In July 2026, a 2.8 trillion parameter model from a Chinese startup few Western executives had heard of a month earlier crashed its own servers under demand, rattled global tech stocks, and prompted a senior OpenAI executive to warn of “full AI communism.” The panic obscures the more useful diagnosis. Moonshot’s Kimi K3 is not a communist weapon. It is a textbook case of predatory pricing executed at civilizational scale, and behavioral economics explains why it is working faster than tariffs, chip bans, or sanctions can stop it. This article applies the zero-price effect from behavioral economics and the solar panel precedent from industrial history to name the three-stage playbook Beijing is running, the Landlord Strategy, and lays out what China’s open-source AI dominance means for any executive whose business model depends on someone eventually paying for AI.
The Landlord AI Strategy
The Weekend the Model Broke Its Own Servers
On the Sunday before markets opened this week, a Beijing-based startup most Western executives had never heard of crashed its own infrastructure. Moonshot AI had just released Kimi K3, a 2.8 trillion parameter open-source model, and demand to use it overwhelmed the company’s servers so badly that new subscriptions had to be paused within 48 hours. By Monday, tech stocks worldwide were sliding on the news, and investors were still working through what a freely downloadable model with frontier-level performance meant for the trillion-dollar bets Silicon Valley has placed on AI.
The reaction from inside the US industry was immediate, and in one case unusually blunt. Dean Ball, a senior OpenAI executive, wrote that China’s strategy of giving away its AI code would likely produce “full AI communism, which is precisely what China proposes.” The phrase traveled fast, because it named a fear that had been building quietly for a year: that China’s open-source AI dominance was no longer a talking point in trade policy briefings but a live threat to the business model every major US AI lab is running.
That fear is well placed. The diagnosis is wrong. What is unfolding is not ideology. It is a pricing strategy Beijing has run to near-monopoly success once already, in an industry that looked nothing like software until it did.
The Numbers Behind China’s Open-Source AI Dominance
Strip away the ideological framing and the numbers alone justify the alarm. Chinese open-source models accounted for roughly 1.2 percent of global AI usage in late 2024. By the end of 2025, that share had climbed to close to 30 percent. Alibaba’s Qwen family of models alone has passed 1 billion cumulative downloads worldwide, spawning more than 180,000 derivative models built on top of it, a rate of adoption that now exceeds Meta’s Llama. China released 1,509 large language models by mid-2025, accounting for 40 percent of every LLM released globally that year, more than any other country including the United States.
The dependency is not confined to hobbyists and open-source enthusiasts. An estimated 80 percent of US AI startups already build products on top of Chinese open-source models, largely because the performance-to-cost ratio is hard to justify ignoring. Harmonic Security, a US-based cybersecurity firm, estimated that one in twelve workers in the US and UK is already using some form of Chinese AI tool at work, a figure that likely undercounts usage buried inside third-party software most employees never realize is built on a Chinese foundation model.
China’s Open-Source AI Dominance, by the Numbers
Source: Moonshot AI, Artificial Analysis, Harmonic Security, RAND Corporation, Alibaba, WAICO founding signatories. July 2026.
Kimi K3 illustrates why the pull is so strong. On the Artificial Analysis Intelligence Index, the model scored 57, ahead of Anthropic’s Claude Opus 4.8 at roughly 56 and OpenAI’s GPT-5.6 Terra at 55. Moonshot’s own pricing data put the model at $0.95 per intelligence-index task against $1.04 for OpenAI’s comparable offering. Kimi K3 is not a discount alternative to the frontier. It is the frontier, and it is free to download, inspect, and modify.
Why Free Beats Better: The Zero-Price Effect
Executives trained to think in terms of total cost of ownership tend to underestimate what happens when a product’s price hits exactly zero. Behavioral economist Dan Ariely demonstrated the effect in a now well-known series of experiments: when the price of one option drops from a small positive number to zero, demand does not rise proportionally, it jumps disproportionately, because free removes the psychological cost of a bad decision along with the financial one. Ariely called it the zero-price effect, and it explains a pattern a straightforward cost-benefit model cannot: consumers and developers alike will often choose an inferior free option over a modestly priced superior one, because free eliminates the fear of loss entirely.
