The FCC is drafting a ban on the part inside most AI data center network gear. That's a new long-lead item nobody's pricing yet.

Reuters reported on August 4 that the Trump administration, through the FCC, is drafting a rule to bar imports of new-model Chinese-made optical transceivers — the modules that move data through data center fiber networks at the speed of light. It's still a draft. But two Chinese suppliers, Innolight and Eoptolink, together control roughly 60% of the global market for the 800G transceivers that AI and hyperscale data centers run on. If the rule lands, a cheap, high-volume network part most GC procurement schedules don't even track becomes the newest entry on a growing list of long-lead items that can stall a data center build.

What is the FCC actually trying to ban?

The draft rule would block imports of new-model Chinese optical transceivers, the small modules that sit in network switches and convert electrical signals into light pulses for fiber transmission. The administration's stated concern is that Chinese-made transceivers could be used to steal data, install malware, or disrupt service inside US data centers that train and run AI models. Officials told Reuters they hope to publish the rule later this year, when it would take effect — but sources also said the FCC could still modify or shelve it before that happens. China has said it will respond if the restriction moves forward.

How exposed is the supply chain, really?

More than most people building data centers probably realize. Innolight and Eoptolink together supply roughly 60% of the global market for 800G optical transceivers, the bandwidth tier used in current AI data center network fabric. Innolight alone holds about 27% of the broader data center transceiver market. Analysts covering the sector have pointed to a 12-to-24-month ramp for Western and Taiwanese manufacturers to absorb that volume, partly because some non-Chinese alternatives still depend on Chinese-controlled indium phosphide, a material used to make the laser components inside transceivers.

Why does this matter to a GC, not just a network engineer?

Because data center construction has already spent the past two years learning this lesson the hard way with bigger equipment. Large power transformers now run lead times of two to four years. Medium-voltage switchgear runs 45 to 80 weeks and climbing. Those components now dictate hyperscale project timelines more directly than design, financing, or site selection — the projects staying on schedule are the ones where procurement got locked in 18 to 24 months ahead, treated as the first phase of construction rather than something that happens in parallel with it.

Optical transceivers have never been on that list. They're a comparatively cheap, high-volume commodity item, installed by low-voltage and structured-cabling subcontractors during network fit-out — one of the last trades to touch the building before commissioning, when schedule float is already gone. A supply restriction hitting the two suppliers that hold 60% of that market turns a part nobody budgets lead time for into one that could behave like switchgear: order it late, and it's the reason the facility doesn't hit its commissioning date.

What should a GC or low-voltage sub do with this now?

The rule isn't final, and it may never take the form Reuters described. That's exactly why this belongs in a precon conversation now rather than after a Federal Register notice forces it. Three moves, in order:

• Ask who's specifying the network gear and with what components. On a hyperscale or AI data center bid, find out whether the electrical or MEP design already names transceiver manufacturers, and whether the spec anticipates Chinese-made modules.
• Flag it as a procurement risk in the same conversation as switchgear. Owners and design teams have gotten used to discussing multi-year transformer lead times. Transceivers haven't made that list yet — raise it before it becomes one more thing found out the hard way.
• Track the rule, not the headline. A draft FCC rule with no publication date isn't a reason to rebid a project. It is a reason to ask your network equipment supplier, this month, what their contingency sourcing looks like if the current path narrows.

The honest read: nothing here is locked in. The FCC could publish this rule, narrow it, or drop it entirely. But the underlying exposure — 60% of a critical network component concentrated in two suppliers now facing a geopolitical restriction — doesn't go away just because the rule stays in draft. Data center GCs have already reorganized how they buy switchgear around exactly this kind of risk. Optical transceivers are the next candidate.