SHANGHAI/HONG KONG — China’s hottest AI hardware trade is getting another reality check, as investors dump some of the country’s biggest chip, optical-networking and robotics winners amid growing concern that valuations raced too far ahead of earnings.
The latest selling extends a broader correction that has already erased a large portion of the gains built during China’s spectacular artificial-intelligence rally earlier this year.
The pressure is being felt particularly hard in companies tied to the physical infrastructure behind AI:
optical transceivers;
networking equipment;
semiconductors;
robotics;
and high-end electronics manufacturing.
That matters because these were precisely the companies investors had treated as the safest “picks and shovels” of China’s AI boom.
Now the market is asking a much tougher question:
What happens when the earnings remain strong—but the valuation already assumes near-perfect growth?
China’s AI hardware trade had become one of 2026’s biggest winners
For much of the year, Chinese AI hardware stocks were among the strongest-performing equities in Asia.
Investor enthusiasm was driven by Beijing’s push for technological self-reliance, explosive demand for AI data centers and expectations that Chinese companies could take larger shares of global semiconductor, networking and robotics markets.
Some hardware indices rallied roughly 75% in only three months before the reversal began, according to Bloomberg-linked market reporting.
By the end of September, the Star Market 50 and ChiNext—both heavily exposed to growth and semiconductor companies—had fallen by roughly 30% from their mid-year levels.
That is the backdrop for the latest selloff.
This is not a sector falling from nowhere.
It is a sector correcting after one of the fastest speculative expansions in the Chinese equity market.
Zhongji Innolight became the symbol of the boom
Few companies illustrate that better than Zhongji Innolight.
The company makes high-speed optical transceivers used inside AI data centers to move information between servers, switches and accelerator clusters.
Its products are essential to the enormous computing systems being built by companies such as Google, Nvidia, Huawei and other hyperscale customers.
Its earnings growth has been extraordinary.
Recent market data showed second-quarter 2026 revenue of around 22.3 billion yuan, with earnings near 7.9 billion yuan and a profit margin above 35%.
The stock’s longer-term performance has been even more dramatic.
Since 2023, shares of Innolight and rival Eoptolink Technology had at one point risen by around 4,500%, according to reporting on China’s optical sector.
That type of rally inevitably creates a valuation problem.
Eventually investors stop asking:
Is this a great business?
And start asking:
How much greatness is already priced in?
Innolight’s valuation had become extremely demanding
At recent levels, Zhongji Innolight traded at a trailing earnings multiple above 40 times earnings, with earlier periods showing much higher forward expectations during the peak of the AI rally.
Other optical names became even more expensive.
Some Chinese AI-networking suppliers traded at valuations approaching or exceeding triple-digit earnings multiples during the strongest part of the rally.
Those prices can be justified only if earnings continue expanding at exceptional rates.
The problem is that extremely high valuations create asymmetric risk.
A company can report good results—
and the stock can still fall if investors expected extraordinary results.
Global investors are becoming less tolerant of AI perfection
That shift is not unique to China.
American AI stocks have also come under pressure.
On October 8, Nvidia fell roughly 2.9%, AMD almost 4%, Micron nearly 5%, Broadcom more than 4% and Intel about 5% amid renewed concerns about whether AI infrastructure spending can justify current valuations.
The Philadelphia-style chip complex suffered one of its steepest recent declines.
The Nasdaq 100 dropped about 1.4%, its biggest fall in seven weeks, after speculation surrounding OpenAI’s revenue growth revived questions over how quickly AI spending will translate into profits.
That weakness spread naturally into Asia.
Chinese hardware stocks are connected to many of the same spending cycles.
If investors begin questioning U.S. data-center capex, they will also question the companies selling components into those systems.
OpenAI’s revenue debate became another catalyst
A Financial Times report said OpenAI’s annualized revenue was tracking around $50 billion, below figures close to $70 billion that had circulated previously.
That triggered another round of AI-stock selling because OpenAI has become one of the most important drivers of global computing demand.
The interpretation is important.
The report did not suggest OpenAI is shrinking.
Its revenue is still growing rapidly.
The market reaction instead reflected how high investor expectations had become.
If one of the fastest-growing AI companies in the world can disappoint simply by growing less explosively than expected, investors naturally begin reconsidering the valuations of its suppliers.
That is precisely the vulnerability facing Chinese hardware names.
Investors are questioning whether AI infrastructure spending can keep accelerating
This is the central global concern.
Technology companies are committing extraordinary amounts of capital to AI infrastructure.
Morgan Stanley estimates the industry may require around $1.5 trillion of external financing by 2028 to support planned data-center investment.
