globalization isn't fleeing one market, but first breaking one market open
"
The so-called besieged fortress isn't a philosophical dilemma of some wanting in and some wanting out; it's a very concrete question: that hammer you forged at home — where else can you still find nails?
On the Singapore AI Agent Summit stage, four people leaned in exactly opposite directions.
A founder who came to Singapore at 15 and has been abroad for over twenty years said: doing globalization this year, you cannot leave the China market. Beside him, a founder who did VC for over a decade domestically said: stepping out of mainland China is a very rational need.
The host Zhu He couldn't hold back and called it out on the spot: "a Singaporean who came over at 15 saying he wants to enter the domestic market — that's real globalization; someone at home saying you can't do it domestically and must go out — that's just the besieged fortress."
This isn't manufactured drama. This is the real situation of doing AI going-global.
What Counts as an "Instantly Fake" Globalization
FOSHO co-founder Tony Sun (from Honor's globalization team) put it in plain language: "doing sales overseas isn't called globalization."
What does he mean? He's referring to many companies' first going-global step: translate the product to English, buy Facebook ads, list on Amazon, then declare they've entered the global market. This playbook may have worked before, but today it basically amounts to opening a sales window overseas, not taking root.
His single test: local consumers don't feel it's a Chinese company — only then has localization succeeded. Product localization is only the most basic layer; branding and channels must localize too, with media partnerships, influencer marketing, and social-media matrix all laid out. Relying on a single channel, ROI will only keep falling.
Mao Hua (founder of 2nd Brain, in Singapore his seventh year, previously eight years at Tencent where Yingyongbao was the product he built) added a B2B view. He says consumer products going global are relatively easy, with low cultural-difference requirements; but B2B is extremely hard — every country's policy, culture, and data-security requirements differ.
He got stuck on this himself. Back then his Singapore company needed a collaboration tool; WeCom had no international plan, so he chose Feishu. Later Feishu did data isolation; Singapore employees downloaded Lark, and once isolated they couldn't log in — all account data cut off. "Going to Feishu president Zhang Nan didn't help either; the only option was having employees switch their App Store country to China and re-download Feishu." A veteran internet hand getting tangled just switching office software — this is the real scene of B2B going-global, not what PR copy describes.
The Reasons to Leave Are Sound, but Sound Doesn't Mean Easy
Li Yinan did VC for over a decade domestically, investing in twenty to thirty companies. He saw two structural problems that make domestic entrepreneurship painful.
The first is procurement-side malice. Once a domestic KA client thinks your product works, their first reaction isn't to renew, but to have someone clone it. The software business can't avoid large clients, but when large clients treat you this way, the ceiling is right there.
The second is a term he coined — "red-basin competition." Everyone lives well in a sector, suddenly a hundred competitors pour in, driving prices to 1% or even 0.1% of the original, from a few players to fifty, until they all die hand in hand. Doing SaaS or software on one leg carries extreme risk.
These two stacked together make going abroad a very rational impulse.
But then he said: once you're out, you'll find more than half of what's been validated at home simply can't be migrated.
He decomposed a product that ran domestically into four layers: market demand exists, product supply is formed, the team can execute, customer channels can push to market. Moved overseas, nearly every one of these four layers needs re-validation. Demand may not exist abroad — bike-sharing and power-bank sharing exploded domestically but have basically no market overseas. Products often don't acclimate; excellent domestic teams lack international-market contextual understanding; domestic channels are useless abroad.
Going abroad is rational, but turning rationality into results is another matter.
How to Choose a Market: Not Planned, but Bumped Into
Agnes AI founder Yang Tong (Bruce, came to Singapore at 15, NUS AI PhD) gave a judgment updated this year, the most counterintuitive in the room.
Last year Agnes had 95%+ of users overseas, in Southeast Asia mainly Indonesia, in Latin America mainly Brazil, plus the North American market. That's already a very "going-global" company. But his conclusion this year: you can't lose the China market.
