In the Sovereign-AI Wave, Taiwan's Real Leverage Isn't "Its Own Model" — It's Being the Node the World Can't Route Around
Over the past year or two, "sovereign AI" has become the hottest phrase on the lips of governments everywhere.
The logic isn't hard: when AI becomes infrastructure — like electricity, like national defense — no country wants to hand its own brain to a company from another country. That fear has a very concrete name: "digital colonization." Ask a foreign mainstream AI what Taiwan's legal obligations are, or Taiwan's historical position, and you often get an answer that reads like "simplified Chinese converted to traditional," or one carrying an outside perspective that isn't quite Taiwan's own. For a country, that isn't just a clunky user experience — it's your power to interpret the world, held in someone else's hands.
So nations are racing for three things: their own compute, their own data, their own models. South Korea bought 260,000 GPUs from Nvidia in one move to build its own sovereign-AI infrastructure; Saudi Arabia and the UAE are pouring state money into "Arabic-first" models; India is building supercomputers while trying to bind data and compute together; Japan is subsidizing domestic supercomputers, explicitly to reduce dependence on U.S. technology. Nvidia's Jensen Huang put it bluntly: the first AI wave was driven by tech giants; the second will be driven by governments.
Taiwan has boarded this train too. In early 2026 the Basic Act on Artificial Intelligence took effect; the government made sovereign AI a national strategy, declaring it will become "one of the world's top five compute centers," pouring over NT$190 billion into a new AI infrastructure plan; the National Science and Technology Council's TAIDE aims to build a large language model that "understands Taiwanese," trained on local corpora — and the Ministry of Digital Affairs has even opened its Taiwan corpus for download by international players like Google, so that Taiwan's values live inside the world's AI too.
The direction is right. But amid all the voices saying "we too must have our own model, our own compute," there's one point that gets easily buried under the heat.
For a small country like Taiwan, the most crucial thing about sovereign AI may not be "we must also have our own large model."
Because if the focus is only on "every country building its own ChatGPT," Taiwan will struggle — we don't have the data scale, the capital, or the electricity to fight the U.S. and China for an all-purpose general model. On that track, a small country can hardly catch a large one. That's not a matter of ambition; it's a matter of size.
And this is something Lee Kuan Yew answered for me more than twenty years ago.
Back then I asked him: how does a country as small as Singapore survive among great powers, and still be heard? His answer wasn't "make yourself as strong as the giants" — impossible. His answer was: make yourself the party the great powers cannot afford not to deal with. Singapore had no hinterland, no resources; it relied on geography, on being a financial hub, on that node value where "without you, things stop turning."
Taiwan, for the past twenty years, has used the same logic — only the leverage became semiconductors. More than ninety percent of the world's advanced chips pass through this island, so no one can talk about the future of technology while routing around Taiwan. Not because Taiwan is "big," but because at that link, Taiwan is irreplaceable.
In the AI era, the problem hasn't changed. Only the leverage needs to be swapped once more.
What Taiwan should really be seizing isn't "owning a model" — it's becoming the node no one can route around in the global map of AI deployment.
And here, Taiwan actually holds cards. As base models become a commodity everyone has, the truly scarce ability moves in two directions.
The first is the ability to "put AI into the physical world." Software chatbots, everyone can build. But to get AI into a surgical medical device, into a factory's automated line, into a robot moving through the real world — that takes extremely tight software-hardware integration, and the most complete semiconductor and systems manufacturing chain. That is exactly what Taiwan has built over decades, and what others can't catch up to soon. When the world's AI moves from "the cloud" to "the physical," Taiwan is the place most capable of installing it.
The second is "proprietary data." When the model itself is no longer scarce, data becomes the real moat. In healthcare, in manufacturing, in public governance, Taiwan has accumulated large volumes of high-quality data that other countries can't replicate. What that data trains isn't one more general model — it's expert models so strong in specific domains that anyone trying to solve the same problem can't route around you.
In other words, Taiwan's position in the AI arms race shouldn't be to become "another country straining to build a big model," but to become "the piece the world's AI can't route around when it wants to land." One is competing on size with giants, and hard to win. The other is making yourself irreplaceable — which is how small countries have always won.
Which brings me to a further point, because it involves two terms that sound alike but are very different.
One is "sovereign AI." The other is "AI sovereignty."
"Sovereign AI" is about owning — do you have your own model, compute, data; behind closed doors, can you run on your own? That matters. It's defense — keeping your door shut, making sure you can't be choked at a critical moment. But it has a ceiling: however well a small country shuts its door, the model behind it may still not beat the giants outside.
"AI sovereignty" is about influence — can you use AI to keep this country a position that cannot be ignored in the world order to come? That's offense — making others' doors open only for you.
The former is guarding your own door; the latter is making yourself the door others can't get past. For Taiwan, what truly decides the long game is the latter.
Lee Kuan Yew's generation guarded Singapore with irreplaceability. Semiconductors keep Taiwan un-routable-around in this era. In this round called AI, the question Taiwan should ask itself may not be "do we have our own model," but — on the day all the world's AI must walk into the physical world, which piece is the one they cannot, no matter what, route around Taiwan for?
Answer that, and you hold leverage. Fail to, and however many compute centers you build, you're just a pawn on someone else's board.