That financing must eventually generate adequate returns.
Investors are beginning to ask:
How many data centers are actually needed?
How quickly will those data centers reach high utilization?
How profitable will AI services become?
And how long can hardware spending continue growing faster than revenue?
None of those questions imply AI is a bubble with no economic value.
They imply that the price paid for AI exposure matters.
Higher interest rates make expensive technology stocks harder to justify
Another reason valuations are under pressure has little to do with AI itself.
Global bond yields remain elevated.
Higher yields increase the discount rate investors use when valuing future profits.
That tends to hurt growth companies most because a larger portion of their value comes from earnings expected years in the future.
The effect has already been visible across Chinese technology stocks.
South China Morning Post reported that elevated U.S. Treasury yields, oil prices and global financing costs helped push the CSI 300 to a 13-month low in late September while the tech-heavy Star Market 50 fell more than 4% in a single session.
That is why even companies with strong earnings can fall when rates rise.
Investors simply become unwilling to pay the same multiple.
Chinese AI hardware has another problem: geopolitics
Valuation is only part of the pressure.
Chinese optical companies face a direct U.S. policy risk.
Washington has been considering tighter restrictions on Chinese optical transceivers and other data-center hardware.
In August, Reuters reported that the Trump administration was considering banning imports of newer Chinese optical modules on national-security grounds.
That news caused immediate damage.
Zhongji Innolight fell around 10%.
Eoptolink dropped roughly 10%.
TFC Optical lost about 6%.
The market’s concern is not simply lost federal contracts.
It is escalation.
Investors worry targeted restrictions could eventually expand into wider bans.
Four U.S. senators have also targeted Chinese optical suppliers
The geopolitical pressure continued in late September.
A bipartisan group of U.S. senators introduced legislation that would restrict government procurement of optical transceivers made by companies including Innolight and Eoptolink.
The immediate commercial impact may be limited.
But Citigroup analysts said the bigger risk is whether government restrictions eventually broaden beyond national-security systems.
That makes the stocks unusually sensitive to political headlines.
A company can deliver excellent earnings and still suffer if Washington threatens access to one of its most important markets.
Overseas revenue exposure is enormous
The vulnerability is especially clear for China’s optical leaders.
Innolight generated a large majority of its revenue overseas.
Eoptolink reportedly gets roughly 96% of sales from outside China.
Innolight’s foreign sales have also represented more than 90% of revenue in some periods.
That international exposure helped create their spectacular earnings growth.
It also creates geopolitical concentration risk.
Companies whose strongest customers are U.S. hyperscalers become vulnerable when U.S.-China technology policy shifts.
The same business model that drove the bull case becomes part of the bear case.
But the U.S. may not be able to replace Chinese optics easily
There is an important counterargument.
Chinese suppliers are not merely cheap alternatives.
They are deeply embedded in the global AI infrastructure supply chain.
Counterpoint estimates Chinese companies control close to two-thirds of global optical-transceiver supply.
Analysts have warned that western suppliers may be unable to replace Innolight and Eoptolink’s volumes quickly.
That means aggressive import restrictions could hurt U.S. data-center operators too.
Companies such as Amazon, Microsoft and other hyperscalers rely on large-scale optical connectivity to keep expensive AI accelerators efficiently networked.
Restricting Chinese suppliers could raise costs and slow deployments.
That interdependence may limit how aggressive Washington can ultimately become.
Jefferies has argued the market may be overreacting
Jefferies previously said it saw a relatively low probability that a sweeping U.S. ban would ultimately materialize.
The bank argued some restrictions could be bargaining tools within wider U.S.-China negotiations.
That is important because geopolitical selling often creates exaggerated market reactions.
Investors may price in the worst possible policy outcome long before that outcome occurs.
If restrictions remain narrow, heavily sold shares could rebound.
But the uncertainty itself justifies some valuation discount.
That risk did not exist—or was not being priced seriously—during the strongest part of the rally.
China may also reopen access to some Nvidia chips
Another complication comes from Beijing itself.
Reports in late September suggested China may allow companies such as Alibaba and ByteDance to purchase certain Nvidia processors again.
That would improve access to AI computing.
But it could also create more competition for domestic Chinese chipmakers.
For investors who had priced Chinese semiconductor companies on the assumption of rapid localization, that matters.
Beijing’s long-term objective remains technology self-reliance.
But if Chinese internet companies can buy competitive Nvidia hardware again, the transition toward purely domestic alternatives may become slower.
That puts pressure on some of the most expensive local semiconductor valuations.
