On June 1, Beijing issued two seemingly contradictory directives that, read together, reveal the clearest picture yet of China's AI strategy. The first: new rules tightening government control over overseas deals involving Chinese investors, technology, data, and anything touching national security. The second: a state-backed push to turn offshore data centers in Shantou into export hubs for AI tokens — turning electricity and compute into a high-value product for Southeast Asia.
One policy says "nothing leaves without permission." The other says "we're selling AI by the token to our neighbors." The contradiction is only surface-deep. What Beijing is actually building is a controlled funnel: lock down the technology and talent that power the models, then monetize the output through state-sanctioned channels. China doesn't want its chips, researchers, or training data walking out the door. But it very much wants Southeast Asian businesses paying for Chinese AI compute, metered by the API call.
This is not openness. It is enclosure with a billing system.
The Foreign Deal Curbs: Sealing the exits
The new rules, published by multiple agencies including the Ministry of Commerce and the Cyberspace Administration of China, expand review requirements for a sweeping range of cross-border transactions. Any overseas deal involving Chinese investors, the transfer of technology or data, or any whiff of national security now faces mandatory government screening. The rules are deliberately broad — "national security" is defined to cover not just defense-adjacent tech but AI models, algorithms, training datasets, and even the metadata that describes them.
The timing is not accidental. The curbs arrive barely a month after Beijing blocked Meta's attempt to acquire Chinese AI startup Manus, a move that signaled China's willingness to use deal review as a tool to prevent foreign access to domestic AI capabilities. The June 1 rules formalize that posture into a standing regime. Where the Manus block was a one-off intervention, the new framework is a permanent filter on the outbound flow of AI-related assets.
For Chinese tech companies, the implications are immediate and painful. Any venture that has taken foreign investment — and most have — now faces a new layer of scrutiny when engaging in overseas partnerships, licensing deals, or joint ventures. The rules effectively give Beijing veto power over any transaction that might transfer AI know-how outside the country, regardless of whether the counterparty is American, European, or Southeast Asian.
Even purely commercial deals are caught in the net. A Chinese AI lab licensing its model to a Thai fintech startup? That now requires review. A data center joint venture in Malaysia with a Chinese cloud provider? Review. A Chinese researcher taking a consulting role with a Singaporean AI company? Potentially reviewable, depending on what they know and what they might transfer.
The message is unambiguous: China's AI capabilities are a strategic reserve, not a commercial product to be traded freely. The state will decide what can leave, when it can leave, and at what price.
The Token Goldmine: Selling AI by the API call
While the exits are being sealed, a new revenue door is swinging open. State media, including People's Daily and China Daily, have begun promoting a concept that would have seemed absurd two years ago: China as an exporter of AI tokens.
The pitch is simple. China has built massive AI infrastructure — data centers, GPU clusters, model training pipelines — and much of it is underutilized or operating at low margins serving domestic customers. Meanwhile, Southeast Asia is hungry for AI capabilities but lacks the capital and technical expertise to build them domestically. The solution? Export not the models, not the chips, not the researchers, but the output: API calls, inference tokens, generated text, embeddings, and synthetic data.
The mechanism is equally straightforward. Offshore data centers in Shantou, a coastal city in Guangdong province, are being positioned as the export hub. These facilities sit outside China's main firewall infrastructure — or at least on a controlled edge of it — allowing them to serve international customers with low latency while remaining under Chinese jurisdiction. The models running in these centers are Chinese. The compute is Chinese. The electricity is Chinese. But the customer is Thai, Indonesian, Vietnamese, or Malaysian.
From Beijing's perspective, this is a near-perfect arrangement. The valuable assets — the model weights, the training data, the algorithmic innovations — never leave Chinese soil. They remain in data centers that the state can monitor, regulate, and if necessary, shut down. What crosses the border is ephemeral, metered, and entirely dependent on continued Chinese infrastructure. If a Southeast Asian customer stops paying, the tokens stop flowing. If a customer misbehaves, their API key is revoked. The relationship is structurally asymmetric in China's favor.
