Optical Stocks Have a China Problem That Most Investors are Missing

Beijing controls a chokepoint that Washington's own crackdown can't address

China controls the permitting for critical optical resources.

Lumentum Holdings recently agreed to pay AXT $87 million in deposits to reserve six years of indium-phosphide wafer supply.

Read that contract closely and it explains the whole artificial-intelligence optics trade. The first $43.5 million is due within 30 business days; the timing and terms of the second half are to be settled in 2028.

Lumentum $(LITE)$ is building its own laser plant in Greensboro, N.C., and its chief executive has spent weeks warning that this material is heading into a shortage worse than the memory crunch. But Greensboro builds lasers; it does not grow the crystals those wafers are cut from.

Lumentum has to buy those. AXT $(AXTI)$ grows them in China. The latter company is based in Fremont, Calif., and is listed on the Nasdaq COMP - but its plants sit in three Chinese locations, and every outbound shipment requires a permit from Beijing's Ministry of Commerce. AXT's finance chief told analysts last week that he cannot predict when those permits arrive, or whether they arrive at all for a given order.

The onshoring story and the actual supply chain point in completely different directions, and most investors who piled into this theme in 2026 have not looked closely enough to notice. Sorting these optical suppliers by who controls the scarce input, rather than by who sells finished hardware, splits the sector into three tiers - and puts one of this year's best-performing stocks in the most exposed one.

Everyone found the bottleneck at once

Inside big AI clusters, copper (HG00) wiring has hit a wall. Push data down a copper link fast enough and the signal survives only a meter or two, and the power needed to push it climbs. Light has neither limit, so the industry is switching to optics, and optics need lasers. On March 2, Nvidia (NVDA) invested $2 billion into Coherent (COHR) and Lumentum each, with purchase commitments attached. Between them, those two companies make most of the world's high-speed data-center lasers. The market bought the whole group.

Those shares have been on a roller-coaster ride. Lumentum's stock traded above $1,000 in early June, fell below $650 in late July and closed near $826 on Aug. 5. Coherent's stock peaked above $440, dropped toward $240 and then was back near $328. Applied Optoelectronics (AAOI) shares swung from $201 to $88, then back to around $129.

Part of the recent rebound came on Aug. 4, when Reuters reported that the Trump administration is drafting a rule, through the Federal Communications Commission, to bar U.S. data centers from importing new Chinese-made optical transceivers. The measure is not final and could still be shelved - but if enacted, it would squeeze out Chinese incumbents such as Zhongji Innolight (HK:3308) (CN:300308), which holds about 27% of the global market, and push U.S. cloud giants toward Coherent and Lumentum.

Don't miss: Lumentum's stock just got a lot more interesting - and Barclays has turned bullish

All three stocks are up sharply for the year; even after the pullback, Lumentum was trading near 150 times trailing earnings, while Coherent was trading at more than 155 times. This is a crowded high-beta trade, and it is not cheap.

Lumentum CEO Michael Hurlston put it plainly at a Paris conference in July, according to remarks cited by Tom's Hardware. Telecom buyers used to order lasers in the hundreds; Nvidia and the cloud giants want hundreds of millions. Lumentum runs five indium-phosphide plants and is still shipping more than 30% below what customers ask for.

"Between the two of us," Hurlston said of his company and Coherent, "I don't think we can service the demand."

So the shortage is real and public. The question that matters most for investors is narrower: Where in the chain does the shortage sit, and who gets paid for it?

The metal is tightening. The substrate is choking.

Let's start with the raw material. The U.S. Geological Survey's 2026 summary put world refined indium output at about 1,100 tons in 2025, with China at 760 tons, or 70%. That is the figure everyone quotes. The one underneath it matters more: World refining capacity is 1,700 tons.

Outside China, plants can refine about 585 tons a year and are producing about 331. South Korea alone runs 180 tons against 310 tons of capacity. Indium is a byproduct of zinc smelting, so how much gets recovered depends on zinc economics, not on laser demand. Higher prices can bring some of that idle capacity back faster than anyone can build a mine, though not overnight.

Demand does not support panic either. Most indium goes into coatings for flat-panel displays, a mature market. Lasers use very little by weight: A four-inch indium-phosphide wafer holds roughly 19 grams. Even allowing generously for waste, the whole wafer industry plausibly uses tens of tons a year, not hundreds.