His logic: AI has entered the 2.0 era; starting this year mainstream attention is fixed on two blocs — China and the US. The speed and depth of AI adoption in these two markets outpace other regions. Doing well only in Southeast Asia, you're still considered uncompetitive in mainstream AI circles. Agnes's daily API Token consumption has already reached 200 billion, possibly 1 trillion next month — without Chinese developers' enthusiasm, this scale is impossible.
He gave DeepSeek as an example: "before it blew up domestically, it was already slowly building a voice in the US market." First go deep in a single market, then the world will know you.
Li Yinan's market-selection logic is completely different, but more executionally valuable. He says: international markets aren't planned, they're bumped into by luck.
First bump around in different markets; once each place has its first ten clients, then look back and judge where to double down — that's when real planning begins. Framing by country is too coarse; what matters more is finding where the "hammer" you forged at home can find similar "nails" — whether the problem is similar, whether willingness to pay exists.
Mao Hua's selection is more conservative: prioritize markets with high AOV, strong purchasing power, and clean regulation — Singapore, Hong Kong, Australia, Malaysia. His logic: once you've run the model in high-barrier markets, radiating to neighbors is far easier.
Tony gave a consumer-goods judgment: physical consumer goods look to Europe and America, whose purchasing power is still the world's strongest; for 2B software they now mainly serve Chinese going-global companies with a Chinese background, where local trust matters far more than the product itself.
Breaking In: Four Playbooks
Market chosen, how do you enter? This was the most heated half of the roundtable; four people surfaced four paths, no standard answer, but each earned with real money.
Path one: Leaderboard-driven
Bruce's acquisition logic is clean — Agnes's three model lines (text, image, video) all broke into the global top 10 of authoritative leaderboards like Artificial Analysis. He calls this "Leaderboard driven": first let your technical strength speak for you worldwide.
For a team with no overseas channels or sales force, leaderboards are the lowest-cost credibility endorsement. Developers and enterprise buyers read leaderboards; have a name on it, and people come to talk.
Path two: Free for market
Bruce's pricing: free for consumers, RPM limited to 20 requests per minute, beyond that you pay. This limit is no obstacle for ordinary users, so conversion jumped and the domestic market grew fast.
Li Yinan took the same B2B path: fully free during the POC stage. A B2B client's biggest concern isn't price, but fearing migration costs and lock-in they can't escape. First let them run it and get results; renewal happens naturally.
This isn't a price war; essentially it lowers the decision threshold — let users feel the value first, and they'll pay for bigger value.
Path three: Sell results, not software
Mao Hua and Tony's business models share one thing: neither charges by software subscription, but by delivered outcome.
Mao Hua's CCMET is an "AI Finance Agent" — real accountants plus AI doing finance outsourcing for SMBs. Clients don't buy accounts; they buy the "result" of books done, taxes filed, compliance clear. SMBs can't afford a full finance team; they buy problem-solving, not a functional UI.
Tony's FOSHO also charges outcome-based; his own term is "result service," not the traditional Marketing SaaS subscription model.
The essence of this model: you don't sell the hammer; you sell the effect after the nail is driven in. The client doesn't care about your model's parameters or inference speed; they only care whether the books are right and sales are up.
Path four: Direct sales first, trust channels follow
Li Yinan's principle is the most direct: "go into any market and do direct sales first. If direct sales can't make it, channels can't either." Channel partners aren't philanthropists; if you can't sell it yourself, they won't lift it for you.
But Mao Hua added a B2B detail: SMB owners are extremely sensitive to data security and privacy and won't hand their books to a strange foreign company lightly. His approach is partnering with local finance firms that already have trust, borrowing their existing credibility to enter the market. "Don't find the one with the strongest channel capability; find the one users already trust."
