In one example from Shenzhen, a former product manager laid off from a major platform company is now running what she calls a business of one, using generative AI to write ad copy, design storefronts, and produce short-form video dramas from a repurposed industrial park where the rent is subsidised by the local government. In Silicon Valley, the going rate for a single frontier AI training run has reached extraordinary levels. Both are described, without irony, as the future of entrepreneurship.
The gap between those two scenes explains something important about where AI innovation is actually being built right now, and by whom. The dominant Western story about Chinese AI has been one of catch-up: cheaper copies, stolen IP, state subsidies papering over a fundamental inability to compete at the frontier. That story is becoming harder to defend. A different one is taking shape, and it looks less like a race for trillion-dollar valuations and more like a slow, distributed reinvention of what an entrepreneur even is.
The word that explains the shift
To understand the new Chinese entrepreneur, start with a word that entered everyday Mandarin around 2020: neijuan, literally “inward curling,” usually translated as involution. It was borrowed from the American anthropologist Clifford Geertz, who used it in the 1960s to describe farming systems in colonial Java that grew more elaborate and labour-intensive without ever becoming more productive. Chinese internet users found the concept described their own lives with uncomfortable precision.
Neijuan names a specific kind of exhaustion. Everyone works harder. Everyone competes more intensely. Nobody actually gets ahead, because all that effort simply raises the baseline that everyone else must now meet. Study more hours to score higher on the gaokao, and the cut-off rises. Work longer at the tech company, and the expected hours become the new normal. The Chinese leadership has grown worried enough about the phenomenon that, according to the South China Morning Post’s account of the central economic work conference, combating neijuan-style competition has become an explicit policy priority.
What makes this more than a linguistic curiosity is what neijuan produces. When the standard paths stop working — when graduate degrees no longer guarantee stable jobs, when the housing ladder becomes unclimbable, when the platform economy stops offering upward mobility — a lot of ambitious people start looking sideways instead of upward. They start building something smaller. And in 2026, the tool most of them are reaching for is AI.
Entrepreneurial workers, not founders
The most useful description of this new group comes from a recent essay in Rest of World, which calls them entrepreneurial workers: digital labourers who are neither traditional employees nor real owners of capital. They exist in the gap between the two. They use generative AI to write copy, run e-commerce storefronts, produce short dramas, edit video, handle customer service, and build small creative businesses across podcasts, blogs, and independent shops.
Compare this to the previous generation of Chinese entrepreneurs. The mass entrepreneurship wave of the mid-2010s, encouraged by the government’s “mass innovation, mass entrepreneurship” campaign, was defined by the language of opportunity: IPOs, unicorn valuations, becoming the next Jack Ma. Founders raised, scaled, exited. The pitch decks looked a lot like the ones being pitched in Palo Alto.
The 2026 version is different. According to reporting on Chinese entrepreneurship, today’s generation has shifted from aspiring to build unicorn companies to simply seeking sustainable income that covers basic living costs. That is not a growth narrative. It is a survival narrative dressed up in the language of entrepreneurship, and the AI stack is what makes it viable.
A single person can now do the work of a small agency. One human, plus a set of AI agents, plus a laptop. Costs get compressed. Speed gets compressed. Margins get compressed too, of course, which is the point — everyone else is doing the same thing. Neijuan follows you into your own business.
Why the DeepSeek moment fits this pattern
The same logic that shapes the entrepreneurial worker is shaping how Chinese foundation models get built. When DeepSeek released its R1 model in early 2025, the reaction in Silicon Valley was closer to shock than admiration. A Chinese lab, working under US chip export controls, with a fraction of the compute budget of OpenAI or Anthropic, had shipped a model that performed competitively on reasoning benchmarks.
The Western explanation quickly settled on suspicion: they must be lying about the costs, or they must be distilling from proprietary US models, or there must be some hidden state subsidy underwriting the whole thing. Some of that critique has merit. But the more interesting explanation is also the more boring one. DeepSeek and its peers cannot burn money and pile on compute the way US labs can, so they have been forced to become more efficient. Model compression, architectural optimisation, engineering efficiency, and open-source ecosystems have become the actual moat.
This is what constrained frontier innovation looks like. It is not the absence of ambition. It is ambition rerouted through resource scarcity. Silicon Canals has explored this pattern before in the context of smaller, sovereignty-focused models built at a fraction of frontier-lab budgets, and the point holds here: the assumption that transformative innovation requires the largest possible capital pool is a specific historical arrangement, not a law of nature.