Opinion: Big Tech is forcing consumers to pay for its AI boom. Voters are pushing back.

Prices have risen sharply, but the move remains modest when considering what has happened with wafers from suppliers already tested and approved for high-volume laser production. The USGS put the 2025 U.S. average at about $390 a kilogram, up 11%. By February, Reuters reported indium in Rotterdam at $500 to $600 - the highest price since early 2015, driven partly by speculation on a Chinese futures exchange.

Now let's look at the compound. The clearest read on qualified-wafer pricing is what it did to a wafer maker's accounts: AXT's gross margin rose from 8% of revenue to nearly 45% in a year. Refined metal has repriced; approved wafer supply has tightened enough to turn a quarterly loss into an $11.1 million profit within a year. That is the trade: not the metal, but the wafer.

The scarce parts are a permit and a crystal

What China controls most directly is the paperwork. On Feb. 4, 2025, Beijing issued Announcement No. 10, putting tungsten, tellurium, bismuth, molybdenum and indium-related items under export licensing. Indium-phosphide wafers are covered. Refined metal is not - though exports still need ordinary licenses - and Reuters reported in June that buyers now face much heavier customs scrutiny on metal, which the industry sees as a sign that tighter controls on the metal itself may follow.

Here is the part few investors have priced. Everyone is watching Nov. 10, when the six export-control measures Beijing suspended under the Busan trade truce are due to snap back. But the controls covering indium phosphide were never suspended in the first place. They are already in force, have no expiration date and will not change when the truce ends.

The November 2025 suspension covered the six measures Beijing issued on Oct. 9, plus the part of a December 2024 rule that had blocked gallium, germanium and antimony sales to the U.S. Announcement No. 10 was not among them; it has run without interruption since February 2025 and has no end date. This Chinese control sits directly on AI-networking hardware, and because it was never traded away, it is not scheduled to lapse or return with anything else in November.

The effects are already visible. Since the controls began, the price of a six-inch InP wafer has jumped 250% to $5,000, Reuters reported. AXT's Chinese unit needed permits in June and August of 2025 before it could ship wafers again. Coherent's chief executive, Jim Anderson, flew to Beijing with President Trump in late May to raise licensing delays, according to the same report.

Underneath the licensing sits a manufacturing limit. Indium-phosphide crystals are brittle and grown in small batches, and the industry is only now moving from three-inch to six-inch wafers, a step silicon took in the 1980s. Merchant supply is dominated by Sumitomo Electric (JP:5802) $(SMTOY)$, AXT and JX Advanced Metals (JP:5016) $(JXAMY)$. Coherent grows some of its own and Chinese producers are expanding, but customers take years to qualify a new wafer supplier.

The framework

Treating this as one "buy optics" trade is the mistake. However, if we sort by who controls the scarce input, three tiers appear:

Tier 1: The assemblers. Least control, highest beta.

Applied Optoelectronics is the clearest case: a transceiver maker with no wafer supply of its own, no epitaxy, no profit over the past 12 months, heavy customer concentration and one of the highest betas in the group. It owns the growth and none of the chokepoints. The FCC import ban described above cuts both ways for this tier: Shutting out Chinese module makers could hand assemblers like AAOI a protected U.S. market, yet it would do nothing to secure their own upstream wafer supply, which still traces back to China. A tailwind on the revenue line, and the same chokepoint underneath it.

Tier 2: The device makers. Partial control, unevenly held.

Coherent posted fiscal third-quarter revenue of $1.81 billion, with data centers and communications now 75% of the total and the backlog stretching into 2028. Its six-inch line makes more than four times as many devices as the three-inch line at less than half the cost, and internal capacity doubled a quarter ahead of plan. Lumentum runs five device plants and is converting the 240,000-square-foot former Qorvo $(QRVO)$ site in Greensboro that's set to start production in 2028. Coherent is the one with real wafer supply of its own. Lumentum is bringing device work home while still buying its crystals, which is why it committed $43.5 million to AXT upfront.

Tier 3: The upstream. Real pricing power - and the concentration risk that comes with it.

