China has to date run two large-scale experiments in capturing global technology market share. Call them the Huawei model and the TikTok model. The contrast between them explains why Chinese AI labs — DeepSeek, Qwen, Kimi, GLM — are succeeding today, and why their winning streak may eventually hit the same impasse TikTok did.
By the mid-2010s, Huawei had built genuinely competitive wireless infrastructure, undercutting Ericsson and Nokia on price and often on performance, and it set out to become the default supplier of 5G worldwide. Then came the first Trump administration’s trade war — export controls, entity list designations, pressure on allies to exclude Huawei gear — and its global ambitions stalled.
Sanctions mattered enormously, but Huawei’s model had a structural vulnerability that made it unusually easy to sanction: it never reached the end user directly. Between Huawei and the handset sat a national telecom carrier, typically licensed by, and close to, its home government. Cut off the carrier, and Huawei’s technology never reached a single phone — a single choke point, which is exactly what trade policy is built to exploit.
TikTok took the opposite path. Riding the open global internet — ironically, one China tightly restricts at home — ByteDance, following the YouTube model, built a product that went straight into users’ pockets: no carrier, no IT department, no gatekeeper. Once hundreds of millions had downloaded it, posting and building livelihoods on the platform, it became sticky in a way infrastructure never could.
You can ban a carrier from buying routers. Pulling an app off that many phones, however, is a political fight, because by then it isn’t really ByteDance’s user base — it’s people with genuine attachment to a product and community. Washington’s actual response bears this out: not a clean ban, but a drawn-out negotiation over ownership and corporate structure, trying to sever the platform from ByteDance while keeping the app and its users intact.
China‘s leading AI labs seem to have absorbed the lesson — reinforced by how the U.S. firm OpenAI’s direct-to-consumer ChatGPT launch reshaped the market in late 2022. Where ChatGPT stayed proprietary, DeepSeek, Qwen, Kimi, and GLM are shipping open-weight models anyone can download, alongside cheap or free consumer apps and APIs anyone can adopt in a few clicks. No carrier-equivalent gatekeeper stands in between. A developer in Bangalore or a hobbyist in Berlin can pull DeepSeek’s weights or call Kimi’s API without asking any regulator’s permission.
The playbook is working: open-weight Chinese models post startling download and fine-tune numbers on Hugging Face, and low pricing pulls in cost-sensitive developers who’d otherwise default to Western labs. Every integration built on top adds a sliver of stickiness — switching costs that, multiplied across millions of users, start to look more like TikTok than Huawei.
The open question is whether this keeps stretching or eventually meets its own TikTok endgame. Open-weight distribution is harder to choke at a single point than carrier-mediated infrastructure, but it isn’t immune to pressure: cloud providers can decline to host models, app stores can delist apps, governments can restrict foreign artificial intelligence in sensitive sectors or procurement — the kind of security carve-outs already surfacing in debates over data handling and model provenance.
Much depends on how the world ultimately judges this: fair competition on price and openness, or a security loophole originating from China. Beijing is already working that judgment at the institutional level, not just the product level. This past week it launched the World Artificial Intelligence Cooperation Organization, a new Shanghai-headquartered body signed into being by 29 founding countries, including Russia, Brazil, Indonesia, Pakistan, and a large bloc of African and Asian states.
It’s styled as an independent intergovernmental body committed to U.N. Charter principles, framed around expanding AI access for the “global south” — Xi Jinping pledged thousands of AI training slots for developing nations at the launch. But analysts have also described WAICO as a governance track built deliberately outside Western-led frameworks like the EU’s AI Act and the G7 process, giving Beijing a forum to set standards and distribute access on its own terms.
That’s the go-to-market logic extended one level up: If DeepSeek and Kimi are winning users one download at a time, WAICO is an attempt to win governments and standards-setting influence at the same time, in the same global south markets where Chinese labs already have the strongest developer traction. AI models add a wrinkle video apps didn’t have — they get embedded deep inside other companies’ products, multiplying the surface area regulators might want to inspect.
THE ‘AI KILLS JOBS’ MYTH IS COLLAPSING — GREAT NEWS FOR JOB SEEKERS
If restrictions come, expect them to look less like a Huawei-style equipment ban and more like the TikTok scenario — a negotiation over structure, hosting, and data flows rather than a clean prohibition, because by the time anyone seriously considers restricting these models, too many users and developers will already depend on them.
Whether that negotiation ends in divestiture, licensing, sovereign forks, or nothing at all remains open. But the direct-to-user strategy has already proven far more durable than the carrier-mediated one — and that durability is exactly what will make the next round of scrutiny so hard to resolve cleanly.
David W. Wang is a senior international business executive, geopolitical affairs consultant, analyst, and writer based in the Washington, D.C., metro area. David can be reached on X @DavidWWang203.