Beijing’s self-reliance push is still real
None of this means China is abandoning domestic technology development.
Bank of America says localization has moved beyond political rhetoric and is producing measurable improvements in Chinese technology standards and market share.
Chinese companies are making progress in:
AI processors;
networking;
memory;
optics;
robotics;
and advanced manufacturing.
The government has strong strategic incentives to reduce dependence on foreign technology.
That remains a powerful long-term tailwind.
But Bank of America also highlighted exactly the two risks now punishing the stocks:
valuation and geopolitics.
The technology story may remain intact even when the investment story becomes less attractive.
Robotics has shown the same pattern
The valuation correction is not limited to optical hardware.
Chinese humanoid-robotics stocks have also experienced dramatic reversals.
Unitree Robotics, one of China’s most celebrated robotics companies, reportedly fell about 55% from its peak only weeks after its blockbuster listing.
Former Alibaba chief Daniel Zhang recently described Unitree as a strong company with visionary leadership but warned that no company could withstand expectations that become too extreme.
That may be the clearest description of the entire Chinese AI-hardware correction.
Great technology.
Real growth.
Too much expectation.
Investors had started valuing potential rather than profits
AI rallies tend to work this way.
At first, investors buy companies because revenue is accelerating.
Then because profit is accelerating.
Then because the total addressable market is enormous.
Eventually, valuations begin depending on businesses that do not yet exist.
Future optical standards.
Future humanoid robots.
Future AI servers.
Future chip market share.
Future export growth.
Once too much future success becomes embedded in today’s price, even minor disappointments can cause violent declines.
That is what China’s hardware sector appears to be experiencing now.
The correction has been particularly brutal for growth indices
The tech-heavy Star Market 50 fell more than 30% during the third quarter, according to market reporting.
That contributed to a broader decline in mainland stocks and helped push the CSI 300 to its weakest level in more than a year.
The decline shows how concentrated investor enthusiasm had become.
When one dominant theme reverses, the entire growth market can suffer.
China has seen this before.
Internet stocks.
Electric vehicles.
Solar equipment.
Biotech.
Each began with legitimate structural growth.
Each eventually experienced periods where valuation became detached from earnings.
AI hardware may now be entering that normalization phase
The word “crash” may be too dramatic.
Normalization may be more accurate.
Hardware companies remain profitable.
Data-center demand remains real.
AI investment remains enormous.
China continues prioritizing technology independence.
But markets no longer appear willing to assume every hardware company will capture unlimited growth.
Investors are differentiating again.
That is healthy for a market.
It is painful for stocks that were priced as if competition did not exist.
Competition inside China is intensifying too
Domestic rivalry is becoming another valuation problem.
China has dozens of semiconductor, networking, optical and robotics companies competing for the same policy-supported markets.
That can accelerate innovation.
It can also destroy margins.
If several companies expand production simultaneously, pricing power weakens.
A huge addressable market does not guarantee huge profits when everyone is chasing it.
That is one of the most important lessons from China’s solar and electric-vehicle industries.
Capacity can grow faster than demand.
Investors are beginning to ask whether AI hardware could eventually face something similar.
The optical market remains one of the strongest AI infrastructure niches
There is still a convincing bullish argument.
AI clusters require dramatically more networking bandwidth than traditional data centers.
As accelerators become faster, communication between those accelerators becomes increasingly critical.
That is why high-speed optical links such as:
800G;
1.6T;
and eventually 3.2T
are seeing rapid demand growth.
Citi has estimated the global optical interconnect market could reach around $92 billion by 2028, with average selling prices supported by the move toward faster products.
If that forecast is close to reality, leading Chinese suppliers still have substantial growth ahead.
The question is whether their stocks already priced most of it.
Strong fundamentals do not guarantee strong shares
This distinction is crucial for investors.
A company can:
increase revenue;
increase profit;
gain market share;
and still produce a negative stock return.
Why?
Because the stock price may have increased even faster beforehand.
If earnings grow 30% but investors previously priced in 60%, the stock can fall.
That is why valuation becomes so important after a major rally.
China’s AI hardware companies may still deliver excellent operating results.
Their shares no longer automatically need to rise.
Innolight itself still has powerful earnings momentum
The company’s latest numbers demonstrate why bears should be cautious about dismissing the sector entirely.
Revenue and profit growth remain strong.
Its global position in optical connectivity remains formidable.
Analyst price targets still imply significant potential upside from recent levels, although estimates vary widely.
That creates a battle between two camps.
Bulls argue the market is temporarily frightened by rates and geopolitics.
