特朗普政府拟限制中国新型数据中心光模块进口的后续研究文章再度扩散

Alright, wow.

What a day in optics land.

Reuters reported that the Trump administration is drafting a proposal that would restrict imports of new Chinese optical transceiver models used inside U.S. data centers.

And optical stocks absolutely exploded.

$AAOI finished up ~20%. $FN and $POET both gained around ~16%. $COHR gained ~12.5%, AEHR up ~20% (although they also have more news) $ALMU and $LITE were both up around 9%.

Meanwhile, $AXTI fell roughly 5%.

And SOXX itself closed up almost 7%.

Obviously, I am stoked.

A lot of these are names I own or follow extremely closely. Any meaningful shift away from Chinese optical transceivers would create a huge amount of durable demand for Western suppliers.

We’re talking more market share, stronger ASPs, more / longer LTA’s, customer-funded capacity, more gov support.

This is what I believe to be the straightforward read and I agree with it.

But you guys should know by now that I like to peel back the layers and look beyond the surface and try to identify alpha/angles that general market commentary won’t.

Once you start digging into this more, the story becomes much larger and convoluted than Chinese transceivers getting restricted and U.S. optics stocks going to the moon.

And if we have been paying attention, this isn’t a new idea.

The FCC publicly discussed this exact product category almost 10 months ago.

The Pentagon recently designated Innolight as a Chinese military-linked company.

Commerce has already proposed funding domestic InP manufacturing, an InP-free alternative, CPO, and advanced interconnect materials and supply-chain traceability.

NVDA invested $2b in COHR and LITE.

Texas is helping fund AAOI expansion.

POET had already started moving manu out of China.

And many more crumbs.

Then we get this proposal and the sequences we are seeing are very delibarte.

The U.S. is increasingly treating optical connectivity like strategic infra.

So the way I see this proposal is multi fold:

-

It can be a real national-security policy

-

It can push hyperscalers toward truster suppliers

-

It can justify more gov and customer funding

-

It can give Western companies even more visibility to expand

-

It can give Trump a major negotiating tool while China still controls several upstream inputs the U.S. desperately needs

I heavily lean into the last one. I think Trump is a negotiator and he is setting himself up for his meeting with Xi.

Let’s dig in.

***Financial Disclosure:** I own positions in several companies discussed here. This article reflects my personal research and opinions for informational and educational purposes only and should never be treated as individualized investment advice or a recommendation to buy or sell any security. These companies carry meaningful execution, valuation, geopolitical, supply-chain and policy risks, and I may change my views or positions at any time.*

## So what really happened?

First off, I woke up at 5:00AM PST to a major green glow from my phone. I don’t know the last time I saw such an aggressive overnight gain. No complaints on my end!

*Checks X*

“Trump bans Chinese optical transceivers”.

Incredible.

But I don’t want to get caught up celebrating.

We’ve got work to do here.
**

Reuters reported that the Federal Communications Commission is developing a proposal that would bar imports of new Chinese optical transceiver models used inside American data centers.

The reported security concerns involve data theft, malware and the possibility of Chinese-controlled devices being used to disrupt data-center operations. The FCC can still change or shelve the proposal, although officials reportedly hope to publish it during 2026.

So we still need the actual rule, the implementation date, an understanding of how existing approved products will be treated, and an understanding of how the gov defines a Chinese supplier.

Is it based on where the finished transceiver is assembled? Or where teh parent comapny is headquarted? Or who controls the supplier? Where each critical componenet was made? And so on and so on. You get the picture. It’s not clean and we need clarity.

All of these could create completely different outcomes.

I won’t unpack them all but let’s just say a rule focused primarily on final assembly could leave a giant loophole. A chinese transceiver supplier could move production into Thailand, Malysia, Vietnam or wherever and argue that the finished module came from somewhere else.

Also of note is that Reuters reported that the FCC may initially restrict imports of all new transceiver models, then exempt many suppliers viewed as being outside Chinese control.

That is a MUCH broader starting point than the headlines are suggesting.

It could potentially mean that COHR, LITE, AAOI etc. would still have to demonstrate why their specific products deserve approval. Their ownership, BOM, firmware, assembly location, and testing location could all become relevant.

Also, what do they consider a “new model”?

Does a new speed gen like 1.6T count as a new model?

Does every new form factor count?

