2026.10.02 | Substack | Optical | Optics Update: China Restrictions Pt.2

What an exciting day!

Optics ripped after Morgan Stanley published a note based on conversations with a senior Washington telecom, FCC and export-control legal team. The big takeaway is that 3.2T is increasingly looking like the generation where restrictions around Chinese optical transceivers could start to matter.

But there was one detail that really changed how I think about this compared to the news we got back in August.

Morgan Stanley heard discussion around a potential framework where a Chinese-built transceiver could still be allowed into the U.S. if roughly 65% of the value inside the module came from U.S.-based companies. Their example was basically a Marvell DSP plus laser content from Lumentum or Coherent potentially getting a Chinese module toward that threshold. They actually came away thinking laser pricing could be the stronger read-through versus a huge transceiver share shift.

I want to make something very clear before getting into this. This is not an FCC rule. The 65% number and the 3.2T framework are coming from Morgan Stanley’s Washington channel work. We still don’t know how U.S. content would actually be defined, what gets counted, or whether anything close to this ends up in a final rule.

But I think this is a really important update from the original Reuters story.

This actually sounds possible

When Reuters first reported in August that Washington was looking at restrictions on new Chinese optical transceiver models, I got very excited about it. I wrote a whole article digging through what it could mean for the optical supply chain.

Trump’s Chinese Optical Proposal Is A Really Big Deal.

Gaetano

·

Aug 5

Read full story

It was very easy to poke holes in it though. And it honestly didn’t really seem plausible. How are we going to ban that much Chinese transceiver capacity?

Innolight and Eoptolink are gigantic and hold massive market share. Western suppliers already can’t make enough optics today. Removing a massive amount of Chinese manufacturing capacity while the U.S. is racing to build AI infrastructure could create an even worse supply problem. The Reuters report itself focused on possible restrictions on new models, rather than ripping out everything already deployed, but the practical question was still sitting there.

So after the initial excitement I think a lot of people basically looked at the story and said... okay, maybe Washington wants to do this, but how?

I think this update could be the answer. Or at least something like this.

You don’t necessarily need Innolight to stop assembling transceivers. Let them keep assembling them. But maybe that Innolight module now needs a Marvell DSP, Lumentum lasers, MACOM analog components and enough other approved content to get across whatever threshold Washington eventually establishes.

That is a MUCH easier transition for me to imagine than trying to recreate tens of millions of units of Chinese module manufacturing somewhere else.

And I think it completely changes where we should look for the beneficiaries. Maybe the biggest winners aren’t the companies replacing Innolight’s transceivers. Maybe the biggest winners are the companies selling the expensive stuff inside the Innolight transceiver.

That is what I have been digging through all day.

It’s also funny because the proposed ban was originally positioned around safety concerns. It was a very big stretch at the time and this just goes to show that it wasn’t ever about safety concerns. But that doesn’t matter for these companies. What we care about is the U.S. gov wants U.S. companies to win. Great.

I think this is a widespread positive for the U.S. optical ecosystem

Before getting into the individual companies, this is the bigger takeaway for me.

Look at how much capacity is being built right now. Lumentum is building Greensboro. Coherent is pouring money into their 6-in ramp. AAOI is expanding like crazy. Tower is massively expanding silicon photonics. Everyone is ordering equipment, locking up wafers and trying to prepare for an optical market that is going to be much larger several years from now.

And one of the risks hanging over all of this has been pretty obvious. What happens if everyone builds too much? What happens when all of this new capacity from the U.S., China, Japan and everywhere else shows up around 2028 and 2029?

I think this policy makes that question more complicated because we may eventually have a very real difference between global optical capacity and capacity that can actually serve the U.S. AI market.

That gives me much more confidence in some of the capex being spent today. It can give suppliers better visibility into future demand, and for some of the really constrained parts of the chain it could support pricing for longer than people currently assume.

Chinese suppliers can still build a ton of capacity. They can still serve China and other markets. But if U.S.-bound products increasingly need approved components, approved manufacturing locations or eventually deeper supply-chain provenance, then you can’t look at one giant chart of global capacity and assume every wafer, laser and transceiver is interchangeable.

That pushes back on some of the (future) oversupply concern I have had myself.

And 2029 really isn’t that far away

This is the other criticism I keep seeing on X.

Why are we talking about something that might happen in 2029?

Because you don’t build a 2029 supply chain in 2029.

Morgan Stanley actually thinks the policy work itself could begin as early as October, while its channel work points toward initial 3.2T volumes around 2028 and broader deployment around 2029. Those are two completely different clocks.

That doesn’t mean some giant 3.2T restriction starts this month. It means Washington could start working through the framework now while these products are still being designed. And that makes complete sense to me.

If Innolight needs to swap lasers, those lasers have to be designed in and qualified. If hyperscalers want more U.S. manufacturing, somebody has to build it. If IQE needs more epi capacity, reactors have to be converted and wafers have to qualify. If everyone suddenly wants more Japanese InP substrates instead of Chinese ones, those substrates need to exist.

The work I did around this suggests something like 6 to 18 months for component substitution and qualification, while meaningful new fab or packaging capacity can take 18 to 36 months or longer.

