Coffee by Day, Drinks by Night, and an Army of One China's "Good Place" Race in the AI Era

Yuandian Academy in Beijing's Wudaokou—fifty open workstations, a "learn/socialize/sell" loop, and a "coffee by day, drinks by night" bar—looks like coworking on the surface, but is actually answering a much larger question: when AI lets one person do the work of an entire team, the office—that underestimated social organ—is being quietly hollowed out. Six months later on a return visit, it has grown from a single bar into an entire infrastructure stack; and across the country, the same question is turning into a "subsidies for efficiency" arms race across a dozen cities.

Walk less than ten minutes north from Tsinghua University's west gate and you reach Wudaokou. Deep in this neighborhood packed with universities, office towers, and cafés sits a space called Yuandian Academy: fifty open workstations, no partitions, reserved for AI developers and so-called "one-person company" founders. The academy has recruited over eighty entrepreneurs and dozens of "curators"—the latter are both service providers and regulars in the room, taking on everything from business registration, tax, legal, and HR paperwork to deep coaching on strategy, marketing, and supply chains. It is also one of the Haidian Night School's 140 teaching sites, with courses ranging from AI technology frontiers to financial management.

What truly sets it apart from an ordinary makerspace is its rhythm. The academy's self-prescribed model is a down-to-earth phrase: a "learn, socialize, sell" loop. Inside is a bar called "Yuandian Bar · Coffee by Day, Drinks by Night"—coffee by day for brainstorming; a drink by night for colliding ideas. According to community accounts, many cross-domain collaborations and cross-technology fusions have happened at this bar, some even turning into formal partnerships. One bar, two faces by day and night—this is an almost exact replica of 18th-century London coffeehouses (see the International Chapter in this series), except this time the regulars are typing Python and talking valuations.

Treat it as ordinary coworking and you'll miss its real significance. Yuandian Academy deserves serious attention because it is responding to a problem far larger than "startup incubation"—but before digging deeper into that question, we first need to work out a more practical calculation: how does the "coffee by day, drinks by night" bar actually make money?

"Coffee by day, drinks by night" isn't Yuandian Academy's invention—it's a time-shared reuse business that has been repeatedly validated in the restaurant industry over the past few years: according to CNR, coffee consumption peaks almost entirely before 4 PM, with stores nearly "idling" after 6; bars are the exact opposite, with no daytime business. Plug both into the same bar and you're using the same rent, the same equipment, and most of the same team to serve two time-slot customer bases—as Daily Economic News reported, citing CIC analysis, selling only coffee or only alcohol makes rent and labor costs crush the single-store economic model, while "capturing both ends" dilutes those fixed costs.

Hidden in this calculation is a more direct margin arithmetic: beverages are inherently a high-margin business. Industry analysis shows craft beer margins typically range from 40%–60%, and cocktail margins reach as high as ~80%; per Winshang data, Helen's mini-pub maintained self-owned beverage gross margins above 70% from 2018–2020. Coffee's repeat-purchase rate is irreplaceable, but beverage average order value and margins are what turn the evening slot into real money—the other half of the equation.

Average order value increases are real: Daily Economic News reported that Haidilao's 2023 "coffee by day, drinks by night" format in Beijing had an average spend of ~¥300; Hangzhou's Ganqishi cross-over BAOBAO mini-pub averaged ~¥119 per person. The common thread in both cases: alcohol lifts average order value, but what really determines business health is whether the daytime coffee crowd is willing to stay into the evening—which is precisely where spaces like Yuandian Academy have an advantage over pure commercial dining: its "repeat customers" are resident entrepreneurs and curators, so it doesn't need to re-acquire customers every day.

This business is far harder to replicate than it sounds. It demands "hybrid" people—baristas who know alcohol are hard to find; bartenders who know coffee are even harder; talent that can fluidly switch between two systems is scarce. It demands "split" spaces—bright by day for table turnover, dim by night to keep people seated; the lighting and soft-furnishing transition is itself a craft. Daily Economic News also noted that even Starbucks Reserve, backed by Starbucks China and operating for nearly three years, has opened fewer than 20 locations domestically—scaling is far harder than imagined. Yuandian Academy is not the first to attempt this problem—Wudaokou has calculated the "good place" ledger several times before, and paid tuition each time.

