AI chip export controls are colliding with subsidies as governments treat leading compute as a strategic asset, not just a commercial input. In late March 2026, South Korea moved to back domestic AI-chip maker Rebellions with a 250 billion won (about $166 million) investment approval, while China’s chip sector accelerated capacity plans in mature nodes as AI infrastructure demand strains parts of the supply chain. In the U.S., policymakers are still revising how export rules should work, with a proposed AI-chip export rule pulled after appearing in the federal rulemaking review process.
What changed across three major chip blocs
South Korea’s Industry Ministry said the Financial Services Commission’s advisory board approved a 250 billion won investment into Rebellions as part of a government-backed push to nurture an advanced semiconductor firm. The approval was reported on March 26, 2026.
In China, executives and analysts described a rapid build-out tied to global AI infrastructure demand, with Reuters reporting that China’s manufacturing capacity for chips on mature nodes (such as 22nm to 40nm) is expected to rise from 37% of global output in 2026 to 42% by 2028. The same reporting described bottlenecks in areas that support AI systems—testing, packaging, optical interconnects, and materials—creating backlogs and strain.
In the United States, the Commerce Department withdrew a planned rule on AI chip exports on March 13, 2026, after it briefly appeared on the Office of Information and Regulatory Affairs website, according to Reuters. That withdrawal followed earlier reporting that officials were debating a new export framework that could tie access to large volumes of AI chips to conditions such as investments in U.S. data centers or security guarantees.
Why it matters for the AI buildout
Subsidies are turning chip startups into policy instruments
South Korea’s move is a concrete example of industrial policy being used to create or scale “domestic champions” in AI silicon. For companies like Rebellions, public capital can improve funding visibility and shorten the timeline to qualify products for local buyers—especially when buyers want a politically durable supply plan, not just the cheapest chip.
For multinationals, the consequence is not only more competition. It is also a more complicated procurement map: customers increasingly ask where chips are designed, where they are fabricated, and whether future controls could disrupt support, updates, or replacement parts.
Capacity growth is happening where the world least expects it
The fastest expansion described in the China reporting is concentrated in mature nodes, not the most advanced process technologies used for top-end AI accelerators. That still matters for AI infrastructure because the buildout is not only GPUs. It also needs networking gear, power management, controllers, sensors, and “everything around the rack.” When demand rises quickly, constraints show up in unglamorous components and the services that connect them.
Reuters’ reporting on China’s expansion explicitly links the current cycle to supply chain strain driven by the global sprint to build AI infrastructure.
Export rules are now a recurring design constraint
In the U.S., shifting rulemaking—debate over new frameworks, followed by withdrawal of a proposed rule—signals that AI chip export controls will remain a moving compliance target, not a one-time policy shock.
This matters because export controls do not only change “who can buy.” They change how companies structure distribution, support obligations, cloud-region planning, and even product roadmaps. If rules can change within weeks, legal and compliance functions start influencing engineering and sales decisions earlier, and more often.
The pressure point: supply chain strain shows up beyond wafers
The AI buildout is stretching the semiconductor supply chain in ways that don’t always track the headline chip shortage. A separate Reuters report in March 2026 highlighted Broadcom flagging constraints, including tight foundry capacity and knock-on shortages in adjacent inputs like optical components and printed circuit boards used in data-center networking.
The consequence is practical: even if a country subsidizes a domestic chip designer, it can still run into bottlenecks in packaging, substrates, optics, and specialized materials. That shifts attention from “who has the best model” to “who can ship complete systems reliably.”
What happens next if the current mechanisms persist
If South Korea continues using public capital to scale domestic AI silicon, buyers in the region may see more local alternatives in targeted deployments, especially where government-backed procurement preferences apply. That does not automatically displace global leaders, but it can reshape the margins and the negotiation leverage in public-sector and regulated industries.
If China’s mature-node capacity expands as projected, global supply dynamics for non-leading-edge chips could loosen in some categories while still tightening in others tied to AI infrastructure, depending on where bottlenecks sit (materials, packaging throughput, optics, or qualification timelines). The more the world builds AI data centers at once, the more the constraint shifts to the next limiting step.
And if U.S. AI chip export controls remain in flux, multinational chipmakers and cloud providers should expect compliance work to behave like an ongoing operating cost: more audits, more documentation, more region-by-region product planning, and more deals shaped by what regulators will permit rather than what customers request.