Kimi K3 does not require that trade-off. It offers frontier-comparable performance and a zero marginal price for anyone willing to self-host it, precisely the combination the zero-price effect predicts will trigger demand disproportionate to the underlying quality gap. A developer choosing between a slightly better paid API and a free, open-weight model performing at 98 percent of its quality is not making a rational value calculation anymore. They are responding to an asymmetry in perceived risk that Beijing’s subsidy structure was built to exploit.
This is the mechanism Western labs have consistently underestimated. OpenAI and Anthropic are competing on capability. Moonshot and Alibaba are competing on the psychology of acquisition, and psychology, not capability, decides adoption curves in the early stage of any platform war.
We Have Run This Play Before: The Solar Precedent
None of this is unprecedented, and that should worry Western boardrooms more than the novelty of the technology does. A decade ago, China supplied roughly 40 percent of the world’s solar panels. Today its share is above 80 percent, and the country now controls 93 percent of global polysilicon production, 97 percent of wafer manufacturing, and 92 percent of solar cell production. The mechanism was not a technological breakthrough. It was state-subsidized overcapacity that drove global prices down faster than competitors in the US, Germany, and elsewhere could sustain margins to compete, followed by consolidation of the entire supply chain once rivals had exited.
Richard Windsor, an independent technology analyst, has already named the parallel directly: China’s AI strategy “looks like a rinse-and-repeat of its highly successful takeover of the solar panel industry.” The playbook does not require ideological conversion. It requires patience, state balance sheet support, and a target industry where price sensitivity outweighs brand loyalty in the near term. Frontier AI, where enterprise customers are actively hunting for ways to cut inference costs, fits that description precisely.
The uncomfortable lesson from solar is not that China cheated. It is that Western competitors correctly identified the strategy in real time and still lost the market, because responding to subsidized overcapacity with anything short of matching state support or radical differentiation has never worked.
The Landlord Strategy: Subsidize, Standardize, Extract
The pattern across both industries resolves into a three-stage model worth naming, because naming a strategy is the first step to defending against it. Call it the Landlord Strategy: give away the building today to control the rent roll tomorrow.
Stage one is subsidize. Beijing backs frontier-grade model development and compute access at a scale no venture-funded lab can match indefinitely, and the resulting models are released as open weights, free to download and modify, undercutting Western subscription pricing to zero for any organization willing to self-host.
Stage two is standardize. Once a model family becomes the default substrate developers build on, in Qwen’s case more than 180,000 derivative projects, it becomes structurally difficult to displace. That standardization is now being extended at the state level: 29 countries, including Indonesia, Brazil, South Africa, and Pakistan, signed onto the China-led World Artificial Intelligence Cooperation Organization this month. Not one major Western democracy joined.
Stage three is extract. This is the stage that has not fully arrived, which is exactly why it is worth naming now rather than after the fact. Alan Woodward, a cybersecurity expert at the University of Surrey, put it plainly: China’s current openness “looks like a strategic decision that could see the Chinese dominate in the long term, at which point they could well start to be more controlling.” Ciaran Martin, the former head of the UK’s National Cyber Security Centre, has flagged a related risk already live today: an open-weight model with frontier capability cannot be gatekept the way a closed model can, so security exposure scales with adoption in a way no single company or government can contain retroactively.
The Landlord Strategy: Give Away the Building, Control the Rent Roll
Extraction does not require a single dramatic policy change. It requires only that enough of the world’s AI workflows already run on Chinese-controlled infrastructure by the time the terms shift, so that reversing course costs more than accepting them.
The Business Model This Breaks
The direct casualty of this strategy is the subscription economics that OpenAI and Anthropic have built their valuations on. Both companies are underwriting enormous infrastructure spend against the bet that enterprise and consumer customers will keep paying premium prices for frontier intelligence. That bet gets harder to defend every time a free, self-hostable alternative closes the performance gap to a rounding error.