The name Wudaokou itself carries the rust of history—named after the fifth railway crossing counted north from Beijing North Station on the Beijing-Zhangjiakou railway, that century-old line built by Zhan Tianyou and its Qinghuayuan Station didn't formally exit the stage until 2016. The nickname "Center of the Universe" truly went national because of a 2013 Weibo post: someone showed a 37 m² apartment in Huaqing Jiayuan selling for ¥3.5M, approaching ¥100K per m². Eight universities converge here; bars, nightclubs, cafés, book stalls, and snack stands cram together—Wudaokou was a high-density "hybrid" from the very start.

A little further west, Zhongguancun staged a more direct "build a good place" experiment. From 1989 to 2014, it was the famous Haidian Book City, a bookstore street. In June 2014, China's first innovation and entrepreneurship-themed street—Zhongguancun Inno Street officially opened, just 220 meters long, flanked by startup cafés represented by Garage Café and 3W Coffee. This was the peak of "mass entrepreneurship and mass innovation": coffee cups refilled, and an investment term sheet was negotiated right at the table; an event could draw 200+ people. Space, coffee, roadshows, hot money—that version's formula is strikingly similar to today's Yuandian Academy's "learn, socialize, sell."

Over a cup of coffee, an investment intention would be sealed.

— Circa 2014, people described Zhongguancun Inno Street's heyday this way

But that version of a good place couldn't stay hot forever. Enthusiasm faded from 2018 onward; by 2024, the former flagship 3W Coffee had moved out of Inno Street, its doors locked, lease expired, only debris left on the worktables. Inno Street itself didn't "cool"—it pivoted to embrace hard tech, and today half its projects are hard-tech. But the "café as incubator" space myth, propped up by foot traffic and hot money, truly receded. This is Wudaokou's own, miniature WeWork lesson (see the International Chapter): when a space sells mainly "vibe," the moment the wind stops, the vibe dissipates.

Since this street has built several generations of good places and watched them recede, why wouldn't Yuandian Academy be just another round of "café incubation"? The difference hides in a new species: the one-person company. The 2014 wave of entrepreneurs mostly still needed to form teams—find co-founders, hire engineers, build a squad; cafés were where they "found people to team up with." Today, leveraging large models and AI agents, one person can solo work that previously required an entire team spanning design, development, copywriting, operations, and legal. AI gave one person the leverage of a team, but not the companions of a team.

The last generation's cafés were for entrepreneurs to find teammates;
this generation's rooms are for "people without teammates" to find their kind.

That is the real subtext of Yuandian Academy's fifty open workstations—on the surface they're for one person to work; in essence they're packing a group of "ones" back into the same room. When machines can nearly do all the "work" for us, why must people still cram into the same room? This question elevates "public space" from a nice-to-have amenity to a real problem of the AI era—and to understand how serious it is, you first have to see what's being hollowed out: the office is actually a severely underestimated social organ.

In 1973, sociologist Mark Granovetter published the paper that would be cited tens of thousands of times—"The Strength of Weak Ties." His finding was counter-intuitive yet robust: what actually brings you new information, new opportunities, even new jobs is often not your closest family and friends (strong ties), but nodding acquaintances, same-floor coworkers, people you occasionally bump into (weak ties). The office's most inconspicuous value is precisely that it's a "weak ties factory"—the water cooler, the elevator, the hallway, the lunch crew, endlessly producing low-intensity, high-frequency chance encounters. Remote work's most hidden cost is shutting down this factory: research shows remote work causes persistent and significant declines in weak ties.

Gallup's 2024 data shows fully remote employees' loneliness (25%) significantly exceeds fully on-site (16%), with hybrid in between (21%); a Science study covering 2011–2024 data is even more direct: remote work increases alone time, worsens mental health across multiple dimensions, and drives up mental health service and prescription drug usage. One survey even found that over half of remote workers go an entire week without stepping outside their home, and a quarter can go days without speaking to anyone.

And the "one-person company" is the extreme end of this curve. It doesn't just lack a water cooler—it eliminates the "colleague" category entirely. When collaboration is taken over by agents, that most routine daily connection node is deleted from life wholesale.

