China OS vs. America OS (2026 version)
Another unfiltered conversation: the bloodline politics of AI talent; open source as strategy? token-maxxing; why OpenClaw hit China harder than Silicon Valley; future predictions
Who are the real power brokers of China’s AI, the equivalent of Sequoia and a16z in Silicon Valley? My friend Lei Du’s answer: the municipal governments of Shanghai, Beijing, and Hangzhou. Diligent, brain-fried officials running on pure FOMO and competitive anxiety, working with the same obsessive hustle as a Sand Hill Road GP. I’ll admit I’m being deliberately provocative, but only a little.
China and America run incommensurable operating systems, as the title suggests. The two worlds have never been more egotistic, never more mutually obsessed — and yet their architectures refuse to map onto each other. The Anthropic-Pentagon drama is only the latest reminder that AI is woven into the deep structure of nations: politics, sovereignty, governance, power, energy, talent, culture. You cannot talk about AI without talking about the nation-state’s body, its operating system, the way blood moves through it.
Last August, I published a Q&A piece called “China OS vs. America OS” with my friends Du Lei and Hua Han. So I summoned the same two minds for an update.
The “real talk” feeling returned the moment we sat down in Lei’s San Francisco apartment, still buzzing from NVIDIA’s GTC — the “Super Bowl of AI”: thirty thousand people in San Jose, I was told, nearly half of them Chinese, many having flown in from Beijing and Shanghai. It felt like the right moment for unfiltered thoughts: how much had shifted, in just a few months, between AI in the US and AI in China. Later that week, they took me to Y Combinator’s W26 demo day in Dogpatch — San Francisco’s old shipbuilding quarter, now repurposed for startup demos. I thought to myself while waiting in a long line for YC-paid ice cream: the most prestigious tech accelerator’s acceptance rate has dropped below one percent. And yet here were hundreds and hundreds of founders, the chosen ones, shoulder to shoulder. The math didn’t make sense. (And I was pilled by the first hotel on the moon idea because why not; for an entire afternoon, I LARPed as an encouraging VC while founders who are running on pure manic energy came at me nonstop).
Because Du Lei and Hua Han are both former engineers turned Bay Area investors, we approached this conversation VC-mindedly: capital markets, government relations, talent flows. We covered the following (I bolded my favorite parts):
The mountain factions: how China’s local governments mirror Silicon Valley’s big VCs
Bloodline politics and factional power in China and Silicon Valley alike
The golden age of China’s open-source AI?
Token-maxxing: the political economy of burning compute
The OpenClaw phenomenon: why it exploded in China
China: Definite but pessimistic? A reassessment
Predictions for the next twelve months
About my guests:
Du Lei: Investor at Sancus Ventures; formerly co-founder of Huma Finance and early team member at Opendoor
Han Hua: Angel investor; former VC at GV (Google Ventures)
The mountain factions: how China’s local governments mirror Silicon Valley’s big VCs
Afra: Lei, you have a fun comparison between China’s regional AI competition and Silicon Valley’s venture capital ecosystem. Can you walk us through it?
Du Lei: This extended trip back to China this month gave me something I’d never quite had before: a ground-level view of what the “regional competition model” actually looks like in practice. The rational understanding was already there: Chinese local governments jockey for position through industrial policy, targeted subsidies, and coordinated investment, each trying to plant a flag in some promising sector and cultivate the next monopoly champion. But watching it unfold in real conversations and real deals gave it a texture that no written analysis could convey.
Coming back to the Bay Area, I found myself reaching for an analogy. In the American AI ecosystem, the primary architects of capital allocation are the mega-GPs — Sequoia, a16z, and a handful of others, each with the kind of concentrated authority to effectively determine who wins in a given vertical. In China, that role belongs to local governments and state-backed LPs in the major economic zones, whose official imprimatur functions as both funding mechanism and seal of legitimacy. In structural terms, China’s local government apparatus rhymes with Silicon Valley’s power broker class.
Think of each regional government as its own investment ecosystem with its own gravity. Shanghai resembles Sequoia. Hangzhou is more Benchmark. This is admittedly a very Silicon Valley lens to apply, but the structural homology holds up better.
