FCC는 외국 의존성으로 인해 현재 AI 데이터센터 공급망 부품의 60%를 금지할 것을 제안하고 있으며, 아직 적격 국내 대체품이 없습니다.

The Federal Communications Commission is quietly drafting a rule that could reshape AI infrastructure costs across the United States. As previously reported, the FCC is considering a ban on imports of new Chinese-made optical transceivers—the compact plug-in modules that convert electrical signals into light pulses for transmission through fiber optic cables within data centers. Chinese-made transceivers could theoretically be modified to install malware or exfiltrate data from facilities central to America’s AI buildout. The supply chain mathematics, however, are anything but straightforward.

An optical transceiver is roughly the size of a pack of gum and plugs directly into switches and servers. Every piece of data moving between GPUs in a training cluster passes through these modules; without them, AI data centers cannot function. Chinese manufacturers currently supply approximately 60% of the optical transceiver modules used in AI data centers worldwide. Zhongji Innolight, a single Chinese company, holds roughly 27% of the global market share. Banning new Chinese models does not merely create a gap—it creates a chasm that Western manufacturers cannot fill overnight.

U.S.-China tech decoupling has played out across semiconductors, AI chips, and now data center components. Each round follows the same pattern: ban first, then scramble for alternatives. The critical difference this time is that the replacement path loops directly back to China.

The dynamics become circular here. Western companies such as II-VI (now Coherent), Lumentum, and Broadcom can manufacture optical transceivers. However, the high-performance photodetectors inside those modules require indium phosphide, a compound semiconductor. China controls approximately 70% of global indium production. Consequently, the proposed solution to Chinese transceiver dependence results in transceivers that still depend on Chinese indium. This is not a novel issue—it mirrors the rare earth dependency that has long plagued electric vehicle and defense supply chains—but it has gained urgent relevance as AI data center buildouts consume transceivers at unprecedented rates.

Microsoft, Amazon, Meta, and other hyperscalers are expected to spend roughly $140 billion on AI-related capital expenditure in 2026. Each new data center requires thousands of optical transceivers. A supply crunch at this scale would directly slow the pace of AI infrastructure deployment.

Wall Street is not waiting for the FCC to finalize the rule. Coherent surged 47% in a single week following the Reuters report, as investors bet that Western transceiver manufacturers will absorb demand currently served by Chinese firms. The logic appears sound on the surface: if Chinese transceivers are banned, orders will flow to Coherent, Lumentum, and Broadcom’s photonics division. But capacity does not scale like stock prices. Building transceiver fabrication lines takes 18 to 24 months. The companies seeing their valuations jump today will not ship meaningfully more units tomorrow.

Experience from the semiconductor sector’s record-setting year underscores this reality—global chip sales hit $120.6 billion in May alone, up 104% year over year. Supply chain pivots at this scale take time, capital, and often government subsidies to execute.

The proposed rule specifically targets new models of Chinese optical transceivers. Existing equipment already installed in U.S. data centers would likely remain unaffected. The FCC could publish the rule later in 2026, though sources told Bloomberg the measure could still be modified or shelved entirely.

China’s response has been predictable but pointed. Beijing has issued guidance requiring new state-funded data center projects to use domestically made AI chips. Chinese regulatory authorities have also ordered data centers less than 30% complete to remove all installed foreign chips—a mirror-image decoupling that could shrink Nvidia’s addressable market.

If the ban takes effect and Western manufacturers cannot scale production quickly enough, AI data center operators face two options: pay premium prices for scarce Western-made transceivers, or slow down their buildout timelines. Either outcome raises the cost of AI compute. Higher infrastructure costs flow downstream to cloud pricing, API rates, and eventually to the products and services built on top of that compute. Companies spending $140 billion this year on AI infrastructure are already operating on thin deployment margins; adding transceiver scarcity to the equation complicates the math further.

The FCC’s ban remains a draft, and drafts are subject to revision. Yet the supply chain arithmetic it exposes—where banning Chinese components leads to alternatives that still rely on Chinese materials—represents a structural challenge that will not resolve by year-end, regardless of the FCC’s final decision.

**FAQs**

**What is an optical transceiver and why does AI need them?**

An optical transceiver is a plug-in module that converts electrical data signals into light pulses for transmission through fiber optic cables. Every AI data center relies on thousands of these modules to move data between GPUs during model training and inference.

**How much of the optical transceiver market does China control?**

Chinese manufacturers supply about 60% of optical transceiver modules globally. Zhongji Innolight alone holds roughly 27% of global market share, making it the single largest supplier to U.S. hyperscale data center operators.

**Could the FCC ban raise AI cloud computing costs?**

Yes. If Western manufacturers cannot absorb 60% of displaced demand quickly, transceiver scarcity would increase data center build costs. Those costs flow downstream to cloud pricing and API rates charged by providers like AWS, Azure, and Google Cloud.

**What is indium and why does it matter for this ban?**

Indium is a soft metal used to manufacture indium phosphide, a semiconductor compound critical for high-speed optical transceivers. China produces roughly 60% of the world’s indium supply, meaning Western transceiver replacements still depend on Chinese raw materials.

**When will the FCC optical transceiver ban take effect?**

The FCC hopes to publish the final rule later in 2026, but the measure remains in draft form. Sources familiar with the process indicate it could be modified, delayed, or shelved. Existing Chinese transceivers already installed in U.S. data centers would likely be grandfathered in.