Bruce added a dimension to channel choice: he splits users into three layers — 1% geeks, 10% power users, 90% the masses. He watches where the first two layers hang out. Geeks and Pro users are the voices, the reviewers, the ones followed in communities; once they endorse, the 90% masses follow. So Agnes's word-of-mouth strategy is: first find where the geek circles are, mix in, let them form a reputation, then let trust ripple outward layer by layer.
Neither Escape nor Pilgrimage
Before the roundtable ended, the host asked each for one closing line.
Mao Hua: "come to Singapore, Hong Kong, Australia, look us up." No sentiment, just doing business.
Li Yinan: "go into any market and do direct sales first." He repeated it, like reminding himself.
Bruce said one last line worth keeping: "do well enough in a single market and the global market will know you."
The reverse holds too — can't break through even one market, and talking globalization is self-deception.
These four on stage weren't talking about "sectors" or "positioning"; they were talking about next month's acquisition, billing, and entering the next market. The so-called besieged fortress isn't a philosophical dilemma of some wanting in and some wanting out; it's a very concrete question: that hammer you forged at home — where else can you still find nails?
Found one? Go in. Not found? Look somewhere else.
First break one market open; everything else comes after.
More Conversation Detail
Panelists: Bruce Yang / 杨通 (Agnes AI Founder); Yinan Li / 李一楠 (BetaAI Co-Founder); Hua Mao / 毛华 (2nd Brain Founder & CEO); Tony Sun (FOSHO Co-Founder)
Host: He Zhu / 朱鹤 (SMZDM CAIO & SVP)
Zhu He: Hello everyone, I'm Zhu He from SMZDM Technology. Over the past few years SMZDM has said it wanted to build a consumer large model. Many questioned us, many supported us. After these two or three years, we do feel we should adjust, but we decided to adjust only the first few words: we want an Agentic Commerce large model — the core is still a large model, not going into apps or Infra. That's my intro. To help Old Wu catch up on time, we decided to discuss only three things. After intros, one: what is real globalization? Coincidentally I have a community called "real going-global"; sitting here it may be "real globalization." Two: how do we choose countries or markets? Three: how do we enter GTM channels or a 2B industry? Let's start from these; first intros, please.
Yang Tong: I'm Bruce (Yang Tong), founder of Agnes AI, based in Singapore, and an NUS PhD. Agnes is both our model name and our app name. This product now has over 10 million global users. Recently we also did important work launching our text, image, and video models, all ranking in the global top 10, including on Artificial Analysis, probably the most authoritative board. And I think every AI company today can try our model, so on another level, rather than competitors, I'm more a partner to everyone.
Li Yinan: Hello, I'm Li Yinan, co-founder of BetaAI. Beta was founded in 2012, starting by serving China's financial industry across all kinds of marketing roles — basically every Chinese financial institution you can think of is our client. From 2024 we set up a new business unit specifically for AI-native products; over the past two years we've roughly kept a pace of a new product every quarter to half year. The first mature product is an "AI Sales Coach"; there's a booth at the door, welcome to learn more; then we did talking-head digital humans; early this year a new product called Xiaoluo — positioned as "a business-relationship-management advisor serving high-AOV sales." I want to use this stage to show the product and ask for your opinions. Simply put, we believe many high-value, high-AOV sales industries can't leave human sales and can't be replaced by AI, but humans today must be assisted by AI. So we built Xiaoluo, dedicated to helping sales in high-value industries work more efficiently daily — freeing everyone from rigid Salesforce-like CRMs, making a truly useful AI Agent for salespeople.
Mao Hua: Hello, I'm Mao Hua, founder and CEO of this Singapore company 2nd Brain. Our product brand is actually called CCMET; it's an AI Finance Agent, and this Agent carries two meanings: a human's Agent plus an AI's Agent. So we're a result-delivering company, delivering results through AI, providing finance services to SMBs. Because any SMB, anywhere, has no finance team of its own, so all its finance is outsourced, usually to traditional accounting firms. Now you can simply understand us as an AI accounting firm; we can provide these SMBs with all finance-related services, solved through AI, so efficiency and accuracy are much higher and cost much lower. I'm in Singapore for my seventh year; before coming I was at Tencent for 8 years, joining in 2011 and leaving in 2019, when I sold a company to Tencent — the product called Yingyongbao; I believe everyone here who has an Android phone and is from China must have used it. Then I just moved to the US, so we're preparing the US market and some Southeast Asian markets here in Singapore. Thank you.