Afra: That analogy immediately called to mind a Hardfork episode I listened to on my way here, about what they’re calling “AI brain fry” — the phenomenon where sustained, intensive use of AI tools leaves people more anxious, depleted, and cognitively frayed. The cohort apparently most afflicted are Silicon Valley’s tech investors. If you mirror that image onto China, the equivalent cohort — the most obsessively, pathologically fixated on the AI narrative — turns out to be local government officials.
Which actually sharpens your comparison in an interesting way. America has fund-versus-fund mountain warfare; China has city-versus-city. But here’s what I’m curious about: do these two systems have analogous rules of engagement? In American venture, there are informal norms — a fund that leads a Series A in Company X generally won’t lead a competing company in the same space. Does something similar operate in China? If Shanghai backs a company, is that company now effectively tethered to Shanghai’s mountain?
Du Lei: My read is that the regional fragmentation in China actually makes late-stage co-investment more exclusive than anything you’d see in the Bay Area. Bay Area mega-funds regularly syndicate at later rounds, and even the biggest players will share a cap table when the stakes are high enough. In China, the late-stage dynamics are different, because so much of what matters — access to large-scale compute subsidies, entry onto government data resource whitelists, favorable infrastructure allocation — flows directly from which regional government has taken a position. Add to that the expectation of return investment, local entity registration, and in-jurisdiction tax contribution, and you have a system that structurally discourages cross-regional co-investment until the field has narrowed dramatically.
Afra: That tracks. Z.AI is the quintessential Beijing company. MiniMax is unmistakably Shanghai. And Alibaba’s Qwen, regardless of venue, seems to hold every major event in Hangzhou. We’ve talked before about how China’s regional competition turns talent recruitment into a zero-sum game — that logic now applies to the AI narrative with even sharper force.
Hua Han: The picture is a bit more fluid than pure faction warfare, though. These regional ecosystems do cooperate when incentives align — Shanghai and Shenzhen, for instance, have a notably closer working relationship; the two cities effectively function as siblings. Beijing operates by its own logic, with its own entrenched networks. The earlier you are in the cycle, the more flexibility exists; the later you are, the more the sheer scale of capital involved forces syndication, whether parties prefer it or not.
Du Lei: And once a decisive contender emerges, the Chinese system has very little tolerance for prolonged mutual attrition. The clearest precedent is the ride-hailing wars from a decade ago.
The consolidation of China’s ride-hailing market took barely two years. In February 2015, the two dominant players Didi and Kuaidi merged, with almost theatrical timing: Valentine’s Day. The combined entity then turned its firepower on Uber China, which had the backing of Tencent, Alibaba, and Baidu simultaneously. Both sides were burning roughly one billion renminbi per month in subsidies. By August 2016, Uber China was finished — its brand, operations, and data absorbed entirely by Didi. What had been a five-player war compressed into a single dominant platform in under three years.
Didi and Uber China shared four common investors, and it was precisely the overlapping shareholders who quietly engineered the merger. This was, in the academic phrase, “an investor-driven acquisition”. The AI sector will likely follow a similar script. Ride-hailing was a single commoditized vertical where the only moats were subsidies and network density. AI is a horizontal capability layer that cuts across dozens of industries at once. Consolidation in China tends to produce a dominant player per vertical, but AI doesn’t reduce to one vertical. The pressure toward merger will be real in specific product categories, inference infrastructure, maybe consumer agents. But the idea that one company swallows the whole AI landscape the way Didi swallowed ride-hailing overstates the parallel. What we’re seeing now is so intense precisely because we’re still in the early innings.
Afra: So the Chinese arc looks something like this: a field of mid-sized oligarchs stake out their territories, consolidate, and then the state steps in to enforce a détente and prevent mutually assured destruction. But from where I’m sitting, China’s AI sector already feels like it’s deep in the scorched-earth phase — everyone is amputating limbs to outlast the competition. Why wouldn’t AI follow the same pattern as the Didi wars?
Hua Han: There’s already a clear policy signal here, and it’s worth understanding its genealogy. The phrase “involution-style competition” first entered the vocabulary of central leadership at the Politburo meeting on July 30, 2024. In June 2025, the National People’s Congress passed amendments to the Anti-Unfair Competition Law, adding a dedicated clause against below-cost predatory pricing; the Central Finance Committee followed in July with renewed emphasis on supply-side discipline.
