Summary

TiDB’s core value is turning the “1,000 cups” that legacy databases must manage into a “magic cup” that expands automatically as the data grows. Founded in 2015, PingCAP spent its first 2 years mostly writing code before putting a distributed relational database behind use cases including Meituan Waimai, WeChat Pay, and bank-card transactions. 黄东旭’s view is that data keeps growing while traditional containers remain finite; TiDB’s real product value is reducing the complexity of scaling and managing multiple containers.
Open source’s moat is not publishing source code; it is using a transparent process to earn trust, attract talent, and retain control of the project’s direction. Top-tier open source does not mean throwing finished code onto GitHub. It means opening the entire build process from the first document, like turning a restaurant kitchen into a glass room. “Giving you the answer to copy is not enough”; the value lies in the thinking that keeps evolving. PingCAP has roughly 300 engineers, annual investment in the hundreds of millions of yuan, and years of accumulated historical know-how. Even a later entrant deploying 500 people at once would struggle to change the direction quickly.
Before revenue scaled, investors were buying non-consensus technical judgment and steadily rising adoption. PingCAP’s last publicly disclosed valuation was $3B, but at Series A the software could not even run and the company had no revenue. 云启’s 陈昱 reached a decision after scanning roughly 5 minutes of source code, while investors who could not read the code had to rely on lagging commercial metrics. One concrete example: a major customer had no budget but sent 4 or 5 engineers earning roughly RMB500K–1M a year to work on the project, creating a form of open-source value exchange.
Cloud services finally made it possible to commercialize open source without breaking the trust underpinning it. PingCAP was “dead set against commercialization” for at least its first 5 years. In the early days, salespeople went to large users to “collect an insurance premium,” creating information asymmetry—a practice 黄东旭 saw as “killing the hen for the eggs.” MongoDB’s growth after shifting from a software company to a services company offered a model to follow. Today, more than 70% of PingCAP’s ARR comes from cloud services; growth has stayed in the triple digits for essentially each of the past 2 years, driven mainly by internationalization and cloud.
True globalization is not “going overseas” remotely from China; it is the founder physically embedding the company, organization, and commercial relationships in the local market. 黄东旭 went to San Francisco in June 2015, just as the company began operating, took several wrong turns from 2017 onward, and moved his entire family to the US in 2022. His experience is that a company may make almost no money during its first 3 years abroad, and that the founder has to go all in personally rather than relying on a global team. English-first documentation from day one, company-wide English for internal IM and meetings in 2022, and local teams with regional CEOs and CTOs were organizational milestones produced by this “working backward” approach.
For US GTM, Chinese founders most often overestimate the language barrier and commercialization difficulty, and underestimate narrative, pricing, and long-term relationships. 黄东旭’s advice to AI founders is to live locally for at least 6 months, hire a local salesperson early, and sell the product personally as the founder. English only requires the courage to speak; before moving to the US, he spent 2 hours a day speaking with a tutor to remove the psychological barrier. The real problem is often having a valuable product but “being afraid to sell it at a high price,” or failing to win mindshare with demos, message frameworks, and stories. In the AI era, “good wine needs no bush” no longer holds.
PingCAP’s core AI bet is not another model, but providing agents with the right context, enterprise data, and a shared memory layer. 黄东旭 believes DeepSeek’s emergence at the beginning of the year showed that LLM fundamentals are now sufficient to support useful applications, with the first wave of restructuring likely to hit enterprise software. His own company still has 3 full-time employees maintaining Salesforce reports, while its data agent can already answer questions about sales activity directly. 黄东旭 first mentioned the interactions covered by NCP and A2A, then later used MCP as an example; a general-purpose shared memory layer has yet to emerge. The next 2 or 3 years will remain a “wild west,” where open standards, real usage, and network effects matter more than designing monetization first.

