FCC Rule on Optical Connectivity Could Slow AI Race

Republished by Newlife - MIM. All rights belong to the original publisher; see Source below.

FCC Rule on Optical Connectivity Could Slow AI Race Escalating tech war between the U.S. and China and the potential ban on importing optical network equipment could force hyperscalers to navigate critical shortages.

The technological trade conflict between the U.S. and China could extend beyond high-end chips. While the initial restrictions focused on advanced semiconductors and the equipment used to produce them, Washington is now eyeing the wider data center infrastructure.

The new strategy is about rack sovereignty. The concept is straightforward: Securing the GPUs is not enough if other components of the networking fabric create potential security or supply chain vulnerabilities. To deal with this, the U.S. Federal Communications Commission (FCC) issued a new rule last week, “Protecting Against National Security Threats to the Communications Supply Chain Through the Equipment Authorization Program,” affecting the small but important components that convert electricity into light for high-speed data transmission.

The FCC has established compliance dates for the new rule March 1, 2027 for online marketplaces that sell their own devices, take title, or have physical access to third-party seller inventory. June 1, 2027 for marketplaces that host third-party listings without physical access/title (must verify FCC ID against the FCC database and collect seller certifications).

The geopolitical weight here comes from an asymmetric dependency. Chinese firms such as Zhongji InnoLight dominate global revenue, but they rely on U.S. cloud giants for their most advanced orders.

The numbers back up this dominance. According to Counterpoint Research estimates, Zhongji InnoLight, which was added to the Pentagon’s list of alleged Chinese military-backed companies in June, sits at the top of the food chain, commanding roughly 27% of global revenue. They’ve locked in a prime spot supplying the high-end 800G and 1.6T modules that North American cloud giants are currently hungry for.

San Jose, Calif.-based Coherent follows in second with a 17% slice of the pie, while Eoptolink—another Chinese heavyweight with deep ties to U.S. hyperscalers—is catching up fast. Lumentum, also based in San Jose, holds about 6% of the finished module market, though its greater strength lies in supplying the laser components that make these modules work.

Meanwhile, Western companies still rely on Chinese manufacturing for high-volume packaging. In this complex environment, Coherent is emerging as a vital U.S. domestic partner as efforts to build more sovereign infrastructure gain momentum.

Closing the component loophole

The FCC’s move targets what it calls the “component part loophole.” Previously, a device made by a company that was not on the Covered List could still receive FCC authorization even if it incorporated a component produced by a Covered List entity.

Under rules adopted by the FCC in July 2026, that loophole is closed.

The new rule further clarifies it.

The FCC has redefined “logic-bearing hardware” to include hardware components that process data at high speeds, including optical transceivers.

Optical transceivers fall under the new rule’s scope because they are physical components that generate and use timing signals or pulses at a rate exceeding 9,000 pulses (cycles) per second and utilize digital techniques.

As a result, if a network device incorporates a covered logic-bearing component made by a company on the blocklist, the whole unit could be denied FCC authorization, potentially limiting its entry into the U.S. market. If optical transceivers fall within the rule’s scope, this could amount to a substantial exclusion of Chinese technology from the core of AI data centers.

Optical transceivers are sometimes considered “dumb” pipes with little security threat, because they merely convert signals and do not run full operating systems suitable for hacking. Yet this is not quite accurate given the speeds needed for modern AI (800G and 1.6T).

Modern modules are complex systems that can include DSPs and firmware. This internal software manages everything from power to data paths. If that firmware is not secure, an attacker could manipulate diagnostics or trigger thermal spikes. In a massive cluster, one faulty module can blind a switch to overheating or inject just enough latency to make a billion-dollar GPU array go idle.

Challenging the symbiotic loop

The problem is that China currently dominates this market. According to Counterpoint Research, China accounts for about 60% of global optical datacom transceiver revenue, with Zhongji Innolight leading the pack. Because U.S. export bans prevent Chinese cloud operators from using the most powerful GPUs, these Chinese manufacturers have little choice but to sell their best 800G and 1.6T modules to U.S. hyperscalers.

