Invest1 publisherNot yet confirmed elsewhere3 min readPublished
Korea's fabless lobby wants orders, not grants: NPU quotas and defence set-asides
The Korea Fabless Industry Association's chairman says R&D money cannot buy a reference customer, and asks the state to be the first buyer instead. One unnamed official backs concentrating the cash.
The Investor · Invest desk

What happened
- The head of Korea's fabless industry association said the state must buy early production volumes in bulk, not just fund research, to grow chip design firms.
- He also wants policy money concentrated at tens to hundreds of billions of won per company instead of spread thin.
- A special semiconductor account launches next year, and the association already expects fabless support to rank low if its budget is short.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- exposure A quota moves early-silicon risk onto the buyer. A defence programme or state data centre obliged to take a first-generation domestic NPU owns the downtime, not the vendor.
- constraint Concentration and broad design-centre subsidies draw on the same account, so funding a few scale-ups means small members lose the cheap verification and prototyping they currently rely on.
- decision The lever being requested sits with purchasing agencies writing specifications, not with grant committees, which changes which part of government has to say yes.
- precedent Justifying domestic preference by pointing at American and Chinese practice hands the same argument to every market a Korean designer later wants to export into.
A grant pays for a tape-out. It does not pay for a reference customer, and that is the gap Kim Kyung-ho, chairman of the Korea Fabless Industry Association, is pointing at when he says you cannot sell into global markets without a record of a product actually being used [3]. His remedy is a purchase order rather than a subsidy line: the public sector buying early production volumes in bulk, in the way he says the United States and China do [1][2].
The ask is specific enough to be costed. A set proportion of neural processing units in public data centres to be domestic, and priority for Korean-designed chips in weapons systems and public mobility [4]. What the interview does not contain is a number: no percentage, no delivery schedule, no ceiling. Quotas are enforced in exactly those terms, so the proposal is currently a direction rather than an instrument.
The money argument is sharper. The unnamed government official quoted alongside Kim frames it as 100 billion won into one promising firm instead of 1 billion won each into 100 places [13], which holds the outlay flat and multiplies per-firm funding a hundredfold [11]. Set that against the stated goal of producing several fabless firms with revenue above 1 trillion won [5], and the largest concentrated cheque on the table is worth about a tenth of one firm's target revenue [12]. Enough to finance silicon. Not enough to create the demand that turns silicon into a trillion-won book, which is why the association is asking for orders as well.
Two of Kim's other requests pull against each other on the same budget. Cheaper access to design assets through upgraded design support centres, plus subsidies for verification and prototyping, is broad help for small firms [7]. Public funds pushing strategic mergers and consortium bids for national projects is consolidation [8]. Both are meant to come out of a special semiconductor account launching next year that he already expects to be too small, with materials, parts, equipment and fabless support competing to avoid the bottom of the priority list [14].
Worth noting who is speaking. The case for guaranteed public demand comes from the trade body whose members would be paid, and the only official endorsement in the material is anonymous and addresses concentration of subsidies, not procurement quotas [13]. The one existing programme Kim praises, the trade ministry's Manufacturing AI Transformation scheme linking anchor firms with robot makers and chip designers [9], works by matchmaking. A quota works by obligation. The distance between those two is who carries the loss when an early domestic NPU underperforms in a state data centre: the vendor, or the agency that was required to buy it.
What to watch
- The first allocation inside the special semiconductor account, and whether fabless and design support get a named line or sit behind materials and equipment.
- Any defence or public mobility tender that actually carries a Korean-designed chip set-aside, with a stated percentage.
- Whether a public fund underwrites a merger or consortium bid among fabless firms, which would show the concentration argument being acted on.