It goes back to what Dan Wang argues in Breakneck: Chinese industrial competition exists in service of national strategy, not as an end in itself. While China hasn’t yet secured a dominant global position in a given sector, the system tolerates and even encourages brutal internal competition as a mechanism for driving down costs and shaking out weak players. But once the consolidation is sufficiently advanced, the state redirects toward oligopoly and profitability. The competition was always a means, not the destination.
Afra: So your read is that EVs have already cleared that threshold, they’re through the consolidation phase and starting to generate real returns, and AI is still fighting for position in the first half?
Du Lei: That’s roughly how I’d characterize it. EV feels like it’s into the semifinals. AI is still early in the game.
Hua Han: With one important caveat: AI has a different competitive geometry from the start. China’s model ecosystem, from day one, has been oriented toward the international arena — not just a carve-up of the domestic market, but a genuine contest for global position. That changes both the stakes and the timeline.
The Bloodline Politics and Factional Power in China and Silicon Valley Alike
Afra: Coming back to something Lei and I chatted earlier: the early competitive dynamics of China’s AI scene have produced intense relationship politics, particularly the “factional” character (门阀) of Beijing’s AI circles: which university you attended, which advisor you chose, which great patron you attached yourself to. These things determine whether you get access to compute, capital, and local government support. How do you make sense of this kind of factional politics (门阀政治) spreading through Chinese AI?
Du Lei: It’s a phenomenon anyone operating in the field can observe directly. In Western terms, you’d call it “bloodline politics” — your academic lineage, your school, your work history directly determine which faction you belong to and who you end up founding companies with. It’s an awkward dynamic for the industry as a whole, because resources end up concentrating heavily at the top. But honestly, Silicon Valley is much the same — are you a Thiel Fellow? Did you come out of Yao’s class in Tsinghua? There’s no fundamental difference.
Afra: Right, Silicon Valley’s own bloodline politics branches outward from the authors of a handful of foundational papers — who studied under whom, who worked at which lab — and that genealogy largely determines whether you can raise money and get your hands on compute.
Du Lei: Let me trace the lineage of this pureblood lineage for a moment, because it’s actually a single tree. The roots were planted by Hinton’s generation, who laid the foundations of deep learning. The 2017 Google Brain paper “Attention Is All You Need” was a critical fork — nearly every one of its eight authors went on to become the seed of a company or a major research team. Add in the DeepMind lineage — itself deeply connected to Hinton — and you’ve essentially accounted for the entire global population of people who have ever had the hands-on feel of training at ten-thousand-GPU scale. When you add it all up, it might be a few hundred people, most of whom have direct mentor-student or former-colleague relationships with each other.
And the deeper reason is: for investors, judging who is genuinely capable is extraordinarily difficult. Those who control the resources lack the ability to make objective assessments, so they fall back on lineage and pedigree as their filtering mechanism.
And this creates a self-reinforcing loop. There’s a fun phrase in Chinese tech circles: training large models is like liandan(炼丹) — “alchemy,” the ancient Chinese art of refining elixirs. The difference in intuitive feel between someone who has actually trained a model on tens of thousands of GPUs and someone who hasn’t is immense. When a company is deciding whether to proceed with a training run that will burn twenty million dollars, that judgment is almost never a scientific question — it’s closer to a craft or certain sensibility. And since only a small cohort has been trained that way, they are the ones who keep attracting disproportionate resources everywhere they go, which only deepens the structure further.
The Golden Age of China’s Open Source AI?
Afra: Speaking of which, I came across something interesting: Premier Li Qiang is still publicly championing open source at the national level, yet on Alibaba’s most recent earnings call, management didn’t breathe a single word about it. Do you think Chinese AI open source is fundamentally unsustainable — that Alibaba has already read the writing on the wall and is quietly pivoting resources toward closed-source products to compete directly with the likes of Doubao from Bytedance?
Hua Han: The thing is, the best models at most of these companies have never been open source to begin with — MiniMax, Zhipu, all of them maintain a parallel open-source branch alongside a proprietary one. Open source is an effective commercial strategy for building reputation, capturing market share, and earning user trust. Meanwhile, many Chinese model companies quietly reserve their most capable models as closed-source offerings sold as enterprise services. So it’s hard to say whether Alibaba or any other company will stick with open source indefinitely — they’ve never actually promised to. Open source is one of the most powerful tools available to challengers trying to close the gap with the leaders. But as a long-term business strategy, the commercial logic simply hasn’t proven itself out yet in the AI industry.