Psychiatrist and former CEO Michael Freeman at UCSF conducted a frequently cited study: 72% of entrepreneurs reported at least one mental health issue, versus 48% in the control group; about three in ten entrepreneurs experienced depression, and when he asked entrepreneurs to rate their own loneliness—the average was 7.6 out of 10. This isn't old data from a decade ago: a 2025 founder survey found that in the past twelve months, 54% experienced burnout, 75% reported anxiety, and 46% rated their mental health as "poor or very poor."

Chinese samples paint an equally sobering picture. Frontline community observations are summarized as "20% profitable, 40% stuck, 40% adrift," most people trapped in 14+ hour workdays; a single platform algorithm change can halve income overnight. Training, health insurance, pensions, and risk that companies once underwrote are now dumped wholesale onto the individual. The one-person company owns the means of production (a computer plus a swarm of agents), and therefore faces the market's full risk and full loneliness alone.

We can finally all be bosses;
the price is that we all become islands.

Zoom out and you see this isn't some industry's minor mood issue—it's a public-health-level ebb. In 2023, the U.S. Surgeon General released a report formally calling loneliness an "epidemic": about half of adults are experiencing loneliness; a third are "almost constantly" online; 73% of respondents believe technology worsens loneliness. And the health harm of chronic loneliness is likened to smoking fifteen cigarettes a day—it measurably raises mortality risk.

The most counter-intuitive part is right here: we live in the most "connected" era in history, yet shared rooms are collapsing fastest. The countless new public spaces the internet promised ultimately produced countless private information streams—each person staring at a screen, viewing a world customized for them by algorithms. We are more "online" than ever, yet less "present" than ever.

We have infinite connections in hand; what we lack is just a shared room.

— The deepest paradox of the hyper-connected age

Generative AI is thickening the bill. When AI companions can chat with you 24/7 with zero friction, the "necessary hassles" that once pushed us to go out and deal with specific people are being eliminated one by one—yet precisely these hassles weave most of the fabric of social connection. String the previous sections together and you get a clear conclusion: connections are not directly destroyed by AI—they're hollowed out when AI removes the free carriers that once sustained them. Loneliness here is no longer an emotion for individuals to process alone; it has become an infrastructure problem that space can respond to.

In March 2026, an office building with over twenty years of history—Dongsheng Building—was officially renamed Yuandian Building. According to Beijing News, the building housing Yuandian Academy is now the landmark of the "AI Yuandian Community": a name change marking this block's upgrade from "a makerspace inside some office building" to an entire officially recognized urban infrastructure stack.

This wasn't an isolated move. On January 5, 2026, at the "Smart New Era: 2026 Beijing AI Innovation Highlands Construction Promotion Conference," Beijing announced its first four AI Innovation Blocks, and the "AI Yuandian Community" was successfully selected—the community covers ~3 km², with 30+ universities and research institutions within 1 km, over 1,000 AI scientists, 13,000 developers, and 100,000 related-major students. By the 2026 Zhongguancun Forum annual conference, the "Century Beijing-Zhang AI Innovation Belt" planning scheme launched a global call for proposals, with Wudaokou designated as the "Wudaokou Pioneer Zone," and the "AI Yuandian Community (Wudaokou) Renewal Area" covering ~1.4 km², encompassing Tsinghua Science Park, Zhongke Keyi Building, and multiple Chinese Academy of Sciences institutes.

Six months later, measure this block again with a finer ruler. People's Daily client's June 2026 on-the-ground report provides a more precise set of numbers: draw a 1 km radius circle centered on Yuandian Building, and it contains 37 universities, 10 new R&D institutions, 52 national key laboratories, 106 national-level research institutions, and 12,300 AI scholars; the community has 440+ resident enterprises, with upstream hardware and data services, midstream frameworks and models, and downstream application products all crammed into the same cluster of buildings—as one resident company VP put it, "upstairs and downstairs is upstream and downstream." Yuandian Academy is on the 10th floor; the first-floor lobby is a large-model ecosystem service station with one-touch access to five modules: policy, career, life, social, and international talent. But the most noteworthy line in this report is one that appears in no planning document: "no forbidding gatekeepers of traditional science parks." Schools, parks, and residential neighborhoods sit right against the office buildings; engineers step downstairs and fall straight into the bustle of daily life—precisely the kind of "publicness" that the second half of this article will repeatedly interrogate.