“Market share of Chinese companies would be significantly affected as these companies depend on overseas customers—mainly the big U.S. hyperscalers—for more than 90% of their high-value 800G and 1.6T module revenues,” Gareth Owen, associate director of Counterpoint Research, told EE Times.

To dodge the potential ban, many are moving production to Southeast Asia. But if the FCC targets ownership rather than just the factory location, those moves won’t matter. It’s a messy, symbiotic loop: Chinese makers need U.S. DSPs, while U.S. companies need Chinese assembly scale.

Packaging deficit

When the FCC news broke in August, Western suppliers such as Coherent and Lumentum saw their stock prices jump, but analysts were not entirely convinced. Western firms lack the specialized cleanrooms and massive scale to replace Chinese production anytime soon. Last year, McKinsey projected a 40% to 60% shortfall in 800G transceivers through 2027, with 1.6T transceiver shortfalls of 30-40% persisting through 2029.

In response to questions from EE Times, Counterpoint Research argued that the potential ban could have severe consequences for the AI arms race

Capacity deficit

Western competitors such as Coherent and Lumentum have advanced photonic designs, but currently lack the cleanroom capacity, automated packaging infrastructure, and yield scale to absorb Innolight and Eoptolink’s volume within a 12- to 24-month horizon.

Cluster turn-up delays

In an AI supercluster, a missing batch of optical transceivers could leave billion-dollar GPU allocations idle. Constraining transceiver imports could delay the commercial turn-up of next-generation 800G/1.6T clusters by multiple quarters.

Capital cost escalation

Restricting market supply in a hyper-growth environment could increase bill-of-materials (BOM) costs for hyperscale data centers, forcing cloud operators to pay higher premiums for non-Chinese allocations.

“This would also cause major problems for hyperscalers such as AWS, Google, Meta, and Microsoft, as a lack of optical transceivers potentially lays idle billions of dollars of the latest GPUs,” Owen told EE Times. “Hyperscalers must therefore move quickly to ensure that they have alternative supply chains in place before the ban is enacted—currently expected before the end of 2026, although this may be pushed back a little.”

Potential laser chokepoint

There’s also a raw material crisis. Lumentum CEO Michael Hurlston warned that the shortage of indium phosphide (InP), used for the lasers in modules, is more severe than the memory chip crisis.

Demand has jumped from hundreds of lasers for telecoms to hundreds of millions for AI. Since China controls 70% of refined indium, it could easily retaliate by cutting off the supply of the very lasers Western companies need to build their own alternatives.

To hedge these risks, giants such as Nvidia are stepping in. It recently invested USD 2 billion in Coherent to bolster U.S. manufacturing. “With Coherent, we are pioneering next-generation silicon photonics to enable AI infrastructure at unprecedented scale, speed, and energy efficiency,” Nvidia’s Jensen Huang said.

Coherent CEO Jim Anderson added that the deal highlights his firm’s role as “a key enabler of next-generation AI data center infrastructure.”

Coherent and Lumentum did not respond to EE Times’ requests for comment.

Rise of silicon photonics

This uncertainty is also accelerating the move toward silicon photonics—integrating optics directly onto the chip to skip traditional packaging altogether. Lumilens, which emerged from stealth on Aug. 6 with a USD 5.51 billion valuation, has begun shipping interconnect solutions to an unnamed hyperscale cloud provider.

“The constraint on AI has shifted from how many GPUs you can buy to how many you can connect,” Lumilens CEO Ankur Singla said. “Hyperscalers need far more optical capacity for the networks they run today, and a path to directly connecting thousands of GPUs together into a single cluster.”

Reconfiguring this global supply chain carries significant risks. The FCC is taking an unprecedented step in the physical layer of internet infrastructure. Still, without sufficient domestic capacity or dependable raw materials, this attempt for greater sovereignty could hinder the very AI leadership that the U.S. is trying to protect.

Source

- eetimes.com (2026-09-17) - Original article: FCC Rule on Optical Connectivity Could Slow AI Race - EE Times - https://eetimes.com/fcc-rule-on-optical-connectivity-could-slow-ai-race