Zhu He: Feels like an Agent for Agents.
Mao Hua: Yes.
Tony Sun: Hello, I'm Tony Sun, co-founder of FOSHO. Our company is an AI Marketing company and a service provider. Our clients are Chinese going-global or globalizing software and hardware brands; as long as they're consumer-facing, we can serve them. We use large models and agents to help brands do better marketing and sales overseas. I think it's a bit like what Mr. Mao said — we charge by delivered results. Traditional marketing services may have many subscriptions, but we now call it result service, charging by outcome.
Zhu He: I'll try to be the host with the fewest words tonight, because recently chatting with Agents I found that fewer words make it smarter. So the next question is very direct: what do you think real globalization is? How do you measure it? At what indicator state do you instinctively say "this is globalization"? Before that, you can first say what you think counts as "instantly fake" globalization.
Yang Tong: At least from our company's experience, last year 95%+ of our users were abroad, i.e. outside China. Among these, the fastest-growing markets were Southeast Asia (mainly Indonesia), Latin America (mainly Brazil), and the North American market — North America goes without saying, it's indeed the fastest AI-adoption market. But actually this year I have a new understanding of globalization. Although from consumer products every region has enough users, from the acceptance of AI adoption, the speed is completely different. The fastest adoption this year has actually returned to China, especially OpenClaw, including Hermes; this OpenClaw wave is absolutely China's adoption speed being the fastest globally. So my understanding is, doing globalization this year, you cannot leave the China market; if you lose the China market, you're not global enough.
Zhu He: I've been wondering — is "domestic" here Singapore or?
Yang Tong: I came to Singapore at 15, but every time I say "domestic" I still mean China. Especially when speaking Chinese, yes. So to sum up, my latest view is: from AI adoption, you can't lose these two big blocs now, the China market and the US market, because their growth, acceptance, and market leadership still far outpace other markets. Although from API and model services we grow fast in China and well in the US, that doesn't mean we're not an international team. Because we believe, in the current market, you must prove you're competitive enough in these two big markets (China and the US) before you can wait for the next wave of other markets worldwide to catch up. So I'll first give a slightly different view: my globalization this year is that you can't leave China and the US.
Zhu He: Very good; I'm personally very inspired. Globalization needn't specially avoid a certain market (like China); that's what truly lets you confidently call it globalization, rather than clamoring about "real going-global."
Li Yinan: Because I've personally spent over a decade in the mainland, and did VC investing domestically for over a decade, we have some own observations. Today talking globalization, I want to slightly turn it; I think some friends coming from mainland China to Singapore to see the world, the question really on everyone's mind is: how do I actually step out of the China market itself? Because I may have invested in about twenty or thirty companies; the China mainland market has a particularly severe problem — a harsh market environment. Two directions: first, from the procurement side, your clients aren't very friendly. If you do a software business, you can't do without KA clients, but domestic KA procurement is terrible; once they think your product is decent, their first reaction is to have someone clone it. So founders in such industries tend to go out, escaping China's somewhat vicious demand soil. Second, once you're in a somewhat fat mid-tier industry, China has another problem — brutally bad business competition, which I call "red-basin competition." Everyone hand in hand was living happily, then it suddenly got hot, a hundred people rushed in, cutting prices to 1% or even 0.1% of the original in a price war, competitors went from a few to fifty, and everyone died hand in hand. So domestically, whether SaaS or software, walking on one leg may be very risky. So today, rationally judging, stepping out of mainland China toward the global market is a rational need.