Deep dive

1. TiDB Turns 1,000 Data Cups into a Container That Grows

黄东旭 explains a database as a “container for data.” The digital life built around WeChat, Xiaohongshu, Weibo, and other services all depends on database software inside data centers. TiDB is a distributed database designed to let multiple machines deliver a unified service to the outside world.
曲凯 asks why there is still room for a database startup when Oracle is a company worth hundreds of billions of dollars and the industry has half a century of history. 黄东旭 answers with water and cups: as the volume of data rises, traditional databases have limited capacity, leaving engineers to manage “1,000 cups.” TiDB lets one cup grow automatically with the volume of water.
PingCAP was founded in 2015 and spent its first 2 years mostly “holding its head down and writing code.” 黄东旭 says TiDB now supports Meituan Waimai, WeChat Pay, bank-card transactions, and other use cases: “If you live in the digital world for a single day, you will definitely become one of our users indirectly.”

2. Open Source’s Moat Is Not Code, but Transparency and Control of Iteration

黄东旭 confirms that PingCAP opened its documentation and build path from day one. “99%” of projects on GitHub simply publish their code, but a top-tier project should be open source throughout its entire life cycle and operated openly.
Infrastructure carries customers’ “livelihoods,” so transparency directly determines trust. He compares it to a restaurant: giving customers the recipe is not enough; ideally, you turn the kitchen into a glass room where they can see the entire process, from sourcing to preparation to cooking.
曲凯 worries that publishing the company’s thinking will let large companies or faster competitors copy it. 黄东旭’s response is that “giving you the answer to copy is not enough”; the value lies in the approach to solving the problem. Software is not a product that stops when it is finished. The evolving understanding, direction, and speed are the core competitive advantages.
Being first does not guarantee victory, but accumulation changes the barrier to entry. PingCAP has roughly 300 engineers, annual investment in the hundreds of millions of yuan, and years of accumulated historical know-how. Even if AWS or Google Cloud decided to enter, it would be difficult to mobilize 500 people immediately and fill in the system’s history.

3. PingCAP Bet on the Hardest, Biggest, Most Global Direction—and Caught a Hardware Inflection Point

Looking back over 10 years, 黄东旭 lists open source, relational databases, and globalization among the day-one decisions. Relational, SQL, and transactional databases are like Mount Everest, but Oracle proved their ceiling can reach hundreds of billions of dollars. By comparison, even a successful vertical market might only reach $10B.
This is also a retrospective acknowledgment of luck: “We were young, our engineering capabilities were relatively strong, and we were also relatively naive at the time.” Today, he believes the challenge was indeed extremely difficult and that reaching this point “was not easy.” But the expected value came from choosing a bigger market: “The direction you take may matter more than how fast you run.”
The real “why now” was hardware maturity. Roughly 10 years ago, bandwidth conditions for data-network switches improved, while SSD throughput and speed became more than an order of magnitude better than older disks and prices kept falling. The algorithms themselves may have been researched for 10 to 20 years, but large-scale distributed systems finally had usable underlying conditions. At the same time, internet technology began moving into banks, governments, and large enterprises that operate half a beat slower.

4. Before Revenue, Capital Was Buying Non-Consensus Technical Judgment and Adoption

黄东旭 calls foundational software a high-leverage moonshot: databases, operating systems, and browsers are used by almost every company. The challenge is not imagining the market, but judging whether unfinished software can grow into general-purpose infrastructure. “The highest-valued projects are always non-consensus.”
PingCAP’s last publicly disclosed valuation was $3B, but at Series A it was merely an open-source project on GitLab, had no revenue, and could not even run reliably. Fundraising was not easy. 云启’s 陈昱 decided to invest after looking at roughly 5 minutes of source code, demonstrating the kind of professional judgment where “you can open a car’s engine and immediately see the difference.”
The company did have users early on; those users simply did not pay. Many internet companies hired engineers to deploy TiDB themselves and used it to solve important problems. 黄东旭’s philosophy is that if you create RMB1T in social value, converting just 1% of it can still make you an extraordinary company.
One concrete example came from a major customer. Its CTO said there was no budget, but offered to put 4 or 5 engineers at PingCAP’s disposal. At annual salaries of roughly RMB500K–1M each at the time, that represented a substantial investment. The customer received project insurance, while PingCAP received outstanding contributors: “a few good men” substituted for cash.