AXT reported second-quarter revenue of $47.6 million on July 30, up 165% - swinging to a profit while seeing gross margins dramatically expand. Cash rose to $412 million from $120 million at the start of the year. The stock jumped sharply and has been one of 2026's best performers.

MW Optical stocks have a China problem that most investors are missing

By Jurica Dujmovic

Beijing controls a chokepoint that Washington's own crackdown can't address

China controls the permitting for critical optical resources.

Lumentum Holdings recently agreed to pay AXT $87 million in deposits to reserve six years of indium-phosphide wafer supply.

Read that contract closely and it explains the whole artificial-intelligence optics trade. The first $43.5 million is due within 30 business days; the timing and terms of the second half are to be settled in 2028.

Lumentum (LITE) is building its own laser plant in Greensboro, N.C., and its chief executive has spent weeks warning that this material is heading into a shortage worse than the memory crunch. But Greensboro builds lasers; it does not grow the crystals those wafers are cut from.

Lumentum has to buy those. AXT (AXTI) grows them in China. The latter company is based in Fremont, Calif., and is listed on the Nasdaq COMP - but its plants sit in three Chinese locations, and every outbound shipment requires a permit from Beijing's Ministry of Commerce. AXT's finance chief told analysts last week that he cannot predict when those permits arrive, or whether they arrive at all for a given order.

The onshoring story and the actual supply chain point in completely different directions, and most investors who piled into this theme in 2026 have not looked closely enough to notice. Sorting these optical suppliers by who controls the scarce input, rather than by who sells finished hardware, splits the sector into three tiers - and puts one of this year's best-performing stocks in the most exposed one.

Everyone found the bottleneck at once

Inside big AI clusters, copper (HG00) wiring has hit a wall. Push data down a copper link fast enough and the signal survives only a meter or two, and the power needed to push it climbs. Light has neither limit, so the industry is switching to optics, and optics need lasers. On March 2, Nvidia (NVDA) invested $2 billion into Coherent (COHR) and Lumentum each, with purchase commitments attached. Between them, those two companies make most of the world's high-speed data-center lasers. The market bought the whole group.

Those shares have been on a roller-coaster ride. Lumentum's stock traded above $1,000 in early June, fell below $650 in late July and closed near $826 on Aug. 5. Coherent's stock peaked above $440, dropped toward $240 and then was back near $328. Applied Optoelectronics (AAOI) shares swung from $201 to $88, then back to around $129.

Part of the recent rebound came on Aug. 4, when Reuters reported that the Trump administration is drafting a rule, through the Federal Communications Commission, to bar U.S. data centers from importing new Chinese-made optical transceivers. The measure is not final and could still be shelved - but if enacted, it would squeeze out Chinese incumbents such as Zhongji Innolight (HK:3308) (CN:300308), which holds about 27% of the global market, and push U.S. cloud giants toward Coherent and Lumentum.

Don't miss: Lumentum's stock just got a lot more interesting - and Barclays has turned bullish

All three stocks are up sharply for the year; even after the pullback, Lumentum was trading near 150 times trailing earnings, while Coherent was trading at more than 155 times. This is a crowded high-beta trade, and it is not cheap.

Lumentum CEO Michael Hurlston put it plainly at a Paris conference in July, according to remarks cited by Tom's Hardware. Telecom buyers used to order lasers in the hundreds; Nvidia and the cloud giants want hundreds of millions. Lumentum runs five indium-phosphide plants and is still shipping more than 30% below what customers ask for.

"Between the two of us," Hurlston said of his company and Coherent, "I don't think we can service the demand."

So the shortage is real and public. The question that matters most for investors is narrower: Where in the chain does the shortage sit, and who gets paid for it?

The metal is tightening. The substrate is choking.

Let's start with the raw material. The U.S. Geological Survey's 2026 summary put world refined indium output at about 1,100 tons in 2025, with China at 760 tons, or 70%. That is the figure everyone quotes. The one underneath it matters more: World refining capacity is 1,700 tons.

Outside China, plants can refine about 585 tons a year and are producing about 331. South Korea alone runs 180 tons against 310 tons of capacity. Indium is a byproduct of zinc smelting, so how much gets recovered depends on zinc economics, not on laser demand. Higher prices can bring some of that idle capacity back faster than anyone can build a mine, though not overnight.