But on the other hand, once out, you find something's off; it isn't as good as imagined. A product validated domestically may have four things behind it: one, market demand exists; two, product supply is formed, with supporting supply chains; three, a team that can really execute; four, customer channels that can push to market. When moved to foreign markets, you find more than half may not migrate. For example, domestic demand may not exist abroad; the most resonant case is bike-sharing and power-bank sharing. At home you have Monster, Meituan bikes; overseas there's basically nothing. Two, products made at home usually don't acclimate and can't be used directly abroad. Three, excellent domestic teams lack international-market Context understanding, unable to understand the client across the table. Four, customer channels — in most cases domestic channels don't work. So stepping out is rationally very reasonable, but in execution you hit every pit; it's a very, very hard proposition.
Zhu He: I think I've seen a human-shaped GPT. See how much he talks like GPT? Covered the pros and cons of going global, one two three four, watertight — a human-shaped GPT. Also I found the first two panelists match a book called Fortress Besieged. A Singaporean who came over at 15 saying he wants to enter the domestic market is real globalization; someone at home saying you can't do it domestically and must go out — that's just the besieged fortress. Over to you, Brother Hua.
Mao Hua: Actually globalization often does want to enter the China market, but many conditions no longer hold, especially now with the emphasis on data security. Traditionally, C-end products go global more easily because cultural-difference requirements are lower.
Zhu He: You're the founding patriarch; Brother Hua founded Tencent's Yingyongbao.
Mao Hua: That was early. Doing C-end products going global is easy, but B-end products are actually very hard, because you're targeting enterprises in each country. Every country's policy and culture differ, so entering is very hard. Look at China's 2B SaaS products — which succeeded? Yonyou, Kingdee etc. are only in Indonesia, Thailand, and so on; tell them to do the US market and that's impossible; the US market has SAP and Salesforce blocking the way, and data security doesn't meet requirements either. Back when I came to Singapore, the company needed a tool; I asked WeCom if it had an internationalization plan, they said no. I ended up choosing Feishu, but later it became Lark, and there was a big pit: I registered with the Chinese Feishu; later because of data isolation, my Singapore employees downloaded Lark. At first data flowed; once isolated they couldn't log in. Going to Feishu president Zhang Nan didn't help either; the only option was having employees switch their App Store country to China and re-download Feishu. So internationalization really is hard.
Zhu He: On what basis do you persuade a foreign company to hand you their books?
Mao Hua: I'll answer that in a moment.
Tony Sun: Our company does Marketing; from our observation, two parts matter a lot: first localization, second diversification. Localization splits into two: product localization and branding/channel localization. On product localization, I previously worked at a big internet company; we made an overseas app, the first page was English no problem, but clicking to the third layer Chinese appeared. So you can't move the product out unchanged. On channel localization and diversification, much early going-global was just buying Google traffic and listing on Amazon; from today's view that's just doing sales overseas. Real globalization needs diversified layout: media partnerships, influencers, social media, both online and offline. Many excellent overseas brands make local consumers not feel they're Chinese companies at all, but local or global brands.
Zhu He: I think your company has this globalization gene; I hear you came out of Honor's globalization team. Good, our second topic is interesting too: country choice. We'll start with Bruce — why are you disguised as a domestic company? Go ahead.
Yang Tong: We're not disguised. Last year our main users were in Southeast Asia, like Indonesia, the Philippines, these populous countries. But this year I think the whole AI era has entered a 2.0 era. Last year was a hundred flowers blooming; this year it's many consensuses: first, Harness as an important Agent component produces more and more value, like OpenClaw, Codex, Hermes; second, it's clear that for mainstream models and products, you must compete in big, recognizable markets. I could be only successful in Southeast Asia, but mainstream would still consider me an AI company unable to compete. North America, for geopolitical reasons, doesn't trust foreign products, making it hard to enter the mainstream. So our path is to prove ourselves in a single big market (China) while expanding others. Recently our API Tokens per day have reached 200B (200 billion), possibly 1T next month. Without the China market we couldn't hit this number, because domestic developers' enthusiasm and acceptance are very high.
