Civics · Law · Geopolitics: Technology and Geopolitics

Advanced chips are the "general-purpose foundation technology" of our age — like electricity a century ago, almost every modern industry and military system is built on them. What matters is not how strong any single country is, but that the production chain is extremely specialized yet highly concentrated: the core design tools and architectures, the extreme-ultraviolet lithography machines (EUV, made by only one firm worldwide, the Netherlands' ASML) required to make the most advanced chips, the most cutting-edge wafer fabrication (concentrated in Taiwan and a few other places), and some critical materials (Japan holds a major share) — no single country holds the whole chain. This "everyone depends, each controls one link" structure turns every link into a chokepoint: block one node and the whole chain can stall. The chip war is, at bottom, the fight over these chokepoints.

• Export controls (choking the chain): From October 2022 the US tightened rules restricting exports to China of the most advanced chips, lithography machines and design software — the logic being "turn the link others depend on into strategic leverage." The Netherlands and Japan later made their own control arrangements on advanced equipment.
• Industrial subsidies (building the chain): Both the US and the EU passed "Chips Acts," using large subsidies to draw advanced manufacturing back onshore — the logic being "keep critical capacity in your own hands."
• Import substitution (patching the chain): The side being choked pours resources into home-grown R&D and domestic supply chains, seeking self-sufficiency at the key links. Each approach has its own cost: controls hurt one's own exporters and the efficiency of the global division of labor; subsidies are expensive and may not out-compete existing clusters; self-reliance takes years and is very costly in the short run.

Efficiency vs. security — both logics hold. For the global division of labor (steelman): chips are the ultimate product of human cooperative specialization; giving each link to whoever does it best is what delivers today's performance and low prices. Forced "decoupling" and rebuilding make everyone pay for redundancy and slow the whole march of technology. For self-reliance (steelman): when a technology bears on national security, placing your lifeline in the hands of a potential rival or a single node hands away your strategic choices — one conflict or natural disaster can cut off supply, and redundancy is not waste but insurance.

Misconception: "The chip war is just about who can make the most advanced chip." A process-node lead matters, but the real lever is control of the chokepoint links — one side holds EUV, another holds high-end foundries, and neither can do without the other. Nor is it only the most cutting-edge chips that count: cars, home appliances and defense use vast quantities of mature-node chips, and if these "unremarkable" supply chains break, the impact can be even broader.

The internet was originally designed to be borderless: data packets ask for no passport and flow freely across borders. But a sovereign state's instinct is "my territory, my rules," so states increasingly assert that the data, network infrastructure, platforms and content within their borders should also fall under their jurisdiction. This is digital sovereignty — moving the physical world's "territory-jurisdiction" logic into the space made of bits. Its levers include data localization, controlling cross-border data flows, constraining foreign platforms and holding key digital infrastructure. The core tension: a system that is technically and naturally globally interconnected meets a world that is politically divided by borders.

An open global network vs. territorialized control — each has its strongest logic. For openness (steelman): the internet's value comes precisely from borderless interconnection; free data flow brings innovation, trade and exchange of ideas, while wall upon wall cuts the network into mutually disconnected fragments (the "splinternet"), leaving everyone poorer. For sovereignty (steelman): letting data and key platforms escape domestic law is to give up protecting citizens' privacy and national security; assertion of sovereignty is not about cutting off the internet, but about building accountable rules in the digital space too. The compromise lies in which flows should stay open and which links must be controllable.

Misconception: "Digital sovereignty just means censorship and blocking the net." That is only one end of the spectrum. Digital sovereignty is a wide spectrum: at one end is the drive for "autonomy" aimed at protecting privacy and shedding dependence on foreign tech giants (as with the EU promoting home-grown cloud computing); only at the other end is "control" centered on content management. Conflating the two obscures the fact that different countries are really trying to solve different problems.

