Zhongji Innolight’s CPO Rally Reversed as Its Market Value Fell Below RMB 1 Trillion

Zhongji Innolight closed 4.29% lower on September 2, pushing its market value below RMB 1 trillion despite record first-half earnings. The CPO favorite has now retreated more than 40% from its June intraday high, according to market data. That reversal is more significant than a bad day for one crowded stock.

Investors are confronting a difficult split. Demand for high-speed optical modules remains strong, but the valuations attached to that demand assumed years of near-perfect execution. Zhongji Innolight must now defend those expectations while optical networking moves from pluggable modules toward integrated silicon photonics.

The central contest is therefore not Zhongji Innolight against another Chinese supplier. It is the company’s exceptional operating growth against a valuation that had already priced in a much larger future market. CPO strengthens the long-term demand story, yet it also introduces new architectures that can redistribute value across the supply chain.

The CPO Selloff Has Become More Than a Market Correction

Zhongji Innolight’s decline turned an ordinary sector pullback into a test of how investors value the AI networking supply chain.

The company’s Shenzhen-listed shares closed at RMB 822.40 on September 2. Its total market capitalization ended near RMB 969 billion, depending on the share count and data provider used. The stock lost 4.29% during the session, while trading value reached about RMB 22.06 billion.

That move followed more than two months of pressure. Zhongji Innolight reached an intraday record of RMB 1,416.88 on June 22. The September 2 close placed the shares roughly 42% below that high.

The company first slipped below the RMB 1 trillion threshold in late August. Its brief recovery failed to hold, and the second breach arrived one day after an announced share repurchase. That sequence made the decline harder to dismiss as a momentary reaction.

The weakness extended across China’s optical networking leaders. Eoptolink fell 4% on September 2, while TFC Optical Communication declined 0.70%. Other optical component and printed circuit board names also weakened during the broader technology selloff.

The overall market provided part of the explanation. China’s major indexes declined, and thousands of mainland-listed stocks closed lower. The ChiNext Index, which contains many growth companies, fell 2.39%.

However, Zhongji Innolight’s longer drawdown began well before that session. A broad market decline can explain the final push below RMB 1 trillion. It cannot fully explain a retreat exceeding 40% from the June peak.

Financing activity adds another layer. The company had a margin financing balance of RMB 32.10 billion on September 1. That was the second-highest balance among mainland-listed companies, behind Ping An Insurance.

Large financing balances do not prove that forced selling caused the decline. They do indicate that leveraged investors remained heavily involved. A volatile stock becomes more sensitive when investors must add collateral or reduce positions during falling markets.

Institutional disclosures also attracted attention. Reports based on Hong Kong exchange filings showed reduced long positions among several global banks during late August. Such disclosures can reflect hedging, client activity, or changing exposure, so they should not be treated as simple directional bets.

Still, the timing reinforced a broader perception that investors were reducing risk around crowded AI hardware trades. Zhongji Innolight had become one of the most visible expressions of that trade in China. Its pullback therefore carried implications beyond its own shareholder register.

This event matters because the company’s business did not collapse alongside its stock. Revenue, profit, product shipments, and customer spending remained strong in its latest reported period. The market instead questioned how much future performance was already embedded in the valuation.

That distinction changes the story. This is not a conventional case where deteriorating earnings drag down a stock. It is a reversal where exceptional earnings have stopped being sufficient.

Record Earnings Could Not Defend a Trillion-Yuan Valuation

The stock weakened because investors moved their attention from current growth to the increasingly demanding assumptions behind that growth.

Zhongji Innolight reported first-half 2026 revenue of RMB 41.78 billion, up 182.49% from the prior-year period. Net profit attributable to shareholders reached RMB 13.65 billion, an increase of 241.70%.

Its gross margin rose to 46.25%, up 6.92 percentage points year over year. Second-quarter revenue reached RMB 22.28 billion, while quarterly attributable profit reached RMB 7.92 billion.

Those figures came from the company’s interim report. They show that the optical module expansion was delivering operating leverage, rather than growth purchased through falling margins.

The company attributed the performance to strong overseas demand for AI computing infrastructure. Shipments of 800-gigabit and 1.6-terabit optical modules increased, and higher-speed products represented a larger share of its business.