What about firmware revision or updated component lists?

If the rule captures future 1.6T and 3.2T, the long term impact could be enormous. If Chinese suppliers can continue selling existing 800G and already approved 1.6T, the near term rev transfer would be slower.

## The Market Gave Us A Clue

If you looked at Optics stocks, you would believe the story. Massive green candles (almost) everywhere.

Then you look at SOXX (semi ETF) and it’s also up almost 7%.

How can both of those things be true?

If the market believed the U.S. was about to remove a huge amount of Chinese transceiver supply, I would expect the broader AI trade to react very differently than it did.

Think about what these modules actually do and how critical they are.

Think about all that CapEX on memory, GPUs, power, construction etc. In order for these clusters to operate at scale, they NEED a massive amount of optical links.

And who supplies a MASSIVE amount of those links?

Bingo. China.

BofA estimates the top three Chinese suppliers (Innolight, Eoptolink, and Accelink) control more than 40% of the broader market.

They also estimated that the three of them generated more than $9b of datacom module rev during 2025, with 80-90% ultimately destined for U.S. customers. (Take this directionally).

So a sudden cutoff here would create a serious problem. U.S. data center / AI progress would falter.

It’s one thing if Western suppliers could close that gap.

But even WITH Chinese suppliers contributing meanignfully, Western suppliers STILL can’t get anywhere close to meeting demand.

So I think the market treated this as a future Western market-share transfer rather than a near term supply shock.

Phased implementation, grandfathering, conditional exemptions etc.

Aside from negotation tactic, these would be my base case.

## This Has Been Building

This felt sudden.

But the gov has been working through this issue for months.

In 10/25 the FCC directly asked whether it should prohibit authorization of equipment containing optical transceivers produced by companies owned by, controlled by or subject to the direction of a foreign adversary.

It was also asking how long the industry would need to replace affected components. Industry groups argued that manufacturers could require at least two years to find replacements, source them, test them and integrate them into finished products without creating new shortages.

Check this out: [FCC ](/api/substack-inbox-db/file/209881225/4bfed9f1d26debf78c193c9d.pdf)

Washington understood that the supply chain could never switch overnight.

It knew replacement components and modules would have to be sourced and qualified.

And it was already asking how much time the industry would need.

Then the other breadcrumbs started piling up.

March 2026: NVIDIA invested $2 billion into each of Coherent and Lumentum. Both agreements included multibillion-dollar purchase commitments and future capacity rights.

May 2026: Coherent CEO Jim Anderson traveled with a U.S. business delegation accompanying Trump to China partly to raise delays involving InP export licenses. Reuters also reported that InP came up during senior U.S.-China trade negotiations.

June 8: The Pentagon added Innolight to its list of Chinese military-linked companies.

June 16: Commerce signed an LOI offering Coherent up to $50 million to expand domestic InP device capacity in Sherman, Texas.

July 29: Commerce proposed another $874 million across seven compute-supply-chain projects, including Aeluma, GlobalFoundries, Thintronics and OBSIDIA.

August 4: Reuters reported the proposed restriction on new Chinese optical transceivers.

## This Is Basically Huawei 2.0, Except Earlier

Raymond James made a comparison that I think explains the strategy really well.

The U.S. allowed Huawei and ZTE equipment to become deeply embedded in communications networks before responding aggressively.

Once the government decided the equipment created a security risk, removing it became extremely expensive. Congress eventually had to support a giant “rip and replace” effort because carriers had already built so much of their infrastructure around the equipment.

The current admin appears to be trying to avoid repeating that mistake inside AI data centers.

Rather than waiting until Chinese optical components are embedded across the next decade of infrastructure, the FCC can create an approval gate on new models today.

Existing products can potentially keep operating for a period of time whereas future products face much more scrutiny.

And the entire procurement chain receives a clear warning before the next several generations ramp.

This helps explain why the proposal appears focused on new models.

The immediate goal may be less about ripping current Chinese 800G modules out of data centers and more about controlling who supplies the 1.6T, 3.2T and eventually CPO-era connectivity.

This also seems to fit Trump’s broader AI export strategy where the White House has already created a program aimed at promoting complete American AI technology packages to allies and partners. Those packages include chips, servers, accelerators, storage, cloud services and networking, along with information about where those products are manufactured.

So this has both a defensive and offensive angle.