Du Lei: In the roughly six months after DeepSeek dropped, there was a palpable “Olympic gold medal” mentality in the Chinese ecosystem: everyone was pouring resources into climbing the benchmark rankings, trying to establish brand dominance for the whole industry in the global competition. Because in the year before DeepSeek, Chinese AI had been in a genuinely suffocating position.
But a year on, the open-source business model still has no answer. From the individual perspective of researchers and lab directors — people with some social media presence who can also publish papers — open source has been enormously effective at building personal influence. But for companies burning tens of millions a day, returns have to materialize eventually.
More importantly, after DeepSeek appeared, local governments across the country flooded the space with industrial policy funding — compute subsidies, training budgets, all tilted heavily toward open source. The logic was: if the resources are there, why not? But from a purely commercial standpoint, if there’s still no visible return in three to five years, the path becomes very difficult to sustain.
Afra: Precisely because open source lacks a clear commercial return, it's become very easy for outside observers to read it as the Chinese government playing some grand AI 4D chess. That's a narrative I hear constantly in AI safety circles in London. Some researchers worry that middle powers like the UK, Australia, Indonesia, and Singapore are quietly building their national AI stacks on top of Qwen or GLM, and that China could one day weaponize that dependency. These countries have a strong desire to build their own national AI stacks, and fine-tuning open-source Qwen models has become the path of least resistance.
My own answer to these Western researchers is: you’re partially right, and partially wrong. Partially right, because local governments genuinely are subsidizing open-source AI at a significant scale. Partially wrong, because Chinese AI companies are ultimately profit-seeking, the moment open source stops working, they will cut it.
Hua Han: The logic of using open source as a strategic counterweight to Silicon Valley — to break up the dominance of the Big Three — that logic holds. It’s the same playbook as when Chinese companies poured effort into open-source software to catch up with Western enterprise software: Linux, Kubernetes, Open Compute. And there’s still capital waiting on the sidelines — Middle Eastern and European money that hasn’t come in yet in any serious volume. Once companies like MiniMax and Zhipu list on Hong Kong’s stock exchanges, Southeast Asian and Japanese and Korean capital flows into Chinese equities in quantity. And if you look closely, even though they’re open source, the internal teams consistently complete parameter tuning three to six months ahead of external developers — which translates into real performance and pricing advantages. So MiniMax and Zhipu’s overseas revenue growth has actually been quite strong.
The US-China competition is real. But the “government’s 4D chess on AI” framing is far too totalizing. There are certainly factions within Silicon Valley that have every incentive to amplify the China AI threat — partly to lock open-source models out of the American market, partly to lobby for more government infrastructure spending at home. But plenty of other players are quite happy to see it all unfold. NVIDIA, for one: its worst nightmare is an OpenAI-Google-Anthropic oligopoly. More competitors, the better. At the GPU infrastructure layer, the lower you go in the stack, the more you love open source. Silicon Valley’s attitude toward open source is entirely determined by where you sit in the technology stack — the higher up you are, the more you embrace it; the lower down, the more ambivalent you become.
Token-Maxxing: The Political Economy of Burning Compute
Afra: Some Silicon Valley companies now track employee token consumption as a performance metric — there are internal leaderboards ranking who burns the most. Does that actually make sense? And given that American companies are running on closed-source models, the costs are enormous — a single engineer might be burning hundreds of thousands of dollars in API calls per month. Is there an equivalent culture in China?
This came up directly at GTC 2026, where Jensen Huang floated a novel compensation model that would give engineers a token budget on top of their base salary. This concept has been circulating widely in Silicon Valley ever since.
Meanwhile, token-maxxing has become a genuine workplace culture. Reports suggest one OpenAI engineer processed 210 billion tokens in a single week — the equivalent of filling Wikipedia 33 times over. Meta, OpenAI, Anthropic, and others have all built internal token consumption leaderboards, turning AI usage into a status signal and performance indicator.
Those are American numbers. What does China look like?