Zhu He: But I'm worried; there are so many domestic giants; by year-end at least 10 will reach trillion-parameter models, the price war has already started, and one well-known large model already stopped growing and started declining after cutting prices. How do you dare rush in? On what?
Yang Tong: Free, free.
Zhu He: Just fighting a price war, always cutting into a bottomless pit? How do you make money?
Yang Tong: We're free, but free for consumers, not for enterprises. We limit RPM (requests per minute) to 20; beyond that you still pay.
Zhu He: You startled me; I thought it was free for B-end.
Yang Tong: That's no usage obstacle for ordinary consumers. After going free, conversion went up, and we're growing very fast in the domestic market. So a recent big realization is: do well enough in a single market and the whole global market will know you; like DeepSeek, before it blew up domestically it was already slowly building a voice in the US. So we want to be a strong competitor in the China market and absolute king in Southeast Asia (especially Singapore), which pre-proves capability for our next Go Global.
Zhu He: Next.
Li Yinan: I'll keep it brief, no long speeches. I think international markets aren't planned, they're bumped into by luck. The first step out of China is bumping around in different markets; once each has its first ten clients, that's when you really start planning which country to grow in. More important is seeing where your workflow actually breaks down — where else the hammer you forged at home can find similar nails. Framing by country is too coarse; you still have to see the real business environment locally.
Mao Hua: Our main markets are Singapore, Hong Kong, Australia, Malaysia, and so on. Each country's purchasing power, culture, institutions, and policy differ, so we choose markets with somewhat higher AOV, stronger payment ability, and clean regulation (like Singapore); once done well, radiating to neighboring countries is easier.
Zhu He: Answer both questions together. One: how to choose a country; two: after choosing, how to choose channels.
Tony Sun: For physical consumer goods, Europe and America's business is the Golden Rule; their purchasing power is still the world's best. For 2B software there's more to consider; right now we mainly serve Chinese going-global companies with a Chinese background. On channel choice, number one is diversification. It used to be the traffic-buying era, just buy Facebook ads, but today doing only one channel, ROI must be low. From ad buying, to media partnership (like big airport billboards), to influencer marketing, to running your own social-media matrix — an omnichannel layout is the long-term winning strategy.
Zhu He: Each give everyone one closing line. Because after this round it ends.
Mao Hua: Then let's do an ad; everyone coming to Singapore or Hong Kong, Australia, look us up.
Zhu He: I didn't ask you to leave a one-liner now; first answer the channel-choice question.
Mao Hua: Channel choice. Actually 2B channel choice differs a bit per country; especially SMB owners care most about data security and privacy, so we partner more with local finance firms that already have trust.
Li Yinan: Go into any market and do direct sales first. If direct sales can't make it, channels can't either. Tonight at 8 we're launching a new high-value relationship-management advisor product, hoping to help all sales find value, remember value, and automatically present it to clients.
Zhu He: My concern is, how many Tokens are given free per value? Infinite?
Li Yinan: Infinite, you can keep using it.
Zhu He: Wonderful; you're the only company on Old Wu's forum giving a lifetime deal on the spot.
Yang Tong: Let me share how to do channels. First is Leaderboard driven — you must rank top 10 on the universally recognized authoritative board. Second is Marketing driven, doing word-of-mouth. We split users into 1% Geeks, 10% Pro users, and 90% the masses; we watch where the first two circles are. Third is when in Rome — be like a Chinese team in the Chinese circle, like an overseas team overseas. Finally a pitch: our models are free indefinitely.
Zhu He: Can 2B API be free?
Yang Tong: We guarantee free at least through the POC stage.
Zhu He: Free POC is already rare. It's been an honor to discuss with four Agent Builders today; thank you all.