The community's positioning has also upgraded: from "startup incubation" to "the world's first stop for global AI talent innovation and entrepreneurship." Per Beijing Daily, Zhipu AI, Shengshu Tech, Gongji Tech and other companies have successively emerged from here, cited as living evidence of this bottom-up innovation ecosystem. These six months of change show that Yuandian Academy can no longer be summed up as "a bar"—it is growing from a space into an entire infrastructure stack. And the newest layer of that infrastructure is a marketplace.

On April 2, 2026, the national first OPC matching trading platform "Jiedanba" (Take Orders) was launched by Haichuangyuan—per Zhongguancun Science Park Network, it has gathered over 1,600 active OPCs and 100+ order-issuing enterprises, and launched two AI capabilities: an "Smart Quote Card" for the OPC side, and an "AI Need Assistant" for the demand side. On the same day, the OPC International Hub launched, leveraging "Jiedanba," Yuandian Academy, Yuandian Yuejie and other carriers, with 1,230 registered users and 78 order-issuing enterprises at launch, and the first offline meetup gathered nearly 100 ecosystem co-builders.

The international path is being paved in parallel: the UK's 48 Group Club indicated it will focus on building a "cross-border two-way channel" resource bridge; LEC AI plans to contribute its enterprise-grade AI platform, engineering capabilities, and international network. Per Jingbao Network, Beijing AI Yuandian Community general manager Li Haitao's statement was blunt: the community will leverage its base of global AI entrepreneurs to continuously link top global developers, startup teams, industry resources, and innovation projects, serving as a "global talent reservoir."

By July 2026, this playbook was placed on a bigger stage. On July 2, the 2026 Global Digital Economy Conference Digital Economy Industry Expo opened, and Beijing launched the "Global OPC Co-creation Festival" for the first time, with Haidian District showcasing 30+ OPC startup representatives, under the slogan "Come to Haidian, let one person's dream have a world-class stage." One Haidian District official's remark is worth copying down: "What we want to do is not 'attract investment,' but 'ecosystem breeding.'" More tangible than slogans is兑现 progress—the April "Measures on Comprehensively Building an OPC Entrepreneurship Ecosystem" (¥50K/year for two years of startup seed funding, up to ¥2M/year in model vouchers, up to ¥100K innovation breakthrough awards) had 406 enterprises applying by early July, with funds expected to be disbursed by end of July. The showcase samples are no longer just concepts: Nezha Entertainment founder Su Kui single-handedly built the "Dragon Bone Animation" platform, generating 2D skeletal animations from a single sentence, winning the "Entrepreneurial Beijing" OPC Feature Competition first prize.

From "academy" to "Jiedanba"—this is the largest paradigm shift in the "good place" narrative over the past four years: measuring whether a space is good is no longer "are the workstations comfortable," but "does the order flow circulate." The Yuandian community is not an isolated case: Shanghai Xuhui's "Model Speed Space," from its 2023 unveiling to now 500 days later, has seen resident enterprises fission from 100+ to nearly 400, with 12× annual revenue growth, completing the same leap from "incubator" to "industry ecosystem"; Shenzhen and Hangzhou have each issued follow-up policies—Shenzhen plans to build 10+ OPC communities by 2027, and Hangzhou's "Digital Intelligence Port" is pushing its AGI Youth Innovation Community to operational peaks. These developments will soon flow into the larger map in the next section.

A Hangzhou young man born in 1997, vocational high school graduate, worked in surveying, hotels, pivoted to software architecture, dabbled in IoT, and finally in August 2025 hit a path almost no one had walked: using AI to mass-produce apps. Five months, 120+ apps launched, of which 90% have paying users, mostly sold overseas. His morning isn't "start work"—it's "take over work": every day he wakes up and AI has already done the market research, requirements analysis, and coding; he just spends less than an hour reviewing and publishing.

He's not an isolated case. Over the past six months, Hangzhou's Shangcheng District has quietly gathered a crowd of such young people—one person, one computer, filling the shared workstations of "China Data Valley · Future Digital Intelligence Port." This is the new species called "super-individuals" or "one-person companies" (OPCs). And behind it stand two bigger names making a bigger bet: OpenAI's Sam Altman publicly bet that "the first billion-dollar one-person company" will appear in 2026–2028; Anthropic's Dario Amodei is more aggressive, betting on 2026 with 70–80% confidence.