5. Only Cloud Gave Open Source a Commercial Model That Did Not Damage Its Foundations

PingCAP was “dead set against commercialization” for at least its first 5 years because databases require years of foundational work, and premature revenue targets would distort investment. The earliest sales strategy proved to be a failure: salespeople naturally went to large free users to “collect an insurance premium,” creating information asymmetry by threatening to reduce support if they did not pay.
黄东旭 saw this as “killing the hen for the eggs.” Open source builds trust through transparency, while commercialization tried to close the information back up. 曲凯 asks whether companies can still afford to go 3 to 5 years without making money. 黄东旭 believes it is actually easier now because “open source builds the ecosystem and community, while the money comes from SaaS and cloud services” has become a standard path.
MongoDB’s share-price growth after shifting from a software company to a software-services company gave PingCAP a model to follow. Today, more than 70% of the company’s ARR comes from cloud services rather than on-prem licenses. The business began accelerating around the pandemic years, and growth has stayed in the triple digits for essentially each of the past 2 years.
Cloud adoption is slower in China. Many companies do not even know how to account for a software SKU, so they price through labor days or a “metal box.” 黄东旭 strongly opposes tying software to hardware. Domestic cloud providers themselves have not made much money, leaving little room for partners: “Don’t look for partners who are starving and have their eyes fixed on food.” The open-source-plus-cloud model works for general-purpose software; vertical software does not need to be open source because its industry know-how is a key asset.

6. Globalization Is Not “Going Overseas”; It Is Growing the Company in the Local Market

黄东旭 believes many companies internationalize only because conditions at home deteriorate, creating a “look at the cards” mindset when playing poker. The first few years abroad are a period of paying tuition, and a company may make almost no money during its first 3 years. Without an expectation of absorbing continuous losses, it is difficult to build a long-term business.
In June 2015, after closing its angel round, renting its first office, and leaving their jobs to start the company, 黄东旭 put a suitcase on his back and went to San Francisco. He says he “really hates the term ‘going overseas’”: it still treats China as the point of origin, like a business trip. True globalization means moving locally, building a local team, and establishing local relationships.
This requires the founder to go all in personally. 黄东旭 stresses that it is difficult to do the job well without personally committing, and reflects on the detours created by keeping the main force in China while setting up a separate global team. PingCAP is now organized into China, Japan, APAC, Europe, and North America, each with its own CEO, CTO, and engineering team. 黄东旭 serves as group CTO, lives in the US, and works with customers and R&D teams globally.

7. English First Is Not Packaging; It Is Working Backward from the Organization

PingCAP’s first document was in English, and its code had no Chinese comments from day one. The software given to its first domestic customer did not even have Chinese documentation and had to be translated on the fly. 曲凯 asks whether globalization is equivalent to “de-Sinicization.” 黄东旭 says deliberately blocking Chinese IP would be “too deliberate”; English first is simply the default setting for a top-tier open-source project.
The second milestone came around 2022, when internal IM and internal meetings switched entirely to English. The third milestone is for overseas new hires to see no sign, from the hiring-market impression and HR systems through onboarding, that this is a company “originating in China.”
Even when every executive in a meeting is Chinese, English must be used if the recording and decisions need to be archived for the long term so that future international teams can review them. 曲凯 interprets this as confidence in the future. 黄东旭 says it is closer to Amazon’s “working backward” method: define the future organization first, then work backward to today’s milestones.
He uses Arrival to explain the relationship between language and thought: an international language is also an international mindset. 黄东旭 had never worked, studied, or even traveled abroad before PingCAP, but believes language “gets better with practice.” Strong professional ability without English is still insufficient for their standard for global executives.