Defensively, the U.S. wants fewer Chinese-controlled components inside its own AI infrastructure.

Offensively, it wants the American AI stack, including the connectivity layer, to become the standard adopted by allies around the world.

## Could This Apply To Every Foreign-Produced Transceiver?

Raymond James thinks the initial restriction could apply broadly to foreign-produced optical transceivers, rather than only modules sold by Chinese companies.

At first that sounded a little aggressive to me.

But as I worked through it I now believe the idea has a real basis.

The FCC’s recent router framework placed all foreign-produced consumer routers behind a new approval gate. Foreign-made models can receive conditional approval after national-security agencies review the company, product and supply chain.

That means a U.S.-headquartered company can still face scrutiny if its product is designed, assembled or tested overseas.

So under a similar optical framework:

-

A Coherent module assembled in Asia could require approval.

-

An AAOI transceiver produced in Taiwan could require approval.

-

A Fabrinet-built product from Thailand could require approval.

-

A POET engine or module assembled in Malaysia could require approval.

But there is a very important difference between an approval gate and a permanent exclusion.

Under the router process, applicants submit detailed information on ownership, beneficial shareholders, foreign-government influence, joint ventures, bills of materials, component origins, software, firmware, manufacturing locations, testing locations and supply-chain concentration.

They also provide a plan for establishing or expanding trusted U.S. manufacturing. Conditional approvals can then exempt specific companies or products for limited periods.

That is where I land on the Raymond James interpretation.

A broad initial approval requirement for foreign-produced new models is completely plausible.

A permanent ban on every transceiver assembled outside the United States would be far too disruptive and highly highly unlikely.

The optical-specific FCC proceeding from October also focused on components produced by foreign-adversary-controlled entities. That points toward China control as the deeper concern, even if the government begins with a broader production-location screen.

So my base case would be:
**

Start with a broad approval gate. Require full supply-chain transparency. Then provide exemptions or conditional approvals to Western and allied suppliers with auditable products and credible U.S. manufacturing plans.

That still changes the value of domestic manufacturing.

AAOI’s Texas lines become much more useful.

Sherman becomes more valuable to Coherent.

Greensboro becomes more valuable to Lumentum.

And allied manufacturing in Thailand or Malaysia remains useful, though potentially subject to more paperwork, disclosures and onshoring commitments.

We’ll see. But think this through.

## The Government Is Funding The Replacement Capacity

The Coherent announcement inJune was one of the clearest clues that this policy was coming.

Commerce signed an LOI offering Coherent up to $50 million in direct CHIPS funding to expand its 6-inch InP manufacturing facility in Sherman, Texas.

The project would add fab equipment and cleanroom capacity for InP photonic devices. Coherent says the expansion will double production space and quadruple wafer-processing capacity.

Now, I want to be very clear about what Sherman does because this gets jumbled. I get frustrated when I see people on X say that AAOI and COHR form their own substrates because they are vertically integrated. They dont.

Coherent still needs to secure the blank InP substrates upstream.

Sherman takes those substrates and processes them into lasers, detectors and other photonic devices.

So the funding expands Coherent’s ability to turn available wafers into optical chips. It does nothing to create the original blank InP crystal supply by itself.

That upstream constraint still exists.

But look at how Commerce described the investment.

It specifically connected InP photonics to moving data between processors, memory and systems inside advanced AI data centers. It talked about solving data-movement bottlenecks, improving energy efficiency and strengthening domestic supply of critical optical technology.

The Sherman project had already entered the CHIPS process during the Biden administration, so domestic InP device manufacturing clearly has bipartisan strategic support.

Trump is now tying that capacity much more aggressively to the AI race and Chinese supply-chain risk.

Then Commerce funded several other parts of the optical stack as well.

On July 29, it announced up to $874 million across seven research and development projects.

GlobalFoundries could receive up to $300 million to accelerate domesticCPO development by two to three years.

Thintronics could receive up to $50 million for lower-loss materials used in next-generation interconnects and packaging.

OBSIDIA could receive up to $34 million for component identification, provenance and secure supply chains.

And Aeluma signed an LOI for up to $30 million to develop large-diameter, InP-free substrate technology for lasers and photodetectors used in AI optical interconnects.

Check this out:

That is almost the entire optical problem being funded from different directions.