The scale mismatch here cuts both directions and is worth stating honestly rather than one-sidedly. US hyperscalers, Alphabet, Amazon, Meta, and Microsoft among them, have committed roughly $650 billion in AI infrastructure spending this year alone. Alibaba’s announced AI investment, by comparison, is about $53 billion over three years. On raw capital, the United States is not losing this race. What it is losing is the distribution race, and distribution, not compute, is what decided the solar outcome.
There is also a legitimate counter-case that deserves airtime rather than dismissal. Open, cheap AI genuinely benefits developers and enterprises outside the largest technology budgets, and a market where only two or three US labs gatekeep frontier intelligence carries its own concentration risk, the kind of principal-agent problem any board would recognize in miniature. Cutting off access to competitive open models, as some in Washington are now proposing, could simply reproduce the closed-market dynamics that made the last AI cycle so vulnerable to price and trust erosion in the first place. The right response to a subsidized competitor is not always a wall. Sometimes it is a business model that no longer depends on a scarcity that has stopped existing.
Why “AI Communism” Is the Wrong Diagnosis for China’s Open-Source AI Dominance
Dean Ball’s framing captured attention because ideology is an easy story. It is also the wrong one, and getting the diagnosis wrong leads executives toward the wrong defense. This is not communism in any meaningful economic sense. The Chinese state is deploying capital-intensive industrial policy to capture a strategic supply chain, exactly as it did with solar panels, batteries, and increasingly electric vehicles. Alastair Paterson, chief executive of Harmonic Security, offered a sharper and more useful warning when he noted that data submitted to these platforms should be considered the property of the Chinese Communist Party given a lack of transparency. That is not an ideological claim. It is a data governance and vendor risk claim, and it belongs in the same conversation your board already has about cloud concentration risk and single-vendor exposure.
Treating this as an ideological contest invites Western labs and policymakers to fight the wrong battle, with export controls and subscription pricing, when the actual battle is over default infrastructure and switching costs. Export controls slow the chip supply. They do not touch a model that has already been downloaded, forked, and embedded into a developer’s toolchain in more than 30 countries. The fight China is actually running is a distribution and dependency fight, and it will not be won or lost on the ideological terrain where the loudest reactions are currently being fought.
What This Means for the C-Suite
Three implications follow directly from the data, and none of them are procurement decisions that can be delegated to an IT department.
First, model dependency is now a board-level supply chain risk. If one in twelve of your employees is already using a Chinese AI tool, sanctioned or shadow IT, that exposure sits on your risk register today, not on a future roadmap. Most executive teams cannot currently answer the basic diagnostic question of what percentage of their AI-dependent workflows run, even indirectly, on Chinese-origin open-weight models embedded inside third-party software.
Second, pricing power in commodity AI inference is eroding faster than most five-year plans account for. Competitive advantage has to move up the stack, into workflow integration, proprietary data, and trust, because raw model access is heading toward a price of zero for a widening set of tasks.
Third, the Global South’s AI dependency curve is being set right now, not negotiated later. Every company with operations or customers across the 29 WAICO member states inherits the infrastructure choices those governments are making today, whether or not the company had any voice in the decision.
A useful diagnostic question for any executive team: if the free, frontier-grade alternative to your current AI vendor improved by another 10 percent next quarter, does your competitive position survive the shift, or does it depend on a price gap that history says China intends to close?
The Move Ahead
China didn’t need to build a better model. It needed a free one, and free just rewrote the economics of the AI race. The executives who treat that as an ideological skirmish will spend the next year fighting the wrong opponent. The ones who treat it as what it actually is, a patient, subsidized bid for default global infrastructure, have a real chance to reposition before the terms change.
nStratagem works with executive teams navigating exactly this kind of structural shift, translating geopolitical and technological disruption into board-ready strategy before the market forces the decision for you. Learn more at nstratagem.com.
The analysis published in The Norm Report is intended for senior executive and board-level audiences as strategic intelligence and editorial commentary. It does not constitute legal, financial, investment, compliance, or regulatory advice. © 2026 Norm Murray / nStratagem. All Rights Reserved.