When "one person + 10K GPUs" can truly sustain a unicorn, whoever can catch this wave of "super-individuals" holds the gateway to the next wave of innovation.

— And so, space became the chip every city fights over

In just one year, "OPC communities" spread from Beijing to Shanghai, Shenzhen, Hangzhou, Chengdu, Suzhou, Nanjing, and 10+ cities, and were written into multiple cities' 15th Five-Year Plans. Interestingly, each city's playbook differs—each is using its own industrial character to build a different kind of room for "super-individuals." (The table below is a mid-2026 snapshot; by May 2026, the actual national count was 65 cities, 426 communities, see § 16's national tally at the end.)

Look closely at these columns of "highlights" and you'll find them highly similar—almost all competing on the "production side": space, government services, compute, subsidies. This race is rapidly sliding toward a "subsidies for efficiency" arms race. And that's precisely where the problem hides.

The Hangzhou guy is the lucky side of the coin. The flip side is the "shovel sellers" rapidly emerging in this gold rush. Li Yizhou, an "AI influencer" with tens of millions of followers, loudly hawked "fully automated unmanned-company wealth" while selling AI courses—his "Everyone's AI Course" sold ~250K sets and grossed ~¥50M in a year, was pulled from all platforms and his account banned for copyright complaints, then he rolled right back and started selling "One-Person Company 3.0." One comment hit the nail on the head: tools are always amplifiers, not printing presses—they can amplify your ability only if you have ability to begin with.

Data is colder than anecdotes. In media reports, frontline community samples are summarized as "20% profitable, 40% stuck, 40% adrift"; some Shenzhen OPC circles have annual retention under 10%; by 2026, "¥30K, six months, deregistered: the first batch of 'one-person company' bosses have exited" became financial headlines. AI solved "can you build it," but not "can you sell it," "can you endure it," "can you sustain it." Training, health insurance, pensions, and risk that companies once underwrote are now dumped wholesale onto the individual.

Measure this arms race against the yardstick of "what the AI era truly needs," and the shared blind spot emerges. The vast majority of OPC spaces treat the "one-person company" as a "smaller company" to serve—so they拼命 optimize the efficiency end: cheaper workstations, faster business registration, more compute and traffic. But the data above already shows: one-person companies rarely fail because "capacity is insufficient"—they fail because of customer-acquisition exhaustion, revenue rollercoasters, solo risk-bearing, and unshared loneliness. And this end is precisely what workstations and compute cannot buy.

● Strength　● Partial　● Clear gap. No space type scores green on all five columns—"anti-fragility" is a collective weak spot.

The efficiency column is already all green; the real gap was never in subsidies. Xinhua nailed the direction: from "solo struggle" to "ecosystem symbiosis." And "symbiosis" can't be bought with more vouchers—it's connection, trust, and time that allows failure.

Don't let a space only add flowers in boom times,
yet fail to catch those who fail and those in the gap.

To be fair, policy-incubator types like Yuandian Academy have already gone further than pure sub-landlord coworking: they've stitched in learning and socializing, and used the night school to lower the barrier—if you check against the seven rules summarized in the International Chapter of this series, Yuandian Academy is already practicing several: Tuesday "AI" study sessions schedule chance encounters into the calendar, and the Haidian Night School turns learning into a public good. What it most needs to add is "anti-fragility" and "useless margin." A space that can't catch its failures is ultimately just a club, not a "good place."

In June 2026, the Beijing Municipal Bureau of Economy and Information Technology issued the "Action Plan for Supporting AI OPC Innovation and Development (Trial)", effective June 18, with eight support measures around cultivating OPCs, including up to ¥2M in subsidies for high-quality service communities, and 50% discounts for compliant enterprises calling base large models. This is real money, but it also quickly drew criticism.

Outlook Weekly's report raised sharp questions directly: many regions have issued OPC support policies, but subsidy rules, investment-attraction logic, and park products show converging trends, and we must guard against "hype concepts, empty parks, subsidies upon move-in, no real industry after landing." A subtler problem hides in the details—the AI tool subscription fees and API call costs that hurt most on entrepreneurs' bills often can't receive subsidies because there's no corresponding budget category; the standard three-piece set of "space amenities, compute support, talent subsidies" lands differently for each person, creating a structural mismatch between standardized subsidy rules and heterogeneous real needs.