Coherent and AAOI and others have already said customers started asking about manufacturing options after the original August policy story. That is exactly the kind of thing I would expect.

So people looking at 2029 and saying this is early are looking at the shipment date instead of the qualification date.

If Washington wants the 2029 optical supply chain to look different, companies have to start changing it now.

Alright. With that, let’s dig into all the companies involved and who I believe is set to benefit the most. You might be surprised on my conclusions.

I think there are a few pretty clear layers here, but the order changes depending on what the eventual rule actually looks like.

1 - Lasers

This is probably the cleanest read-through for me.

The laser can be a huge chunk of the value inside an EML-based 3.2T module.

$LITE is probably the cleanest merchant beneficiary to me right now. They have EMLs, CW lasers, demonstrated 400G/lane differential EMLs and Greensboro starts coming online right into the 3.2T window. And I really like the economics of this setup for them. Lumentum may not need to win the transceiver. Innolight can still build the thing. Lumentum can sell them the expensive laser that helps make the module compliant. And this is the highest margin business for LITE anyway. I don’t really want them making transceivers.

$COHR obviously belongs right there as well. They have 400G/lane differential EMLs, 400mW CW lasers and their own 3.2T work. But there is an important wrinkle here because Coherent has basically said its own module business can eat a lot of the InP capacity it is building. So I think we need to separate who can makes good lasers from who actually has merchant lasers available to sell into somebody else’s module. That could make Lumentum cleaner than Coherent specifically for this setup.

Then you have $AVGO, which might be the wildcard because Broadcom can bring way more than a laser. DSP, driver, EML, photodiode. They already publicly work with Eoptolink on Taurus. If compliance becomes a BOM exercise, being able to supply a giant chunk of that BOM from one vendor gets pretty interesting.

Then I would have $SMTC/HieFo, $AAOI, SIVE and some of the earlier programs like AEVA in the next group. There are real products and real programs here, they just aren’t all at the same level of qualification, scale or merchant availability yet.

If AAOI can ramp meaningful capacity and become a merchant laser supplier, or if we do still try to shift more finished transceivers to domestic manufacturing then they could be a massive beneficiary.

2 - DSPs and PICs

If lasers are one giant chunk of the BOM, DSPs are the other obvious one.

$MRVL immediately jumps out. Marvell is already deeply embedded in optical DSPs and just demonstrated 2nm 400G/lane technology at ECOC.

The funny part is that Chinese module companies already use U.S. DSPs in some designs. So the policy upside isn’t necessarily that Marvell suddenly wins every one of these sockets. Maybe they already have a lot of them.

The benefit could be protecting those sockets from future Chinese substitution, giving customers another reason to lock in supply and potentially improving visibility around pricing. Still very valuable, just a different type of win.

Broadcom obviously shows up again here, then $MXL and $CRDO become interesting and very relevant if their 400G/lane roadmaps line up with the timing which I believe they will.

Then for NOK they are developing ICE-D around an integrated InP photonic circuit (PIC) that combines light generation, modulation and detection, and it is already talking about 448G and 3.2T. That makes Nokia a legitimate PIC / optical-engine supplier into this generation, with meaningful veritcal U.S. manufacturing exposure on top of it.

3 - Analogers

Let’s say you build the whole module and end up at 60% eligible content. You need another five points.

On an $800 BOM, replacing around $40 of foreign content with eligible content gets you there.

So what is easier? Redesign the whole optical architecture, or qualify a U.S. TIA, driver and photodiode?

That is where $MTSI gets really interesting. MACOM has already demonstrated 448G drivers, TIAs and photodiodes. $SMTC has a similar angle through the analog portfolio and then HieFo adds light-source exposure too.

These might not be the biggest individual parts inside the module, but if they are the parts that get you from 60% to 65%, suddenly they can become pretty valuable.

4 - Foundries and merchant epi

$TSEM already sits directly inside the optical supply chain through the silicon photonics chips going into these modules. Tower has a major relationship with Innolight that already covers current high-speed products and is being extended toward 3.2T, it has shipped more than 5 million coherent PICs with Marvell, and it has demonstrated silicon photonics capable of supporting 400G/lane with Coherent. So under a moderate rule where Innolight keeps assembling modules and changes other parts of the BOM around it, Tower could keep a lot of that volume. A hard Innolight ban is actually messier because then you have to figure out who replaces those modules and whether the same Tower silicon follows with them.

$GFS has a similar silicon photonics opportunity, but I think the domestic manufacturing angle makes it especially interesting here. GlobalFoundries can manufacture the actual SiPh chips, and its relationship with Marvell now includes additional U.S. capacity in Vermont for the analog components going into pluggables, NPO and CPO. Commerce has also signed an LOI for up to $300 million specifically to accelerate GFS’s domestic CPO work by roughly two to three years. A lot of GFS’s optical growth was already coming without this policy, so I don’t want to double count it. But if Washington increasingly favors optical silicon actually manufactured in the U.S., there is already a platform, customer relationships and government support sitting there.

Then you have the merchant epi foundries sitting one step farther upstream.