An optical transceiver converts electrical signals into light, sends that light through fiber, and converts it back at the destination. These modules connect servers, accelerators, switches, and other equipment inside data centers.

AI clusters need far more of those connections than conventional enterprise computing systems. Thousands of accelerators must exchange data quickly during model training and inference. Network delays can leave expensive processors waiting instead of computing.

Higher transmission rates therefore support both faster clusters and better utilization. The shift from 400G to 800G, followed by 1.6T, has increased the value and complexity of the optical equipment surrounding AI accelerators.

Zhongji Innolight entered this cycle with substantial production scale and relationships with large overseas customers. Its Hong Kong listing documents described it as the world’s largest optical interconnect supplier by revenue since 2021.

The company’s listing prospectus cited a 21.2% share of the overall optical interconnect market in 2025. It placed the company’s high-speed data-center optical interconnect share at 28.1%.

Those estimates support a clear bullish case. A leading supplier can capture disproportionate benefits when cloud companies accelerate network upgrades. Scale can also improve purchasing, manufacturing yields, and delivery reliability.

Yet the same figures reveal the valuation challenge. A company already holding more than one-fifth of its market needs rapid industry expansion, additional share gains, or both. Maintaining existing leadership is not enough when the stock reflects much larger future profits.

Consensus estimates cited by Chinese financial media placed Zhongji Innolight’s expected revenue at RMB 110.2 billion in 2026. They projected RMB 216.6 billion for 2027 and RMB 310.4 billion for 2028.

Those forecasts are not company guidance. They are market expectations that can change quickly. However, their steep trajectory shows what investors had begun to demand from the company.

The projected 2028 revenue would be more than seven times the company’s first-half 2026 sales. Converting that target into profit would also require durable margins during a period of fast capacity expansion and technological change.

Strong earnings can disappoint when expectations rise even faster. The market does not reward the absolute level of growth alone. It reacts to the difference between reported results and the future already reflected in the share price.

That mechanism helps explain why the first-half report failed to restore the June valuation. Investors had already accepted that 800G and 1.6T shipments would grow rapidly. The harder question concerned the duration and profitability of that growth.

The company began repurchasing shares on September 1. It bought approximately 374,100 shares for about RMB 318 million, with transaction prices ranging from RMB 838.16 to RMB 870.

A repurchase can signal management confidence and create incremental demand. It cannot remove uncertainty about customer concentration, capacity, technical transitions, or long-term margins. The September 2 decline demonstrated that limitation.

Co-Packaged Optics Changes Who Captures the Value

CPO expands the networking opportunity while challenging the pluggable module structure that created Zhongji Innolight’s recent earnings surge.

CPO means co-packaged optics. The architecture places optical engines close to a switch application-specific integrated circuit, instead of relying entirely on removable transceivers at the system’s front panel.

Traditional pluggable modules offer a clear operational advantage. Technicians can remove a failed module without replacing the entire switch assembly. Suppliers can also upgrade modules while retaining familiar system designs.

The disadvantage appears as bandwidth rises. Electrical signals must travel between the switch chip and the front-panel module. At higher speeds, those paths consume more power and become more difficult to manage.

CPO shortens the electrical connection by bringing the optical conversion closer to the switching silicon. That can reduce signal loss, increase bandwidth density, and lower power consumption.

This shift has moved beyond laboratory demonstrations. Broadcom said its 51.2-terabit-per-second Bailly platform combines eight silicon photonics engines with a Tomahawk 5 switch chip. The company claims 70% lower power for optical interconnects than conventional pluggable designs.

Nvidia has also committed to the architecture. It introduced Quantum-X Photonics for InfiniBand and Spectrum-X Photonics for Ethernet, placing optical engines beside its networking silicon.

Nvidia says the design uses four times fewer lasers and delivers 3.5 times greater power efficiency than traditional methods. It also claims ten times better network resilience at scale.

These are vendor claims tied to specific systems and comparison methods. They should not be generalized across every network. They nevertheless show why hyperscale operators are willing to evaluate an architectural change.

Nvidia’s photonics roadmap originally placed Quantum-X availability in 2025 and Spectrum-X Ethernet systems in 2026. Later product materials described Spectrum-X Photonics availability during the second half of 2026.