Meanwhile, "OPC Cooling and Local Positioning," "$20K rakes in $400M" and similar controversial reports have started appearing, reflecting that some projects may have bubble tendencies in valuation or investment attraction—when "building communities" itself becomes a measurable political achievement, distinguishing real demand from investment-attraction rhetoric becomes the same task.

Hype concepts, empty parks, subsidies upon move-in, no real industry after landing.

— Outlook Weekly's question on the convergence of national OPC support policies

Put it all together and the "good place" arms race is reprising the old script of the "mass-entrepreneurship café" arms race—from Zhongguancun Inno Street's "close a deal over coffee" to 3W Coffee moving out and doors locked took a full decade. This time, the distance from policy subsidies to real order flow may be the true variable that determines how long this street stays hot.

On July 23, 2026, Professor Yao Jianhua's team at Fudan University School of Journalism released the "2026 China One-Person Company (OPC) Survey Report", the first to count this arms race clearly: as of May 2026, there are 426 OPC communities nationwide, covering 28 provincial-level administrative regions and 65 cities. The "10+ cities" map from the previous section was multiplied dozens of times over in a single year.

But spread the distribution out and the gradient is startlingly steep: East 335 (78.6%), Central 52 (12.2%), West 31 (7.3%), Northeast 8 (1.9%). Jiangsu alone has 158, ranking first nationally, followed by Shandong, Guangdong, Zhejiang, Beijing, Hubei, and Shanghai. "Super-individuals" sound like the form of entrepreneurship least dependent on geography, yet community distribution clings to industry corridors even more tightly than manufacturing—because communities compete not for people, but for orders, compute, and scenarios.

The truly valuable part of the report is the "health report"—its five listed dilemmas match this article's judgments from a year ago almost point for point: policy tools are homogenizing; some physical spaces face idle risk, "heavy on construction, light on operations"; tech tools can't replace business capability, with some entrepreneurs saying they feel more like "all-around employees" than "bosses"; work-life boundaries are increasingly blurred; cost centers have shifted from labor to compute, with some super-individuals reporting compute costs may exceed 80% of total costs; legal compliance mechanisms are incomplete, with risk pushed onto individuals. The report's recommendation to government is almost verbatim from § 09 of this article—shift from "give workstations, emphasize subsidies" to "give orders, build ecosystems," letting OPCs move from "islands" to "archipelagos."

Concurrent with the tally, this business is starting to be written into standards. On March 12, the Suzhou AI Industry Association联合 nearly 40 units released the nation's first OPC community group standard "OPC Community Construction and Service Evaluation Guide" (T/SZAI 001-2026), setting admission thresholds and a "POWERS Six-Force Model": pioneer talent aggregation, opportunity engine drive, full-journey accompaniment, ecosystem network linkage, return-growth acceleration, and professional service support. In June, Ningbo followed with an "AI OPC Community Construction and Evaluation Specification" (T/NBSIA 023-2026), grading across five dimensions—the project lead's metaphor is apt: "you can think of an OPC community as a classroom, where each 'classmate' is an OPC entrepreneur." Beijing is also aligning: Chaoyang District released an 11-OPC-community matrix at once, covering digital humans, AI film & TV, industrial AI, AIGC audio-visual, and data elements, with Beichen Commercial Management as the first state-owned-enterprise-led community built to Suzhou's group standard.

Standards are good: they force "subsidies for investment attraction" toward "quality improvement," and give entrepreneurs a ruler to measure communities by. But hold the Six-Force Model up against the traffic-light table in § 13 and you notice something—the dimension closest to "sanctuary" is "full-journey accompaniment," and it still measures helping you build and sell something—there's no line that directly says "catch the people who can't build it." The evaluation system illuminates efficiency and growth; the anti-fragility column remains in shadow.

So, one year later, the answer forks: on scale, "good places" won—426 communities, 65 cities, 106 policies; public space has never been built as infrastructure at this scale; on quality, the Fudan team's survey report delivers the same sober verdict as this article—spaces are idle, individuals are overloaded, risk is sinking downward. Building blocks has proven replicable; building a place that can catch people has not.