IQE is interesting because its historical problem has been underutilization. They have had a lot of infrastructure sitting there without enough demand to really fill it. That is starting to change. Tower signed a multiyear InP epiwafer agreement, MACOM has a strategic relationship with them, and the whole idea here is that IQE can potentially help companies ramp qualified InP devices without every supplier needing to build all of that upstream capacity internally.

$IQE . Right Place. Right Time. But Can They Execute?

Gaetano

·

Jul 22

Read full story

LandMark gives us a pretty great example of what this model can look like when demand is really strong. They signed a four-year agreement with a U.S. customer covering CW laser supply from 2027 through 2030, while at the same time locking up their own InP substrate supply. Customers are planning years ahead, then the epi supplier is planning years ahead so they can actually fulfill those commitments.

IQE Major Read Through. Also AAOI, SMTC, SIVE, MTSI

Gaetano

·

Aug 28

Read full story

I wouldn’t put IQE or LandMark at the top of the policy beneficiary list because the device makers can always decide to do more epi internally. But if Western laser and PIC production has to ramp faster than internal epi capacity can handle, merchant epi becomes a pretty useful release valve.

5 - Manufacturing

This depends heavily on what the final rule actually says.

If Washington outright restricts Chinese module manufacturers, then $FN, $SANM, $AAOI and other alternative manufacturers get much more interesting because actual module production has to move.

If China can keep assembling modules as long as enough eligible content sits inside them, that share shift gets a lot smaller.

That is why I think SANM could actually be more interesting than FN under some versions of this. Sanmina has real U.S. optical packaging and assembly capability. If U.S. manufacturing value itself counts, that could become useful.

Fabrinet has an enormous optical manufacturing footprint, but most of it is in Thailand. That is great if trusted allied manufacturing qualifies. It is much less important if the answer is simply stuffing more U.S. components into a Chinese-built module.

6 - Substrates

This might eventually become the messiest part of the whole chain.

Let’s say Lumentum makes a laser in the U.S., but the blank InP substrate underneath it came from AXT’s operation in China.

Does that still count as a U.S. laser?

Maybe under a shallow company-based definition. Maybe not if Washington eventually starts tracing deeper provenance. I doubt theyd take it that far. Im just using this to illustrate the grey areas.

That is why $AXTI is such a weird one. More Western laser production is obviously good for InP demand, but AXT makes the wafers in China.

At the same time Japan is expanding aggressively through Sumitomo and JX, while China now has this new Xianrui/Guangzhi substrate operation ramping 6-inch material and claiming shipments to major optical customers.

So AXT could end up sitting in a strange middle ground where Western customers increasingly want allied substrates while Chinese customers increasingly have domestic alternatives, even as total InP demand continues exploding.

That’s going to be a really interesting one to watch.

If AXT can form these long term relationships and qualifications with U.S. based companies ahead of AXT’s Chinese rivals, I could see that being very strategic.

When I first heard the August 4th update I thought Trump was going to use the proposed ban as a way to get more access to substrates. But I didn’t hear any updates on that since their meeting.

What I am keeping lower on the list

Companies like VIAV, KEYS, AEHR, FORM and TER obviously benefit from the normal 3.2T test cycle. The FCC-specific benefit is narrower.

You need actual redesigns, duplicated production lines or manufacturing moving to new geographies before I start getting really excited about incremental test revenue.

Same thing with $CIEN. I really like Ciena’s broader AI networking setup, but if Chinese coherent modules can stay in the market by changing their BOM, the FCC-specific benefit to Ciena may be pretty small.

That is another reason Nokia sits higher for me in this particular discussion. Nokia has the PIC/device angle too.

Closing Thoughts

I think today’s Morgan Stanley note made this whole story much more believable.

The original Reuters story gave us the goal. This gives us a possible way to actually get there without blowing up the optical supply chain in the process.

Keep the Chinese manufacturing scale, push more of the valuable content toward approved suppliers, then give the industry a couple of years to redesign and qualify the next generation.

If something resembling that framework becomes real, I think it is broadly positive for the U.S. optical ecosystem. It gives me more confidence in the capex being spent today, it can improve visibility and pricing power for some of the constrained parts of the chain, and it makes me much less comfortable looking at one giant chart of future global optical capacity and immediately concluding that everyone is going to overbuild.

We still need the actual rule. The 65% number could change completely. How Washington defines “U.S. content” could move the beneficiary list around a lot.

But if the policy process really starts moving now, then I don’t think we are early talking about this.

Financial Disclosure / Disclaimer

I am long some of the companies discussed in this report and may buy or sell shares at any time without notice. Any positions referenced reflect my holdings at the time of publication and may change.

This report is for research, educational, and informational purposes only. Nothing in this report should be considered financial advice, investment advice, a recommendation, an offer to buy or sell securities, or a solicitation of any kind. I am not acting as your financial adviser, investment adviser, broker, fiduciary, or tax adviser.

The analysis reflects my personal opinions and interpretation of publicly available information, including company filings, earnings calls, presentations, press releases, industry research and third-party sources. Some conclusions involve assumptions around future regulation, customer qualification, capacity, pricing, manufacturing location and supply-chain changes. Actual outcomes may differ materially.