So this definitely looks like an attempt to build a full domestic AI interconnect ecosystem.

And I don’t think they are nearly finished.

## AAOI Was Already Being Built Out Too

AAOI received a $20.9 million Texas Semiconductor Innovation Fund grant in April.

The grant supports an additional 210,000-square-foot facility beside AAOI’s Sugar Land headquarters. Texas expects the project to involve more than $279 million of capital investment and create around 500 jobs.

AAOI is also adding almost 400,000 square feet across two Pearland properties for production of high-speed optical products.

This happened before the proposed restriction became public.

Texas was already helping fund domestic transceiver capacity, and AAOI was already committing to a giant U.S. ramp.

Now hyperscalers have another reason to prioritize that future output.

## Why This Is Bullish Even If The Proposal Gets Watered Down

I don’t think that Hyperscalers can wait for the final rule before adjusting their supply chains.

A new optical transceiver requires reliability testing, thermal validation, firmware work, interoperability testing and high-volume manufacturing qualification.

That can take several quarters.

The FCC proceeding itself showed that replacing affected components can require two years or more.

So every future design awarded to a Chinese-controlled supplier now carries another layer of risk.

Maybe the supplier gets restricted, the module loses authorization, the finished system becomes harder to approve because the module sits inside it, the hyperscaler has to redo the entire qualification process a year later. Whatever it is.

That changes purchasing behavior before any formal restriction takes effect.

Hyperscalers now have another reason to qualify Western and allied suppliers, allocate more volume to them, reserve capacity earlier and help finance their expansions.

That can improve long-term visibility for Coherent, Lumentum, AAOI and other qualified suppliers even if the final rule receives a long transition period or broad grandfathering.

## Customer-Funded Capacity Is Already Everywhere

NVIDIA’s investments in Lumentum and Coherent are the most aggressive examples, but the broader model has already spread throughout the optical stack.

TSEM signed contracts supporting $1.3 billion of silicon-photonics revenue for 2027 and received $290 million of customer prepayments for capacity reservations.

META signed an agreement with Corning worth up to $6 billion and became the anchor customer for additional U.S. fiber and cable capacity in North Carolina. Amazon later signed its own multiyear, multibillion-dollar Corning agreement.

Fabrinet and Amazon also entered into manufacturing and development arrangements, with Amazon receiving warrants for up to 381,922 Fabrinet shares.

POET’s Lumilens agreement has a version of this structure too where Lumilens placed an initial $50 million purchase order under a framework that could scale above $500 million over five years. In exchange, Lumilens received warrants that vest as cumulative payments increase.

So we are already seeing customer prepayments, capacity reservations, long-term purchase agreements, anchor-customer relationships, equity investments, warrants and dedicated manufacturing expansions.

The China proposal should accelerate the trend no doubt about it.

## Hyperscalers Will Probably Push Back

The policy may be great for optical suppliers.

That does very little to make it great for hyperscalers.

BofA points out that these customers already rely heavily on Chinese transceiver supply while dealing with rising costs across memory and other parts of the AI stack.

Rosenblatt expects hyperscalers to lobby against a harsh version of the rule because the industry could quickly move from worrying about eventual oversupply back into severe undersupply.

I think that is absolutely right.

Amazon, Microsoft, Google and Meta need more optics today, rather than fewer.

They may support long-term diversification while strongly opposing a short implementation period.

That lobbying could result in:

-

A longer phase-in

-

Broad treatment of existing products

-

Conditional approvals for critical programs

-

More time for Western suppliers to qualify capacity

-

Faster government support for domestic expansion

This also helps explain the SOXX reaction.

The market appears to expect a managed transition rather than a policy that suddenly removes a large portion of available supply.

And the government has its own reason to listen.

Trump wants less Chinese exposure, but he also wants the U.S. AI buildout moving as quickly as possible.

Those goals can conflict with each other.

There is also the CapEx risk.

Yes I love when hyperscalers increase their expense on AI infra. Because I am heavily exposed to the beneficiaries. But if we keep squeezing them on memory ASPs and now optics, at a certain point we are putting too much pressure on hyperscaler returns and earnings at the margin.

Great near term for optics to have higher margins and everything, but long term potential bearish hyper CapEx ROI. I’ll touch on this again later.

## But does this make any sense?

The U.S. wants to reduce its dependence on Chinese optical transceivers.