Du Lei: A similar informal standard exists in China’s indie developer community: can you burn a billion tokens in a week, in a month? It’s an unofficial signal of capability. But it doesn’t map linearly onto dollars spent. If you configure your workflow carefully — Claude Code on the $200/month plan, Codex X Pro for certain tasks, cheaper models you pay for yourself handling the lightweight work, each tool placed where it performs best — you can reach a very high level of productivity for around $500 a month. Run everything through the API or on a company expense account, and the costs scale up dramatically.
But when we talk about token-maxxing, the real question worth asking is: given the same volume of tokens, how do you generate the most value? American companies have gravitated toward the highest-conversion use cases, because the US has a mature B2B playbook and enterprise customers with real willingness to pay — the economics work.
China’s internet has historically been strongest in consumer and entertainment. The best Chinese AI products are accumulating methodology, data, and creative ambition in that direction. So in China, the prevailing belief is that the next great AI opportunity will be a super-consumer product — but until that product materializes, the industry will keep pressing forward along its existing strengths, in a state of productive anxiety, waiting for the category to be defined.
Hua Han: There’s also a fundamental asymmetry in the value a token can produce — and behind it sits a very straightforward labor substitution economics.
Consider American salary benchmarks: a senior associate at a major New York law firm earns between $80,000 and $110,000 a year; a senior engineer in Silicon Valley typically earns $200,000 to $300,000, with equity on top. If an AI agent can absorb even 30% of either role’s workload, the daily value it generates is quantifiable — and that value vastly exceeds the marginal cost of the tokens consumed.
China’s white-collar compensation structure is entirely different. This gap in human capital costs means that each token used to automate Chinese white-collar work produces proportionally less “savings value” in the first place.
That doesn’t mean AI applications in China lack opportunity, but the opportunity doesn’t lie in replacing expensive talent the way it does in America. It lies, as we’ll get into later, in manufacturing and industrial-scale efficiency gains.
The OpenClaw Phenomenon: Why It Exploded in China?
Afra: We talked about Manus last time — looking back now, has it essentially been eclipsed by OpenClaw? I also want to get into the OpenClaw phenomenon in China, which has been widely analyzed but is worth unpacking properly, especially now that the Chinese government has started calling for the hype to cool down.
Let me set the scene. OpenClaw is an open-source AI agent framework released in November 2025 by Austrian developer Peter Steinberger, designed to let AI models actually do things on a computer. In February 2026, Steinberger was recruited by OpenAI. At GTC 2026, NVIDIA featured him in the pre-show segment, and Jensen Huang publicly called OpenClaw the next ChatGPT.
In China, the tool’s popularity has far outstripped anything seen in the US. Tencent and Baidu organized OpenClaw configuration workshops in Shenzhen that drew retirees and students alike; developer meetups in Beijing sold out instantly; China’s daily token consumption has surpassed 140 trillion — more than a thousandfold increase from the 100 billion daily tokens of early 2024. According to cybersecurity firm SecurityScorecard, China has now overtaken the US in OpenClaw adoption. So how did this happen?
Du Lei: There’s a recurring pattern in China’s AI scene: every year around the Chinese New Year, a pulse of tech hype crests. Last year it was Manus, this year it’s OpenClaw. I vividly remember Manus’s team coming to the Bay Area for a gathering during last year’s GTC. It felt like a genuine cultural moment. A year later, the entire narrative has migrated elsewhere — which is telling in itself.
What’s worth examining more closely is that both Manus and OpenClaw generated far greater reverberations in China than in Silicon Valley.
There’s a structural reason for this. In Silicon Valley, OpenClaw’s arrival didn’t produce any cognitive shock — because it emerged from a complete evolutionary sequence. We’d already had Devin, various agent frameworks, and a step-by-step progression of Claude and other tools steadily advancing in capability. OpenClaw felt like a natural next layer — a more flexible agentic glue. Most of what it enables, people in the Valley had already approximated six months earlier through other means. Exciting, yes. Directionally significant, yes. But perceived as an organic next step within a familiar lineage.
In China, that entire year of evolution had essentially not happened. The domestic conversation had stayed at the level of “how do I use AI to break down a PowerPoint” — consumer entertainment tracks like image and video generation, extremely sophisticated in their own right, but irrelevant to anyone who needed AI to actually assist with work, anyone who needed a genuine intelligent agent.