That timing makes Zhongji Innolight’s current valuation debate unusually important. CPO is approaching commercial deployment while demand for 800G and 1.6T pluggable products remains high.

The two architectures will coexist rather than switch overnight. Nvidia specifically identified Innolight and Eoptolink among suppliers supporting pluggable optical transceiver technologies around its networking platforms.

This coexistence creates both opportunity and tension. Pluggable modules can continue growing across conventional scale-out networks, where accelerators communicate across racks. CPO can gain adoption in the highest-density switches and scale-up systems.

Scale-up networks connect processors within tightly integrated computing domains. They require high bandwidth, low latency, and predictable performance. Those requirements make them attractive early targets for integrated optics.

LightCounting expects CPO to emerge as a leading option for scale-up connectivity because of its bandwidth density and reliability. Its market forecast projected 30% to 35% annual optical market growth in 2025 and 2026.

The firm projected slower annual growth of 15% to 20% between 2027 and 2030. It also warned that optical markets have experienced periodic declines when cloud spending slows or inventories accumulate.

That forecast captures the central contradiction. The long-term market can expand significantly while individual suppliers still face shorter cycles, architectural shifts, and pricing pressure.

Zhongji Innolight has invested in silicon photonics and remains positioned to supply high-speed modules. It could also participate in external laser sources, optical engines, or related CPO components.

However, participation does not guarantee that the economics will match those of premium pluggable transceivers. An integrated switch architecture gives semiconductor and system vendors greater control over design choices and procurement.

Broadcom can combine its switching silicon, photonic engines, packaging, connectors, and manufacturing processes within one platform. Nvidia can coordinate networking chips, software, accelerators, and photonics around its own AI systems.

That integration can shift technical decisions upstream. Optical suppliers may win large component orders, but they may operate within architectures controlled by a smaller number of platform companies.

The transition also creates serviceability questions. A removable transceiver can be swapped quickly after failure. Optical engines placed near expensive switching silicon require different approaches to testing, redundancy, lasers, fibers, and repairs.

External laser modules and detachable fiber connectors address some of those problems. They do not eliminate the manufacturing and maintenance complexity associated with tightly integrated systems.

CPO is therefore not simply a larger version of the existing module market. It changes the product boundary. It can expand total optical content while redistributing control, margins, and technical risk.

For Zhongji Innolight, the best outcome would combine continued pluggable growth with meaningful roles in integrated photonics. The more difficult outcome would involve CPO adoption accelerating while platform vendors capture more of the system value.

What the Bull Case Still Has to Prove

The selloff does not invalidate AI optical demand, but the valuation requires evidence that growth can survive concentration, competition, and product transitions.

The strongest bullish argument begins with cloud capital spending. Large technology companies continue investing in accelerators, data centers, power infrastructure, and networking. Every deployed AI cluster requires connections inside racks and across facilities.

Network bottlenecks also become more costly as accelerator prices rise. An operator gains little from adding processors if congestion leaves those processors idle. That gives high-performance optics strategic importance within the overall system budget.

Zhongji Innolight’s first-half results confirm that this demand reached its factories. Revenue nearly tripled, profit more than tripled, and gross margin expanded. Those are measurable outcomes, not distant projections.

The rise of 1.6T modules provides another growth step. More bandwidth per module supports denser clusters and faster switching platforms. Customers moving from 800G to 1.6T can increase the revenue opportunity even without proportional unit growth.

CPO can extend the cycle further. Integrated photonics does not remove optical components from the system. It moves those components closer to the switch chip and changes their packaging.

A large optical manufacturer with silicon photonics capabilities can adapt to that transition. Existing customer relationships, production knowledge, testing capacity, and component sourcing remain valuable.

Yet four uncertainties deserve equal attention.

First, customer concentration creates negotiating risk. A relatively small group of cloud and AI platform companies accounts for much of the advanced optical market. Their purchasing schedules can quickly affect factory utilization and inventory.

Those customers also possess significant leverage. They can qualify multiple suppliers, redesign network architectures, or delay deployments. Strong demand does not automatically protect supplier margins.

Second, competition is expanding across technologies. Eoptolink competes in high-speed modules, while TFC Optical supplies optical components and assemblies. International vendors such as Coherent and Lumentum also participate across lasers, modules, and photonics.