But it also desperately needs Chinese optical supply today.

If Trump pushes Chinese supply out faster than Coherent, Lumentum, AAOI and others can replace it, the U.S. risks slowing its own AI buildout.

And Chinese hyperscalers can continue receiving supply from their domestic vendors.

The U.S. could end up paying more for fewer optical links while China directs more production toward its own buildout.

That would be a terrible outcome.

So this to me feels much more like a capacity clock.

Signal the direction now, push hyperscalers to move future design wins, give Western suppliers enough visibility to spend aggressively, fund the domestic fabs and then phase restrictions in as replacement capacity becomes available.

## The Chinese Oversupply Bear Case Just Got Weaker

One of the major long-term bear cases for $LITE, $COHR and $AAOI has been that Chinese laser and transceiver capacity eventually ramps too far.

I wrote all about this a month ago:

TLDR on the idea;

The shortage ends.

China floods the market with lower-cost supply.

Pricing collapses.

And all of the Western capacity being added today arrives into a much worse market.

Chinese transceivers are the main vehicle through which future Chinese-made InP lasers would reach U.S. hyperscalers.

If those finished modules lose access to the U.S., a large portion of future Chinese laser overcapacity becomes much harder to push into the American market.

That could keep Western laser and transceiver supply tighter, support ASPs for longer and extend the current favorable pricing environment.

I think that argument is directionally right, though it depends on the final component rules.

A module-focused restriction would still leave open the possibility that a Chinese laser chip gets sold into an approved Western module.

But that route involves more qualification, supply-chain disclosure and customer scrutiny. And if the FCC eventually reaches deeper into component provenance, the pathway becomes even harder.

So the proposal may do more than transfer module share.

It may reduce the ability of China to solve the global shortage with fully integrated, low-cost supply.

That is a major change to the long-term bear case.

Sorry bears!

## China’s Leverage

Yes, I know.

I have talked about substrates about a million times.

But you cannot understand this policy without understanding the upstream problem.

China controls a huge amount of the supply chain that Western optical companies still need.

China began restricting InP exports in February 2025. It also produces around 70% of global indium, the metal used to create InP.

The controls have already pushed companies to build buffer inventory and secure supply years ahead.

Reuters reported that Coherent CEO Jim Anderson traveled with the U.S. business delegation accompanying Trump to China partly to raise delays involving InP export licenses.

InP also came up during senior U.S.-China trade negotiations.

That is huge!

The substrate shortage was already being discussed at the highest levels of U.S.-China negotiations.

This was far from an obscure issue sitting inside optical earnings calls that Washington had somehow missed.

The administration knew and Coherent was directly involved.

And the issue was already on the negotiating table before the transceiver proposal became public.

## Trump Needs Leverage Too

China has something the U.S. wants.

Access to upstream materials and substrates.

The U.S. has something China wants.

Access to the largest and most valuable hyperscaler market in the world.

China can push for continued access to American data-center customers.

Trump can push for more predictable treatment of indium, InP substrates and other critical inputs.

I am simplifying this, obviously.

And there is zero public proof of an explicit substrates-for-transceiver-access agreement.

But the bargaining setup is legit.

Raymond James directly tied the timing of this proposal to the planned September 24 Trump-Xi meeting. It also argued that using the FCC Covered List gives Trump a more calibrated tool than immediately escalating through a harsher Commerce action.

That makes sense to me and was inline with my expectations as well.

There is also direct precedent for technology restrictions moving alongside broader negotiations.

Reuters reported in February that the Trump admin had already prepared restrictions on Chinese equipment used inside U.S. data centers, then shelved those measures during a wider trade détente ahead of Trump-Xi discussions.

So we already know the administration is willing to advance data-center restrictions, pause them during negotiations and bring them back later.

But that don’t mean this is a bluff.

I think Trump genuinely wants to reduce Chinese exposure across the AI stack.

But the timing, implementation period, exemptions and enforcement severity can all become negotiating tools.

Both things can be true.

## Company Read-Throughs

###

For Coherent, the relevance is direct and powerful.

The proposed restriction could direct more future transceiver programs toward the company while increasing the strategic value of its U.S. InP manufacturing capacity.