So when OpenClaw arrived in China, it compressed an entire year of product evolution into a single moment. The final exam answers appeared on the table overnight. OpenClaw settled the debt that the domestic AI productivity track had been accumulating for a year — all at once, in one decisive strike.
Hua Han: On the supply side: every model company has enormous incentive to accelerate this. AI capability has climbed in distinct steps — first ChatGPT-style single-turn conversation; then reasoning models, where thinking time extended but the interaction remained fundamentally one-shot; then Claude Code and open computer use, where single interaction is no longer the design target and multi-turn, long-horizon task collaboration becomes the paradigm, with token consumption rising exponentially. For model companies, this is an enormous business — their entire imperative is to sell tokens.
Shenzhen is home to vast numbers of hardware and model companies. Every laptop or phone that ships with an OpenClaw client pre-installed becomes a persistent token consumption entry point. Token-maxxing, from a pure business logic standpoint, is entirely rational for every model vendor.
Afra: I have my own theory as well. A lot of local governments in China are desperately hungry for new tech narratives right now. Against the backdrop of economic slowdown and high youth unemployment, keywords like “digital nomad” and “open source” have become the latest in a string of buzzwords that local governments race to adopt. From Liangzhu and Anji in Hangzhou to Huangshan in Anhui, cities and towns are competing to attract AI digital nomads. Local governments are anxiously latching onto each new narrative, trying to shore up their talent pipelines and fiscal resources.
Hua Han: From the user side too, the appeal is real. OpenClaw is a genuine productivity tool. Chinese corporate culture — whether in state enterprises or major internet companies — is notoriously process-heavy. Even if you’re already cycling through DingTalk and Feishu (China’s Slack) all day, a tool that can actually automate your routine work has authentic pull for ordinary users.
Doubao had actually launched an AI phone that could order food delivery, hail rides, and operate every app installed on the device. But the Doubao phone was controlled by ByteDance, and none of China’s other major tech companies wanted ByteDance acquiring their users’ data through it or routing around their advertising ecosystems. So WeChat and Alibaba both blocked it. OpenClaw is open source — a fundamentally different proposition. There’s no single controlling party; every Android manufacturer can build their own version on top of it without worrying about data sovereignty.
Afra: Right — I remember people being genuinely stunned by the Doubao phone when it launched. It also earned the distinction of being perhaps the shortest-lived phone in history.
China: Definite but Pessimistic? Let’s re-evaluate
Afra: We discussed a framework back in last August: “indefinite optimism” describes America, “definite pessimism” describes China. Let me briefly explain where it comes from. It’s Peter Thiel’s four-quadrant model from Zero to One, which cuts attitudes toward the future along two axes: optimistic versus pessimistic, and definite versus indefinite — meaning whether you have a concrete vision of the future and a plan of action to match.
When we last applied this framework, Hua Han, you argued that China’s definiteness — this capacity for clear-eyed anticipation of where things are heading and active mobilization of resources accordingly — is itself a competitive advantage, even if the underlying register is pessimistic. Meanwhile, American optimism was starting to look increasingly unmoored: a belief that things will get better, without anyone taking responsibility for making them better, and no one actually designing that better future. Does that reading still hold?
Hua Han: Broadly, yes, but it resists simple binary attribution. The biggest variable since we last spoke is that Chinese equity markets have recovered. The economic fundamentals remain poor, real estate is still at the bottom. But every time market sentiment lifts, it shifts the social mood — what Keynes called animal spirits. The critical question, though, is: whose recovery is this? The people benefiting from the AI boom, from this round of capital appreciation and market gains, are a small minority — small enough that they barely register as a statistically meaningful share of society. For the vast majority, this rally has no intersection with daily life. They haven’t actually made money.
Du Lei: And the concentration of wealth this time is even more extreme than during the mobile internet wave in China. In the mobile internet era in the 2010s, people were posting offer letters online to show off — “just graduated, just joined a major tech company, here’s my package.” That kind of showing-off had a social foundation, because the beneficiary pool was wide enough that the envious outnumbered the resentful. Now no one dares showing off.
A Tsinghua or Peking University PhD who lands a spot at a top AI lab doing model training — if they quote their offer number out loud, it will only draw anger. When only a tiny fraction of people have cut themselves a large slice of the pie, everyone else’s first reaction to seeing that slice isn’t “I am inspired by that” — it’s “why you?”