Demand does not support panic either. Most indium goes into coatings for flat-panel displays, a mature market. Lasers use very little by weight: A four-inch indium-phosphide wafer holds roughly 19 grams. Even allowing generously for waste, the whole wafer industry plausibly uses tens of tons a year, not hundreds.

Opinion: Big Tech is forcing consumers to pay for its AI boom. Voters are pushing back.

Prices have risen sharply, but the move remains modest when considering what has happened with wafers from suppliers already tested and approved for high-volume laser production. The USGS put the 2025 U.S. average at about $390 a kilogram, up 11%. By February, Reuters reported indium in Rotterdam at $500 to $600 - the highest price since early 2015, driven partly by speculation on a Chinese futures exchange.

Now let's look at the compound. The clearest read on qualified-wafer pricing is what it did to a wafer maker's accounts: AXT's gross margin rose from 8% of revenue to nearly 45% in a year. Refined metal has repriced; approved wafer supply has tightened enough to turn a quarterly loss into an $11.1 million profit within a year. That is the trade: not the metal, but the wafer.

The scarce parts are a permit and a crystal

What China controls most directly is the paperwork. On Feb. 4, 2025, Beijing issued Announcement No. 10, putting tungsten, tellurium, bismuth, molybdenum and indium-related items under export licensing. Indium-phosphide wafers are covered. Refined metal is not - though exports still need ordinary licenses - and Reuters reported in June that buyers now face much heavier customs scrutiny on metal, which the industry sees as a sign that tighter controls on the metal itself may follow.

Here is the part few investors have priced. Everyone is watching Nov. 10, when the six export-control measures Beijing suspended under the Busan trade truce are due to snap back. But the controls covering indium phosphide were never suspended in the first place. They are already in force, have no expiration date and will not change when the truce ends.

The November 2025 suspension covered the six measures Beijing issued on Oct. 9, plus the part of a December 2024 rule that had blocked gallium, germanium and antimony sales to the U.S. Announcement No. 10 was not among them; it has run without interruption since February 2025 and has no end date. This Chinese control sits directly on AI-networking hardware, and because it was never traded away, it is not scheduled to lapse or return with anything else in November.

The effects are already visible. Since the controls began, the price of a six-inch InP wafer has jumped 250% to $5,000, Reuters reported. AXT's Chinese unit needed permits in June and August of 2025 before it could ship wafers again. Coherent's chief executive, Jim Anderson, flew to Beijing with President Trump in late May to raise licensing delays, according to the same report.

Underneath the licensing sits a manufacturing limit. Indium-phosphide crystals are brittle and grown in small batches, and the industry is only now moving from three-inch to six-inch wafers, a step silicon took in the 1980s. Merchant supply is dominated by Sumitomo Electric (JP:5802) (SMTOY), AXT and JX Advanced Metals (JP:5016) (JXAMY). Coherent grows some of its own and Chinese producers are expanding, but customers take years to qualify a new wafer supplier.

The framework

Treating this as one "buy optics" trade is the mistake. However, if we sort by who controls the scarce input, three tiers appear:

Tier 1: The assemblers. Least control, highest beta.

Applied Optoelectronics is the clearest case: a transceiver maker with no wafer supply of its own, no epitaxy, no profit over the past 12 months, heavy customer concentration and one of the highest betas in the group. It owns the growth and none of the chokepoints. The FCC import ban described above cuts both ways for this tier: Shutting out Chinese module makers could hand assemblers like AAOI a protected U.S. market, yet it would do nothing to secure their own upstream wafer supply, which still traces back to China. A tailwind on the revenue line, and the same chokepoint underneath it.

Tier 2: The device makers. Partial control, unevenly held.

Coherent posted fiscal third-quarter revenue of $1.81 billion, with data centers and communications now 75% of the total and the backlog stretching into 2028. Its six-inch line makes more than four times as many devices as the three-inch line at less than half the cost, and internal capacity doubled a quarter ahead of plan. Lumentum runs five device plants and is converting the 240,000-square-foot former Qorvo (QRVO) site in Greensboro that's set to start production in 2028. Coherent is the one with real wafer supply of its own. Lumentum is bringing device work home while still buying its crystals, which is why it committed $43.5 million to AXT upfront.