The government was already moving in this direction before the Reuters headline. Remember, Commerce signed the LOI for up to $50 million of CHIPS funding for Sherman, while NVIDIA invested $2 billion and secured multibillion-dollar purchase commitments and future capacity rights.

Coherent’s future capacity is becoming something both the government and major customers feel they need to secure.

If hyperscalers begin assigning greater regulatory risk to Chinese-controlled suppliers, Coherent should receive more qualification activity, earlier capacity reservations and better long-term visibility.

They can benefit even if the transition happens gradually. A future 1.6T or later-generation program awarded today can create revenue for years while older Chinese modules remain deployed.

But of course the biggest complication remains upstream supply.

More demand for Coherent’s lasers and transceivers increases how much blank InP substrate the company needs to secure. The Trump proposal strengthens Coherent’s downstream position while making the substrate negotiations with China even more important.

And again, no Coherent doesn’t make InP substrates.

### $LITE

They will also benefit of course.

NVIDIA had already invested $2 billion and signed a multibillion-dollar purchase commitment tied partly to future capacity. They are also preparing their Greensboro facility for U.S. production.

One think to remember here is that Lumentum currently sells lasers to Chinese module companies, including Innolight and Eoptolink. If those customers lose future U.S. programs, they may purchase fewer Western components or retaliate by shifting toward Chinese laser suppliers.

I think this is more of a customer and product-mix reallocation issue than a real demand problem of course.

We know demand way outstrips supply and Lumentum is the laser GOAT.

Any modules displaced from Innolight or Eoptolink still need lasers.

Lumentum can redirect capacity toward its own Cloud Light products, other Western and allied module makers, SiPh customers, CPO programs and external light sources.

That reallocation may take time because every product and customer requires qualification. But I have a hard time seeing any high-performance laser capacity sitting idle in this environment.

### $AAOI

One of the primary beneficiaries as they are aggressively expanding U.S. production of complete 800G and 1.6T transceivers.

Texas already awarded them nearly $21 million to support their Sugar Land expansion, and they are also adding a large amount of manufacturing space across the Houston area.

A restriction on Chinese transceivers could push more future module programs toward that capacity.

It could also make customers more willing to provide prepayments, minimum purchase commitments or other support for the ramp.

I would pay especially close attention to any large agreement tied directly to the Texas buildout.

A major hyperscaler becoming an anchor customer would reduce the risk attached to all of this new floor space and equipment.

The foreign-produced question and clarity is relevant here too.

AAOI has meaningful U.S. manufacturing, along with facilities in Taiwan and China. If the FCC starts with a broad foreign-production gate, the location and bill of materials for each specific product will become important.

But I do believe that AAOI’s onshoring efforts are going to be a very strong selling point and we should hear all about it in their call tomorrow.

### $FN

Fabrinet could benefit as Western optical suppliers need more manufacturing, packaging and testing capacity outside mainland China.

Its Thailand footprint gives customers an established location for scaling complex optical products while reducing direct China exposure.

If COHR, LITE or other Western-controlled suppliers win more programs and outsource part of the volume ramp, FN could become one of the companies helping absorb that production.

Rosenblatt specifically highlighted Fabrinet’s relationships with NVIDIA and Amazon as a reason it could gain additional hyperscaler-related manufacturing work.

The exact FCC structure will be especially important here too though.

A rule focused mainly on Chinese ownership and control would strengthen the value of Fabrinet’s allied manufacturing platform.

A broader structure that initially captures all foreign-produced models could create another approval process for products assembled in Thailand.

So FN’s opportunity comes from helping Western suppliers physically scale displaced production rather than replacing Chinese transceiver brands itself.

The customer-funding angle may also become more relevant here. Amazon already has a manufacturing and development arrangement alongside equity warrants tied to the relationship.

### $POET

POET could benefit by helping Western and allied module companies scale optical engines outside China.

They established their manufacturing footprint through Globetronics and NationGate in Malaysia as part of an effort to diversify and secure its supply chain.

That decision looks increasingly well timed under this proposal.

A Western or allied module supplier first needs to win the program, and then that company needs to select POET’s optical engine.

That makes POET’s partnerships and current customer programs especially important.

The Lumilens framework begins with a $50 million order, although the larger and broader opportunity still depends on qualification, development and manufacturing milestones.

POET also has an interesting connection to the InP constraint.