Afra: The asymmetry keeps deepening. The Matthew effect has reached some kind of threshold. That’s true in China — and honestly, even more true in America.
Predictions for the Next Twelve Months
Afra: Last question — any predictions for the next twelve months?
Hua Han: First, the open-source model competition is nowhere near its final form. China will keep pushing hard on open source, and the question worth watching is whether the best domestic open-source models can close or even erase the benchmark gap with closed-source frontier models. That gap currently runs three to six months. Can it be shortened?
Second, semiconductors, Huawei in particular, and the broader domestic compute ecosystem. Open-source models matter precisely because software and hardware are deeply coupled. How far the domestic silicon stack advances will shape everything else.
Du Lei: Three predictions from me. First, the OpenClaw battle will be ferocious, but when the dust settles, the major players’ structural positions will be largely unchanged.
Second, enterprise AI adoption and revenue in China will genuinely break out. Traditional SaaS has always struggled in China — it’s a market where digital transformation was never fully completed. But that’s exactly why AI-native delivery models fit better here. More importantly, Chinese enterprise AI won’t follow the SaaS subscription logic. It will move toward outcome-based “Result as a Service” — I don’t charge you a software license fee; I help you cut five billion in procurement costs by two billion, and I take 5% of the savings. I’m already seeing more Chinese companies skip waiting for mature products and deploy AI agents directly into their production workflows. Bairong (HKEX: 6608) launched exactly this model in December 2025: they deployed voice agents for a major Chinese EV automaker to reactivate dormant customers, and the fee is tied directly to connection rates. No outcome, no payment.
Third, China’s AI for Science track will produce major breakthroughs — I’m not saying we will see Nobel-caliber results in twelve months, but results in genomics, new materials, and agriculture can be quite significant.
Afra: There’s one more thread I’ve been following closely: the intersection of China’s vast manufacturing base with AI, and why that combination is particularly potent here. Du Lei walked me through a great example earlier: a Chinese luggage company that sources whole hides of cowhide for every production run. Natural leather is inherently uneven — some areas are uniform enough for the exterior face of a handbag, others have flaws and can only be used for lining. Traditionally, experienced craftsmen would eyeball each hide and sketch out a cutting pattern by hand, trying to maximize the number of usable pieces while minimizing waste.
The problem is that the optimal layout for leather is actually a complex two-dimensional combinatorial puzzle: an irregular curved surface onto which you need to fit dozens of pieces of varying shapes, where the orientation and placement of each piece affects the final yield. After introducing AI, the factory uses image recognition to scan each hide’s texture and contour, then runs an optimization algorithm to compute the optimal cutting path in real time.
This kind of application is simply invisible to Silicon Valley — because Silicon Valley doesn’t have the leather manufacturing substrate. The problem domain doesn’t exist there.
Du Lei: And the significance extends beyond leather. China has an enormous manufacturing base, and every industry within it carries the same accumulated store of “master craftsman’s intuition” waiting to be translated into AI. Steel heat treatment parameter optimization, food factory quality inspection, textile color consistency control — the pattern repeats across sectors. What these problems share is that they used to require expensive bespoke algorithm development. Now a single person equipped with the right AI tools can tackle them independently. These opportunities will keep surfacing across Chinese manufacturing — and China is exceptionally well-positioned to capture them.
Afra: I’ve been watching Bambu Lab along exactly these lines — they’ve deeply integrated AI image recognition into their 3D printers to monitor the print head for clogs in real time, automatically distinguishing normal operation from the dreaded “spaghetti” failure mode where the print spirals out of control, and halting immediately to prevent further material waste. For the 3D printing industry, it’s a meaningful leap.
Hua Han: Exactly. One final note: the Chinese market will always produce applications that Silicon Valley never saw coming — some of them strange, all of them worth watching. I for one can’t wait.
“You cannot talk about AI without talking about the nation-state’s body, its operating system, the way blood moves through it.”
Insightful.
Same with the Internet, the Internet is a Federation that organically arose from the organic and so Constituted Federation of America.
Would you have any recommendations for reading on how China funds deep tech development and commercialization - both from govt grants or private funding.