Artificial intelligence poses a classic governance puzzle — the pacing problem: the technology iterates by the month, legislation by the year, and by the time rules are written the target has already changed shape. On top of that, AI is a general-purpose technology (one model can serve medicine, hiring, the military, creative work), so its risks and benefits are diffuse and hard to gauge, and it is inherently cross-border. So countries disagree not on "whether to govern," but on what philosophy to govern with: rules up front or accountability after the fact? A single horizontal law, or sector-by-sector governance?

• EU: risk tiers, ex-ante rules. The AI Act, in force since August 2024, sorts AI into four risk tiers by risk — unacceptable (e.g. social scoring, banned), high (e.g. hiring, medical, must meet strict obligations), limited (e.g. chatbots, must tell users they are talking to a machine) and minimal (largely unregulated). The higher the risk, the more the rules come up front.
• US: sectoral, largely light-touch. It leans against a single unified statute, relying on existing sector regulators and voluntary corporate commitments — the logic being "don't cage a technology in its infancy."
• Some states: rules for specific applications. They issue dedicated rules for particular settings like recommendation algorithms and generative AI, often dovetailing with existing content and data governance. Each path has its cost: too strict stifles innovation, too loose lets harm run.

Prudent precaution vs. permissionless innovation — both stand. For precaution (steelman): bias, deepfakes and loss-of-control risks, once they cause real harm, are often hard to reverse; patching up after the fact is too late, and putting guardrails in place before the technology sets is cheapest. For innovation (steelman): laying down rigid rules too early for a technology not yet understood can strangle the best opportunities in the cradle, and one-size-fits-all can wrongly kill harmless uses; better to let the technology run and target confirmed, specific harms precisely. What is really at stake is which way to set the direction of error in the face of uncertainty.

Misconception: "AI governance is about stopping robots from waking up, preventing 'the Terminator.'" Long-run loss-of-control risks are seriously discussed by some, but the great majority of governance work today deals with problems right in front of us: algorithmic discrimination, opaque decisions, data misuse, allocation of responsibility, disinformation. Imagining AI governance only as a sci-fi scenario ignores these issues that are happening right now and truly affect ordinary people.

A standard is a set of shared technical conventions that let different products work together — the shape of a plug, a communication protocol, a file format. It looks dull, but it holds enormous power: whoever's technology is written into the standard sees their approach become the default that everyone else must adapt to; whoever holds "standard-essential patents" (patents unavoidable in implementing the standard) can collect licensing fees on FRAND (fair, reasonable, non-discriminatory) terms; and once a standard is widely adopted, path dependence locks it in for the long run. So the battle over standards is a fight over who writes the technological world's "common language." The venues for setting them include the International Telecommunication Union (ITU), 3GPP and the IETF — historically industry-consensus-driven and bottom-up.

• 5G mobile communications: A single standard set by 3GPP, one network worldwide. The fight over shares of its standard-essential patents is fierce; Chinese firms (such as Huawei) became major patent holders — which is why the 5G contest takes on a geopolitical hue, since patent share maps directly onto long-run economic returns and a say in the rules.
• A historical mirror: Standards battles are old. In the early videotape era, VHS beat Betamax — the winner often decided not by whose technology was better, but by who could build the larger coalition and seize network effects. In recent years countries have more actively pushed their own proposals within international standards bodies; whoever leads on standards gets a head start on the next generation of technology.

A single global standard vs. competing/national standards — each has its cost. For unity (steelman): one globally common standard delivers the greatest interoperability and economies of scale, lowest cost and largest market, and every device in the world can talk to every other. For plurality (steelman): handing standards entirely to a single system is to outsource the power to define a key technology; once dominated by a few parties, other countries are both dependent and unable to guarantee security — keeping competition and a domestic backup is a necessary cost of strategic autonomy. The trade-off is how much "autonomy and security" to surrender for "interoperability and cheapness."

Misconception: "Technical standards are purely technical, neutral and objective." Setting standards is riddled with interested choices: whose patents to adopt, whose existing products to accommodate, which performance to prioritize. That is also why "the technically best option doesn't always win" — network effects, coalition size and first-mover advantage often decide the outcome more than technical specs, and behind it all lies an economic and geopolitical contest.