It still relies on InP lasers, but its 1.6T Teralight engine uses four dual-channel laser chips instead of the standard eight separate chips. That reduces the number of individual laser dies that need to be sourced, attached and aligned.

The opportunity is significant for POET. The next step is proving that the customer programs convert into meaningful volume revenue and that Malaysian manufacturing can deliver the promised scale.

### $AXTI

AXT has a very mixed read-through. Especially depending on how you are viewing this whole thing.

More Western laser and transceiver production creates even more demand for InP substrates, which should help AXT’s underlying business.

They already hves intense demand, long-term agreements and aggressive capacity expansion underway.

The proposed restriction could also give Trump greater leverage when pushing China for more predictable substrate export permits.

Any meaningful improvement in direct U.S. permits would materially increase AXT’s ability to serve the market.

But AXT manufactures in China and sits directly inside the supply-chain conflict.

China can delay permits.

The U.S. could eventually expand its concern farther upstream.

And customers will continue trying to secure more non-Chinese substrate supply wherever they can find it.

So AXTI could become a major winner from a negotiated agreement or suffer from a deeper separation between the two supply chains.

The negative stock reaction tells us which side the market focused on after the headline.

I think the next major signal will come from China rather than the FCC. If permits begin moving more freely, the bargaining-leverage thesis becomes much stronger.

### $ALMU

Aeluma becomes more strategically relevant because the government is actively looking for domestic alternatives to the constrained InP supply chain.

Only days before the transceiver proposal became public, Commerce signed an LOI offering Aeluma up to $30 million in proposed CHIPS funding for its large-wafer, InP-free laser and photodetector platform.

That timing shows the administration is thinking beyond replacing Chinese finished modules.

It is also funding technologies that could reduce upstream dependence over time.

Aeluma’s relevance comes from the possibility of producing photonic devices on much larger alternative substrates, potentially creating more manufacturing throughput and reducing reliance on the small native InP wafer ecosystem.

But theyremains years away from replacing meaningful commercial supply.

The technology still needs to prove qualification, yields, reliability, throughput and economics at scale.

And the proposed funding contemplates equity issued to Commerce, meaning shareholders need to account for dilution.

So ALMU fits into the longer-term policy direction far more than the immediate transceiver-share transfer.

The government is effectively helping fund an alternative path while established suppliers expand the conventional InP chain.

## Other Read-Throughs

There are several other names that can benefit, but the path is less direct.

$VIAV, AEHR and other relevant testing cos because more Western transceiver and CPO production creates more demand for testing and validation.

$CIEN, $NOK and $ADTN could eventually receive more opportunity if Chinese suppliers lose share. (I do have another angle here but I might be overthinking it. Basically CIEN and NOK are supply constrained because LITE/COHR can’t get them enough pump lasers and such. So if the economics make more sense for LITE and COHR to prioritize their capacity lines for transceivers, this could limit the supply for CIEN and NOK. I really haven’t dug into this too much yet so this might make no sense and I am gettign ahead of myself. But I just want to share where my brain is going and stress testing).

$GLW benefits from the broader move toward trusted U.S. optical infrastructure, although fiber and cable sit outside the specific transceiver restriction. The Meta, Amazon and NVIDIA agreements already show how aggressively hyperscalers are securing domestic connectivity supply.

$MRVL, $CRDO and $ALAB can receive some scarcity premium if the market places a higher value on U.S. connectivity technology. But higher transceiver prices and reduced module availability could also slow the broader AI deployment cycle, creating a much more mixed read-through.

The same applies to $MTSI and $SMTC.

More Western module production can increase demand for their drivers, TIAs, photodiodes and other components. But some of those products already go into Chinese transceivers. BofA estimates roughly the mid-20% range of MACOM’s revenue comes from Chinese customers, so Chinese retaliation or customer substitution could offset part of the benefit.

And then there is $IQE. Its allied manufacturing footprint and external InP epitaxy capacity could become more strategically valuable if customers seek geographic diversity or additional overflow capacity outside China. But IQE still needs access to the same constrained blank substrates, and new customer programs still require lengthy qualification. The benefit depends on whether laser companies actually turn toward external epi partners as they try to fill the capacity gap.

I think that covers the general read through. Let me know if you have any other company specific reads of course.

## The Less Bullish Second-Order Effect

All of this is obviously great for optical suppliers.