Bears argue current earnings are already embedded in the price and future risks are being underestimated.
Both positions can be rational simultaneously.
China’s wider stock market could actually become more attractive
Interestingly, the hardware correction does not necessarily imply a bearish outlook for all Chinese equities.
Reuters Breakingviews argued on October 9 that China may be approaching a broader “slow bull market,” supported by structural reforms, improved private-sector policy and enormous household savings that could eventually move into equities.
That highlights an important rotation possibility.
Money leaving expensive AI hardware does not necessarily leave China.
It can move into:
banks;
consumer stocks;
industrial companies;
dividend payers;
and cheaper technology names.
That would make the market healthier by reducing dependence on one crowded theme.
Valuation dispersion is now becoming more important
During the strongest phase of the AI rally, investors often bought entire sectors.
Now the market is becoming more selective.
Bank of America has explicitly differentiated among Chinese AI companies based on valuation, technological positioning and geopolitical exposure.
That is likely to continue.
Companies with:
real earnings;
strong balance sheets;
diversified customers;
reasonable multiples;
and limited geopolitical exposure
may outperform.
Companies trading primarily on narrative may struggle.
That is typical after speculative phases mature.
Global AI weakness makes the Chinese correction more difficult to escape
China cannot isolate itself from Wall Street sentiment.
The global AI supply chain is deeply interconnected.
When Nvidia falls, Asian suppliers react.
When Broadcom guidance raises questions, optical stocks react.
When OpenAI spending expectations change, networking companies react.
That connection became particularly obvious this week as global semiconductor shares fell across the U.S., Europe and Asia.
China may have its own technology ecosystem.
Its stocks still trade inside a global financial ecosystem.
Rising oil prices and bond yields add another layer
Technology valuations are also being squeezed by macroeconomic conditions.
Brent crude has remained above $100 per barrel amid Middle East instability.
Higher energy prices can sustain inflation.
Higher inflation can keep interest rates elevated.
Higher rates reduce the valuation investors are willing to assign to growth stocks.
That chain matters because Chinese AI hardware stocks are among the market’s longest-duration assets.
Much of their valuation depends on profits investors expect years from now.
The higher the discount rate, the less those future profits are worth today.
The correction could ultimately make the sector healthier
Rallies based entirely on expanding valuation multiples are fragile.
A correction forces investors back toward fundamentals.
Which company has real orders?
Which company has real cash flow?
Which products have genuine technological advantages?
Which customers are diversified?
Which businesses can survive weaker AI spending?
Those questions become much more important when stocks stop rising automatically.
In that sense, the current selloff may ultimately improve market discipline.
China’s technology ambitions have not changed
Beijing still wants leadership in:
AI;
semiconductors;
robotics;
advanced manufacturing;
and digital infrastructure.
Those industries remain strategic priorities.
Chinese companies will continue receiving policy support.
Domestic demand for computing power will continue rising.
The long-term structural thesis remains powerful.
What has changed is investor willingness to pay any price for exposure.
That is an important difference.
The biggest danger is assuming every AI company will be a winner
History suggests transformative technologies can create enormous economic value while destroying investment capital along the way.
The internet changed the world.
Many dot-com stocks still went bankrupt.
Solar energy became globally dominant.
Many solar manufacturers produced terrible shareholder returns.
Electric vehicles continue growing.
Numerous EV stocks have collapsed.
AI can become equally transformative without every AI hardware company deserving an extreme valuation.
That is the lesson markets are starting to relearn.
The selloff is really a test of expectations
China’s hardware companies now face an unusually high bar.
Investors want strong growth.
But strong growth may no longer be enough.
They want earnings beats.
Margin expansion.
Large AI orders.
Geopolitical resilience.
And proof that enormous valuations can be justified.
Every quarter becomes a test.
Every U.S. restriction becomes a risk event.
Every change in global AI spending becomes a catalyst.
That is the price of becoming the market’s favorite trade.
This is not the end of China’s AI boom—but it may be the end of easy money
That may be the clearest conclusion.
The technologies remain important.
The orders remain real.
The earnings remain substantial.
China’s manufacturing advantages remain formidable.
But the market has moved from optimism to scrutiny.
Investors no longer appear willing to assume that every company linked to AI infrastructure deserves a premium indefinitely.
China’s AI hardware stocks became some of the biggest winners of 2026 because investors believed they were selling the essential machinery behind the next technology revolution.
That argument may still be correct.
But after extraordinary rallies and increasingly demanding valuations, the market is now confronting a harder question:
even if AI changes everything, how much of that future did investors already pay for?