Frugal innovation goes upmarket
There is a name for what is happening. It comes from India, not China: jugaad, the practical wisdom of creating more value with fewer resources. Frugal innovation used to be dismissed in Western business schools as a coping mechanism for poor countries — clever, sure, but not really where the frontier gets moved.
That framing is aging badly. The reporting around DeepSeek’s rise made clear that the efficiency gains coming out of Chinese labs are not marginal. They are structural, and they are being open-sourced, which means they diffuse rapidly. Silicon Canals reported in a piece on the Global South building its own AI on $50 hardware, and the technical stack for running capable models on modest infrastructure is now available to anyone willing to do the engineering work.
What used to be a bottom-of-the-pyramid strategy is becoming a general-purpose innovation strategy. And China is the largest laboratory in the world for running that experiment at scale.
The state is not absent
The romantic version of this story — plucky micro-entrepreneurs building elegant one-person businesses out of pure ingenuity — misses half the picture. The Chinese state is deeply embedded in the whole arrangement, and the shape of that embedding is changing.
Local governments are experimenting with support programs for solo entrepreneurs: computing vouchers, low-rent office space in repurposed industrial parks, access to models and data, sometimes small direct subsidies. The explicit goal is to absorb laid-off workers from major tech platforms and re-channel them into a new form of employment that sits inside the state-led AI framework rather than outside it.
This is a very different arrangement from Silicon Valley’s founder mythology, which imagines the entrepreneur as a solitary hero detached from family and state, changing the world through personal force and venture capital backing. Chinese entrepreneurs are almost never truly alone. They are embedded in policy trajectories, capital allocation regimes, and governance logic. And they are tethered, tightly, to the family.
Vulnerability inside the model
None of this is a clean success story. The one-person AI company is a lower-barrier form of entrepreneurship, but it is also a more precarious one. Platforms can cut off traffic at any moment. Algorithms shift without notice. AI itself is rapidly compressing the premium on the very skills these workers are trying to monetise — the copywriter using ChatGPT is competing with clients who are also using ChatGPT and beginning to wonder whether they need the copywriter at all.
The policy support helps, but it comes with strings. A model built on computing vouchers and repurposed industrial parks rises and falls with the policy cycle. When the local government’s priorities shift, or the fiscal situation tightens, or the political weather changes, the scaffolding can be pulled out from under thousands of one-person businesses at once.
The neijuan logic also does not stop at the border of your own company. When a million people are all running the same AI-powered e-commerce store, competing for the same customers on the same platforms, margins collapse toward zero. The tools that were supposed to liberate the worker become the mechanism through which the pressure gets more intense.
A different profile of the founder
What kind of person thrives in this environment? Not the archetype Western venture capital tends to fund. The new entrepreneurial worker is not looking for an IPO. They are looking for stability, autonomy, and enough margin to keep going.
That maps interestingly onto research on founder success that has been quietly accumulating for years. A Kellogg School study of 2.7 million company founders found that the average founder of the fastest-growing tech firms was 45, and that a 50-year-old is nearly twice as likely to build a runaway success as a 30-year-old. The mythology of the 22-year-old dropout founder is a specific cultural export, not a description of what actually works.
Related work suggests that the strongest predictor of founder success may not be visionary charisma but psychological consistency: the ability to keep showing up, to make similar decisions across long stretches of uncertainty, to not blow up the business by chasing the next shiny thing. That is precisely the profile of the Chinese entrepreneurial worker: not swinging for the fences, just running the shop day after day, adjusting to what the tools and the platforms allow.
What this means for the rest of the world
The interesting question is not whether Chinese AI catches up to American AI. It is whether the model of entrepreneurship being incubated in the subsidised industrial parks of Shenzhen turns out to be more portable than anyone in Palo Alto expects.
Because the ingredients are not uniquely Chinese. Cheap, capable models. A generation of educated workers the old economy no longer has room for. Tools that let one person do the work of a small agency. Those conditions are arriving, in different forms, almost everywhere the promise of stable upward mobility has quietly stopped being kept — in the graduate-heavy job markets of southern Europe, in the gig-saturated cities of South Asia and Latin America. China is simply running the experiment first, and at the largest scale.
If that holds, the one-person business running on a stack of AI agents is not a Chinese curiosity. It is a preview. And the questions it raises — whether someone tethered to a platform, a policy cycle, and a family’s savings is really an entrepreneur, or just a worker who now carries all of the risk alone — are the questions the rest of the world is about to start asking about itself.
Silicon Valley is watching China’s AI story for the wrong plot. The interesting development was never going to be another Jack Ma. It is the quieter, more uncomfortable discovery that you can now build a business of one and still not be free.
