At a Glance
What happens if an export tax is imposed on foundation models — the very top of the AI service supply chain? A proposal to bring tokens, the billing unit for APIs offered by OpenAI, Anthropic, and other US providers, under an export-control framework has entered the discussion. What makes this especially relevant for Korean investors is that even allied nations may not be exempt.
Why It Matters Now
Breaking the AI service supply chain into stages makes the structure clear: foundation models (OpenAI, Anthropic, Google DeepMind) → API layer → Korean service integrators → domestic end users. Most Korean IT service companies currently sit between the second and third stages, sourcing the foundation models needed for inference through US firms' APIs. If an export tax is layered on top of charges that already run from a few won to several dozen won per token, the most direct impact would be concentrated at this stage.
This discussion isn't coming out of nowhere. Just as semiconductor export controls expanded in scope from chips to equipment to materials, the logical progression of AI controls runs from physical hardware (GPUs, HBM) down to software weights, and now to the inference API layer. A token export tax would be the final layer in that progression. Nothing has been legislated yet, but the mere fact that the discussion has surfaced puts pressure on Korean companies to reexamine their AI sourcing strategies.
This moment is making visible the cost asymmetry between companies that own proprietary large language models and those that don't. Naver serves HyperCLOVA X from its own data centers, keeping its reliance on foreign APIs limited. By contrast, companies that wrap and resell foreign models, or that depend entirely on external APIs for backend inference, have a cost structure directly exposed to US policy variables.
FAQ
- What exactly is an AI token export tax — It's a proposal to impose a tax or export-control conditions on token usage (the billing unit for input/output text) when US AI companies provide API access to foreign firms. It is currently at the pre-legislative policy discussion stage.
- Would this apply even if Korea is an ally — This is precisely the crux of the debate. The view that strategic-asset control could take precedence over alliance status under an AI-nationalist posture is gaining traction, backed by the precedent set by semiconductor export controls.
- Does owning a proprietary LLM provide full protection — Not completely, since dependence on US-origin resources remains in training data, GPU infrastructure, and foundation-model fine-tuning pipelines. However, it can shield a company from the direct cost hit at the inference stage.
- Would higher costs be passed on to service prices — Given how competitive the AI service market is, passing costs through won't be easy. A situation where costs rise but prices can't follow leads to margin compression.
Related Stocks and Sector Impact
- Naver — With HyperCLOVA X in-house, its reliance on inference APIs is low. Under an export-tax scenario, its relative cost resilience could stand out, though the cost of foreign GPUs and training infrastructure needed to advance its models remains a variable.
- Kakao — Relies heavily on foreign model APIs as it expands its AI services. If cost increases materialize, a scenario in which its profitability-improvement momentum stalls should be considered.
- SK Telecom, KT — Both have B2C and B2B services, such as AI personal assistants and enterprise call centers, that are often integrated with foreign LLMs. Rising large-scale inference traffic costs would directly affect the profitability of their AI businesses.
- Samsung SDS and the IT services sector — Paradoxically, growing demand for AI sovereignty could boost demand for on-premise GPU servers and domestic AI cloud investment, creating potential structural upside for enterprise IT infrastructure build-out demand.
Investment Considerations
- The token export tax remains at the discussion stage. With the legislative timeline, scope of application, and any allied-nation exemptions still unconfirmed, a stock decline that overly prices in the cost impact in advance could turn out to be noise — though conversely, the actual pace could also move faster than expected.
- Companies' reliance on foreign APIs is often not disclosed explicitly in filings. The actual scale of cost exposure needs to be verified through the cost-of-revenue breakdown in quarterly earnings releases or the footnotes of business reports.
- Accelerated investment in domestic AI model development is a near-term drag on earnings but a mid-term positive for cost savings. This time lag should be evaluated separately.
- Changes in US AI companies' own API pricing policies are another factor at play simultaneously — foundation-model unit prices can rise or fall independent of any export tax.
Overall Outlook
The optimistic scenario is one in which the token export tax discussion never translates into actual legislation. If the AI cooperation framework with allied nations strengthens and Korea negotiates preferential terms as an AI supply chain partner, the discussion could settle without any cost shock. In that case, companies that invested early in advancing their proprietary LLMs would be recognized for their competitiveness without an unnecessary regulatory premium.
The risk scenario looks different. Just as AI chip export controls spread faster than expected, the timeline from bill proposal to enforcement for a token export tax could also be compressed. If the pace of in-house model advancement fails to keep up with rising API costs, margin compression could become prolonged. Three things are worth watching: whether the US Congress introduces AI export-control legislation, progress in US-Korea AI policy talks, and the quarterly trend in Naver's and Kakao's AI service cost-of-revenue ratios.
Naver: Real-Time Data Snapshot
Naver's most recent closing price was 197,400 won (0.00% vs. the prior day), and the composite signal combining foreign/institutional order flow with news momentum reads 🟢 Buy-leaning. Positive foreign-investor flows and news momentum make it worth watching.
- ▼ Trend Alignment — Short- and mid-term downtrend alignment (intraday +0.0% · 1-week -1.0% · 1-month -27.3%)
- ▼ 52-Week Position — Near the 52-week low, at the 6% mark
- ▲ News Flow — 3 positive catalysts vs. 1 negative catalyst — positive catalysts lead
Recent related news skews favorable, with 3 positive catalysts versus 1 negative catalyst.
※ Price and foreign/institutional order-flow data are provided by Korea Investment & Securities (KIS), as of the time of publication.
This article was automatically summarized and analyzed based on the original news report. View original (Yonhap News, Industry)