8. Where the Body Is, at Least 70% of the Energy Follows

The first trap was keeping the main force in China while creating a separate global team. The second was a mismatch of people. If overseas commercialization is still handled mainly by a China-based team repeatedly visiting customers and emphasizing a “local product,” the interaction does not feel natural. PingCAP’s commercialization team is now largely local, and more than half of the customer-facing relationship team is local.
Chinese teams also tend instinctively to make products broad, complete, and complex. 黄东旭 reflects that adopting a “pure Silicon Valley” product and commercialization approach earlier would have avoided many detours. The biggest mistake was not moving to the US sooner.
He began taking wrong turns in 2017 and did not move his family until 2022. His observation is that wherever the body is, roughly 70% of one’s time, relationships, and thinking are naturally pulled there. Living in China while focusing wholeheartedly on overseas business is not impossible, but he has seen almost no especially successful cases.
The US came first because it is one of the two peaks of the software world. Taking Silicon Valley’s “hilltop” first and then radiating into regions such as Japan and Singapore is often easier than advancing country by country. When he first arrived in Japan, customers were willing to meet because they viewed TiDB as a “top-tier American project.”

9. The US GTM Barrier Is Overestimated; the Real Gaps Are Narrative and Pricing

His advice for an AI startup’s first year is concrete: regardless of visa status, live in the US for at least 6 months; once the product is ready, start commercialization early, find a local salesperson to visit customers with you, and sell personally as the founder rather than relying only on technical judgment to read the market.
US teams are good at stories, GTM message frameworks, and mature communications methods, while Chinese founders are often “too earnest.” Language is only a tool; the key is having the courage to speak. Before moving to the US, 黄东旭 found an English tutor in Beijing and talked for 2 hours a day, first proving that he could sustain a conversation with foreigners and then removing the psychological barrier.
Another overestimated obstacle is selling. US customers are “not embarrassed to talk about money” and will pay if the product creates value. The widespread problem is that founders fail to recognize their own value and are afraid to charge a high price. Pinterest paid directly because it used the product heavily, without the circuitous negotiations common among domestic open-source customers.
黄东旭 once became obsessed with secret sauce. After studying the field, he found that the answer was mostly common sense. Databricks, Snowflake, and MongoDB had already repeatedly validated the cloud business model. “Almost every fancy trick we tried failed.” Inviting people to dinner, having coffee, solving strange problems, and building long-term relationships work the same way everywhere.
His analogy for sales is that a salesperson who is excellent in China may still be a top salesperson when dropped into an unfamiliar environment. The ceiling of North American business is higher, but he also acknowledges that PingCAP has not entered the very top circles, and that Chinese founders may find that layer harder to access. That is a question to consider after the company reaches $10B. The impact of US-China relations may become more visible only after ARR exceeds $1B. For now, the priority is to win customers beyond the 5-kilometer radius of the Silicon Valley office.

10. There Is No Shortcut Around the Organizational Tax of Globalization

A startup may have to manage subsidiaries across countries, different ethnicities, time zones, products, fundraising, and business from its first or second year. 黄东旭’s conclusion is blunt: “This is the tax we have to pay. There is no way around it.” Compared with local competitors, Chinese teams start with “negative points”; neither the language gap nor organizational complexity can be eliminated by methodology.
When he first arrived in the US, his most extreme day involved 19 consecutive hours of meetings from the moment he woke up, moving from the China time zone into the US time zone. Organizations gradually develop resilience, like a guitarist’s fingers repeatedly forming calluses. “It hurts a lot at first,” but later becomes a necessary part of learning to play well.
That pain can also become an advantage. Chinese founders who have experienced complex organizations, cross-cultural collaboration, and high-intensity work can turn the pain into a future competitive advantage. 黄东旭 is optimistic about young Chinese founders who build their businesses in the US from day one because they have both engineering fundamentals and a greater sense of having “nothing to lose.”