And the economics are compelling. AI tokens are high-margin digital goods. The marginal cost of an additional inference call is a fraction of a cent in electricity and depreciation. The price charged to a Southeast Asian enterprise customer might be ten or a hundred times that. The difference is profit that accrues to Chinese infrastructure providers, Chinese cloud platforms, and ultimately to the Chinese state through taxes, equity stakes, and regulatory leverage.
Two sides of the same coin
The foreign deal curbs and the token export push are not contradictory. They are complementary. Together, they describe a strategy of controlled openness — a model where China engages with the global economy on terms that preserve state control over strategic assets.
Think of it as the AI equivalent of China's approach to rare earth minerals. Beijing doesn't sell the mines. It doesn't export the refining technology. But it will sell you the processed material, at a price it sets, in quantities it controls, with the implicit threat that supply can be cut off if you misbehave geopolitically. The token economy is the same playbook applied to intelligence rather than minerals.
The curbs ensure that the underlying technology — the models, the data, the training pipelines, the human expertise — remains trapped in China. The token exports monetize that trapped asset without giving away the asset itself. It is a rentier model for the AI age: China owns the means of production, and Southeast Asia rents the output.
This has profound implications for the region. Southeast Asian countries have been actively courting Chinese tech investment, viewing it as a shortcut to AI capabilities they cannot build themselves. Singapore, Malaysia, Thailand, and Indonesia have all welcomed Chinese data center projects, cloud partnerships, and AI research collaborations. The token export model fits neatly into this pattern — it looks like partnership, but the power dynamics are fundamentally colonial.
The dependency being created is not just economic. It is epistemic. As Southeast Asian businesses integrate Chinese AI tokens into their workflows — customer service chatbots, document analysis, code generation, financial modeling — they become dependent on Chinese models for their cognitive infrastructure. The underlying assumptions, biases, and blind spots of Chinese-trained models become embedded in Southeast Asian business processes. Over time, this shapes decision-making in ways that align with Chinese interests, not necessarily with local ones.
Why Southeast Asia is buying in
The appeal for Southeast Asian customers is genuine and immediate. Building frontier AI capabilities from scratch requires billions of dollars, thousands of GPUs, and access to technical talent that is concentrated in the US and China. For a mid-sized Thai bank or an Indonesian e-commerce platform, the choice is not between Chinese tokens and self-built models. It is between Chinese tokens and American tokens, or between Chinese tokens and no AI at all.
Chinese tokens have advantages. They are cheaper, often dramatically so. Chinese providers are willing to price aggressively to gain market share, subsidizing inference costs with state backing. They are also more willing to customize — to fine-tune models on local languages, local regulatory frameworks, and local business contexts. A Chinese provider will happily train a model on Thai financial regulations or Indonesian consumer behavior in a way that American providers, focused on larger markets, might not.
And there is a geopolitical dimension. For Southeast Asian governments wary of American dominance in digital infrastructure, Chinese AI offers a diversification play. If your cloud is already Alibaba, your 5G is already Huawei, and your e-commerce is already Shopee (Tencent-backed), adding Chinese AI tokens is a logical extension of an existing dependency, not a new risk.
But the long-term cost is strategic vulnerability. The more deeply integrated Chinese AI becomes into Southeast Asian economies, the harder it becomes to decouple. A bank that runs its customer service on Chinese tokens cannot switch providers overnight. A government that uses Chinese models for document analysis cannot easily audit what the model is doing. An e-commerce platform that relies on Chinese embeddings for product search is structurally tied to Chinese infrastructure.
The global AI supply chain splits further
These June 1 moves accelerate a bifurcation that has been underway for years. The global AI ecosystem is splitting into two parallel systems: one American-led, one Chinese-led, with decreasing overlap and increasing friction.