Tier 3: The upstream. Real pricing power - and the concentration risk that comes with it.

AXT reported second-quarter revenue of $47.6 million on July 30, up 165% - swinging to a profit while seeing gross margins dramatically expand. Cash rose to $412 million from $120 million at the start of the year. The stock jumped sharply and has been one of 2026's best performers.

(MORE TO FOLLOW) Dow Jones Newswires

August 07, 2026 07:00 ET (11:00 GMT)

MW Optical stocks have a China problem that most -2-

Anyone buying AXT to reduce China exposure is doing the opposite. Every wafer is grown in China, every shipment needs a Beijing permit, and the company holds stakes in 10 Chinese suppliers. The risk section of its own earnings release opens with export permits. AXT is a good operator with genuine scarcity value and the biggest political dependency in the chain. The cleanest way to own the shortage is also the most direct way to own the licensing risk.

The cleaner non-Chinese options are duller. IQE (UK:IQE) $(IQEPY)$, a Welsh wafer maker with plants in Britain, the U.S. and Taiwan, raised its 2026 growth guidance to above 30%, from 20%, on July 21. It is still in the midst of a turnaround: 2025 revenue fell to GBP97.3 million from GBP118 million, and the company lost GBP37 million before tax. On the metal, Korea Zinc (KR:010130) is among the largest refiners; it bought Nyrstar's (BE:NYR) U.S. assets on April 1 and is developing a $7.4 billion Tennessee smelter with U.S. government backing - with phased operations, and indium among 13 planned products, targeted for 2029. That is how long a real fix takes.

What next week tests

Lumentum reports earnings on Aug. 11, with Coherent following on Aug. 12. Both are guiding to sharp revenue growth: Lumentum has guided to $960 million to $1.01 billion, up from $500 million a year earlier, and Coherent has guided to $1.91 billion to $2.05 billion, up from $1.53 billion. Strong growth is already the consensus, so the beats won't move their shares much. What will is how much of each company's wafer supply runs through China, which is what the disclosures below reveal.

The useful disclosures sit lower down, in terms of how much of each company's wafer supply is grown in-house and how much comes from China. Investors should look to see whether anyone else pays cash up front to lock in wafers, which would be a measure of how little faith buyers have in the open market. And it's worth listening for whatever management says about export permits - the one thing neither company controls.

How this trade breaks

The bear case is not that the shortage is fake - it is what happens afterwards. Market-intelligence firm LightCounting's April forecast had demand for optical transceivers - the modules that use these lasers to move data over fiber - running 30% above supply, with the squeeze easing by the end of 2026. Its July report offered revised estimates on both counts: Growth for 2026 went up to 73%, and the laser shortage now looks likely to run into the middle of 2027.

That is good for near-term revenue, but it makes the backlog harder to trust.

LightCounting's reason is that almost every transceiver component is now scarce and customers have started double ordering, which inflates backlogs. The firm says the industry is starting to feel like 2000 and 2001. Backlog built on double orders is not the same as backlog built on demand, and the difference only shows once supply catches up. Coherent's book into 2028 deserves that caveat.

It also supplies the timing: LightCounting's data show transceiver sales turning down about six months before cloud capital spending does - and its model has that spending slowing from roughly 70% growth this year to 30% next year. If so, the turn in optics would arrive before the news that would explain it. Lumentum CEO Hurlston, meanwhile, runs a company that benefits the longer the shortage lasts.

The risk runs the other way too. If Beijing extends Announcement No. 10 to refined metal, everything downstream reprices at once. The Defense Logistics Agency has already asked for bids to stockpile up to 403 tons of indium over three years - more than a third of a year's global output. A purchase of that scale shows the U.S. government considers a supply disruption credible enough to prepare for now.

The market was right to identify optics as a bottleneck. Its mistake was treating every company in the chain as the same trade. Scarcity rewards businesses that control qualified wafer supply, but AXT's advantage comes bundled with the sector's largest geopolitical dependency.

From here, returns will depend less on who reports the biggest backlog than on who can keep shipping when supply catches up - or when Beijing tightens the gate.

-Jurica Dujmovic