11. Chinese AI Teams Are Strong at Engineering but Weak at Being Heard by the Market

黄东旭 believes the engineering fundamentals of Chinese AI products are “very, very strong,” with DeepSeek as the clearest example. When its engineering details are presented to US peers, they are often seen as “magic.”
The weakness remains go-to-market. As a heavy Cursor user, he has seen several domestic AI IDEs whose products are “not bad at all,” yet he had never heard of them in the US. This is not a technical failure; the products simply never entered the local developer network of awareness and discussion.
“Good wine needs no bush” is the wrong assumption in the AI era. The industry is still shouting in the street to attract attention. Brand, messaging, and a sufficiently cool demo all affect who gets the early lead. 黄东旭 says plainly that Chinese founders sometimes cannot compete with Indians because the gap lies in storytelling and the ability to amplify their voices.

12. Shared Memory Could Become Data Infrastructure for the Agent Era

黄东旭 believes the core problem AI has not solved well is “memory”: how to make an LLM understand the “you” inside the quotation marks—whether an enterprise or an individual—and continuously set the right context. A database company is naturally close to this layer of data and context tools.
Traditional database interfaces are designed for programmers, DBAs, and developers, but future users may be agents. From an agent’s perspective, what makes a good database and how data should be accessed and organized are different from direct human operation. PingCAP has allocated some resources to explore this direction, which still lacks an industry standard.
黄东旭 first mentions NCP as defining the interaction between agents and tools, and says A2A is also about interaction. 曲凯 then uses MCP as an example, and 黄东旭 discusses the MCP launched by Anthropic. A general-purpose shared memory layer has yet to emerge.
What determines whether a standard succeeds is sufficient openness, real users, several industry-leading triggers, and ultimately network effects. 黄东旭 says MCP seemed to attract little attention when Anthropic launched it, but appears to have been popularized by Cursor over the past 2 months. He calls the next 2 or 3 years a “wild west” and argues for taking a position early while staying open, rather than distorting a protocol for future competitive advantage: “The people who built TCP/IP surely were not thinking about making money from TCP/IP.”

13. Enterprise Software Could Break into Small Capabilities Linked by Agents

黄东旭 sees DeepSeek’s emergence at the beginning of the year as a turning point: LLM fundamentals have reached the stage where they can “do some useful things.” The first wave of restructuring may occur in enterprise software because the core competitive advantage of enterprises remains industry know-how. LLMs do not know that proprietary knowledge; they can only help accelerate its delivery.
His concrete example is Salesforce. PingCAP has at least 3 full-time employees producing reports and answering data questions, while the Chat BI/data agent he demonstrated can already answer operational questions such as “Which customers has this salesperson been looking at recently?” and “Who has been out in the field the most over the past month?”
Enterprise software interfaces may therefore move from Salesforce-style, broad, static systems to discrete capabilities linked together by LLM agents. Each person could have a personal assistant that accesses a large data pool underneath, eliminating the need to navigate complex menus personally.
曲凯 asks whether OpenAI will eventually become the unified gateway connecting all enterprise data. 黄东旭 breaks enterprise operations into “common sense” and “data.” LLMs already have common sense, but an enterprise’s most valuable asset is its own data. Companies with customer-facing interfaces will try to preserve information that was previously discarded and will be more cautious about handing data to OpenAI. His conclusion is that the database business may actually expand.

14. The Methodology That Remains After 10 Years: Patience, Energy, and Common Sense

黄东旭 used to “do whatever came to mind” and move too quickly. After 10 years of entrepreneurship, he is more willing to remain patient once the direction is confirmed. In the early days, uncertainty made him panic and act erratically. Now his choice is: “If this direction is OK, then you should be a little more patient. Leave the rest to time.”
He attributes his ability to keep going to energy management, physical health, and outlets outside work: playing music, building synthesizers, and making art. The CEO plays badminton. In the end, he returns to “respecting common sense—there are no fancy tricks.” The confidence he had in his twenties was equally simple: “Nothing to lose. Worst case, I fail and go back to a big company and get another job.”

AI研究员写科幻：对话Meta田渊栋，畅想智能未来 Trump's Big Week: Middle East Trip, China Deal, Pharma EO, "Big, Beautiful Bill" with Ben Shapiro