Better ASPs, stronger pricing, longer agreements and more customer funding improve the economics for LITE, COHR, AAOI and others.

But those benefits increase the cost of the data-center buildout.

Optics alone should remain manageable.

The larger concern is that optics is one of many cost categories moving higher at the same time.

Memory is already squeezing hypers enough. Then youve got power, cooling, construction, grid access etc.

At some point, all of those increases begin pressuring hyperscaler returns and earnings at the margin.

I still expect the hyperscalers to spend aggressively of course. Maybe even more so now.

But every constrained supplier capturing better economics reduces the return on each incremental dollar of capital spending.

That can also accelerate customer funding, deeper vertical integration and architectures designed to reduce power and cost.

Strong pricing is great for pluggable suppliers today.

It also gives customers more motivation to redesign the system around the expensive pieces over time.

## China Can Hit Back

China has already warned that it will respond to actions that cause material harm to its interests.

The most obvious response would involve tighter controls on InP substrates, indium or other upstream materials.

China could also prioritize its own optical companies.

It could direct more module capacity toward Chinese hyperscalers.

It could subsidize aggressive pricing throughout the rest of the world.

And it could make life much harder for Western manufacturers that still rely on Chinese upstream inputs.

So here is the central paradox of this whole thing.

The more transceiver share moves toward Western suppliers, the more lasers and substrates those suppliers need.

And a meaningful portion of incremental upstream supply still involves China.

The U.S. needs to move downstream production away from Chinese control while preserving enough upstream access to keep the AI buildout running until alternatives arrive.

That is a very difficult needle to thread.

Let’s see if they can do it.

## What I Think Happens

My current base case is a gradual transition.

The FCC creates a new approval gate for future optical transceiver models.

Existing approved products receive some form of continued treatment.

Foreign-produced products face detailed disclosure requirements, with conditional approvals for Western and allied suppliers that have auditable supply chains and credible U.S. manufacturing plans.

Hyperscalers shift more future programs toward Coherent, Lumentum, AAOI and other trusted suppliers.

Federal and state funding continues moving toward substrates, lasers, CPO, advanced packaging, raw materials and component traceability.

Meanwhile, Trump can use access to the U.S. transceiver market during broader negotiations over InP, indium and other strategic inputs.

Could the FCC move faster than I expect?

Of course.

Could China retaliate much more aggressively?

Yes.

Could the proposal eventually get shelved during another trade agreement?

Absolutely.

We already have precedent for that.

But even in that scenario, the direction has changed.

Every future optical design awarded to a Chinese-controlled supplier now carries more regulatory risk.

That should improve the long-term position of trusted Western and allied suppliers regardless of where this exact proposal ends up.

## *Financial Disclosure / Disclaimer*

*I currently own positions in several companies discussed in this article. *

*I may buy, sell, add to, trim, hedge or exit any position at any time, including without providing an update in real time. My views may also change as new information becomes available.*

*This article reflects my personal research, opinions and interpretation of publicly available information and private sell-side research shared with me. It is provided solely for informational and educational purposes and should be treated as personal commentary rather than individualized financial, investment, legal, tax or accounting advice.*

*The companies discussed carry substantial risks involving execution, valuation, dilution, customer concentration, geopolitics, export permits, material availability, capacity ramps, manufacturing yields, qualification timelines, pricing, technology transitions and broader AI capital spending.*

*POET and Aeluma remain early-stage businesses relative to the established optical suppliers discussed. Their announced orders, development programs, letters of intent, customer frameworks and government funding proposals may face delays, fail to convert into meaningful commercial revenue or require additional capital and dilution.*

*Several government actions discussed remain proposals, letters of intent or conditional policy frameworks. Funding amounts, equity terms, regulatory language, implementation timelines, exemptions and enforcement mechanisms can change, face delays or fail to reach completion.*

*Sell-side market-share estimates, customer-exposure assumptions and industry forecasts can vary materially by source and methodology. They should be treated as estimates rather than audited company disclosures.*

*Forward-looking estimates and scenarios are inherently uncertain. Actual outcomes may differ significantly.*

*Everyone should conduct their own research, review relevant filings and primary-source disclosures, and make decisions based on their own financial situation, portfolio construction, time horizon and risk tolerance.*

*I have received no compensation from any company discussed for publishing this research. Any errors are my own.*
Does this change everything?