On the American side, export controls have choked off Chinese access to advanced chips. Nvidia's latest architectures are banned from direct sale to China, and the Commerce Department is now closing loopholes that allowed Chinese subsidiaries in third countries to acquire them. The US has also restricted Chinese AI labs from accessing American cloud infrastructure and has pressured allies to limit Chinese tech investment.
On the Chinese side, the response is the strategy we see formalized on June 1: lock down domestic capabilities, prevent foreign acquisition of Chinese tech, and build alternative export channels that monetize AI without transferring it. The token goldmine is the monetization layer. The foreign deal curbs are the enforcement layer.
For the rest of the world, the choices are narrowing. Countries that want access to American AI technology must align with US export controls and limit engagement with Chinese entities. Those that want Chinese market access and cheap AI infrastructure must accept the dependencies that come with it. Neutrality is becoming harder to maintain, and the middle ground is shrinking.
Southeast Asia, in particular, is caught in the crossfire. The region has benefited from investment from both sides, with American chipmakers building packaging facilities while Chinese firms pour money into data centers and cloud infrastructure. Now, with the US signaling tighter restrictions and China offering subsidized tokens, these countries may be forced to choose — or find themselves dependent on both in ways that leave them vulnerable to pressure from either.
🔥 Hot Takes
1. The "token goldmine" is digital colonialism with better marketing. Beijing isn't exporting AI partnership — it's exporting dependency. Southeast Asian countries are being sold a shortcut to modernity that comes with a structural leash. The tokens flow as long as the political relationship is warm. The moment it sours, the API calls get throttled, the models get updated with "regional maintenance," and the customer discovers they don't actually own anything. This is the Opium War playbook updated for the API economy: create dependency, then extract rent. The only difference is this time the product is intelligence, not narcotics, and the addiction is economic, not chemical.
2. The foreign deal curbs prove that China's AI self-sufficiency narrative is mostly theater. If Chinese AI were truly world-beating and self-sufficient, Beijing wouldn't need to panic-lock the exits. The fact that they're slapping emergency controls on technology transfer tells you everything about where they think they stand. They're terrified that their best researchers, their most promising startups, and their most valuable models will walk out the door the moment they get a better offer from Singapore, Dubai, or London. Confidence doesn't build walls — insecurity does. These curbs are an admission that China's AI ecosystem is still fragile enough to hemorrhage talent and IP if the exits aren't guarded.
3. Southeast Asia is sleepwalking into the most lopsided technology dependency since the colonial era. The region's leaders talk about digital sovereignty and AI sovereignty, but their procurement decisions tell a different story. Every Chinese token consumed is a brick in a wall that will eventually enclose Southeast Asian digital economies. In ten years, when these countries try to assert independent AI policy, they'll discover that their banks, their governments, and their platforms are running on Chinese infrastructure they cannot replace without catastrophic disruption. The token goldmine isn't a business opportunity for Southeast Asia. It's a debt trap dressed as a service contract, and the bill comes due in geopolitical leverage.
The Bottom Line
Beijing's June 1 directives are not a contradiction. They are a coherent strategy for the AI age: control the means of intelligence production, prevent foreign access to the underlying technology, and monetize the output through state-sanctioned channels that create lasting dependency.
The foreign deal curbs ensure that China's AI capabilities remain trapped inside the fortress. The token goldmine ensures that those trapped assets still generate revenue and geopolitical leverage. It is enclosure economics applied to artificial intelligence: the commons are fenced off, the shepherds are barred from leaving, and the wool is sold to neighbors who are slowly convinced they cannot survive winter without it.
For Southeast Asia, the choice is not between Chinese and American AI. It is between dependency and the hard work of building genuine digital sovereignty. The tokens are cheap now. The price will be paid later, in leverage, in vulnerability, and in the slow realization that the "partnership" was never a partnership at all.
The loopholes are closing. The walls are going up. And the world is entering an era where AI is not traded freely but rationed by the state, metered by the token, and weaponized by the geopolitical bloc that controls the infrastructure. The goldmine is real. But the miners are not the ones getting rich.