Different companies are pursuing silicon photonics, linear pluggable optics, near-packaged optics, and CPO. The eventual market will probably contain several architectures, chosen according to distance, bandwidth, reliability, and service requirements.

Third, manufacturing becomes harder at higher speeds. A 1.6T module requires tighter control over optics, thermal behavior, signal integrity, packaging, and testing. Rapid capacity expansion can expose yield problems or component shortages.

CPO increases integration further. A defect in one element can affect a more valuable assembly. Suppliers must demonstrate that they can manufacture integrated optical engines at scale without sacrificing economics.

Fourth, the industry remains cyclical. Cloud companies can build ahead of demand, suppliers can accumulate inventory, and component shortages can turn into excess capacity. Optical markets experienced corrections in 2019 and 2022.

The current cycle is larger because generative AI created a new source of demand. That does not repeal inventory cycles. It can instead amplify them when companies simultaneously expand capacity against optimistic forecasts.

Valuation magnifies each risk. A company valued for several years of rapid expansion has little room for slower shipments, lower prices, or delayed product ramps. Even excellent results can disappoint if they merely match expectations.

The 40% retreat can therefore coexist with a positive industry outlook. Investors are adjusting the price paid for growth, not necessarily rejecting the demand itself.

There is also an important classification problem. Chinese market coverage often labels optical module companies as CPO leaders even when current revenue comes mainly from pluggable products.

That shorthand connects the stocks to a popular technology theme. It can obscure which companies already generate revenue from integrated optical engines, which remain in qualification, and which mainly benefit indirectly.

Readers should separate three questions. How quickly will co-packaged optics ship at scale? Which components will Zhongji Innolight supply? What margins will those components generate compared with existing modules?

Until companies disclose clearer CPO revenue, shipment, and customer data, broad leadership labels remain less useful than product-level evidence. The market decline reflects that demand for greater precision.

Three Signals Will Decide Whether the Reversal Lasts

The next phase depends on shipment quality, commercial CPO deployment, and whether earnings expectations stabilize after the valuation reset.

The first signal is Zhongji Innolight’s next financial report. Investors should focus on 1.6T shipment growth, gross margin, inventory, receivables, and operating cash flow.

Revenue growth alone will not settle the debate. Higher inventory could indicate preparation for demand, but it could also signal slower conversion into customer shipments. Cash collection will help distinguish those possibilities.

Gross margin will reveal whether supply constraints and premium products continue supporting profitability. A stable margin would strengthen the case that the company retains pricing and manufacturing advantages.

A sharp margin decline would weaken that argument. It could indicate pricing pressure, an unfavorable product mix, lower factory utilization, or higher costs associated with new production.

The second signal is commercial deployment of Nvidia’s Spectrum-X Photonics systems. Nvidia has described systems offering up to 409.6 terabits per second of switching capacity and 512 ports at 800 gigabits per second.

The important milestone is not another product announcement. It is evidence that customers are deploying those systems in volume, followed by disclosure about the participating optical supply chain.

Broad deployment would strengthen the long-term CPO thesis. It would also force investors to identify which suppliers provide optical engines, lasers, connectors, fibers, packaging, and testing.

If deployments slip or remain limited, pluggable modules will retain their central role for longer. That outcome could benefit existing module revenue while delaying the integrated opportunity attached to the CPO label.

The third signal is the direction of customer capital spending and orders for 2027. Zhongji Innolight has said that key customers provided order visibility into next year, according to local reporting.

Investors need that visibility to become shipments and cash. Sustained spending from cloud companies would support 800G and 1.6T demand while giving suppliers time to prepare integrated products.

Reduced spending, delayed data centers, or inventory adjustments would weaken the high-growth assumptions behind earnings forecasts. The effect would spread across optical modules, networking chips, printed circuit boards, and photonic components.

These signals matter more than whether the stock immediately recovers RMB 1 trillion in market value. That threshold is psychologically significant, but it is not an operating metric.

The selloff has already exposed the market’s revised standard. Investors want proof that Zhongji Innolight can convert a historic optical cycle into durable earnings across two architectures.

For technology buyers, the outcome will shape equipment availability, power requirements, and vendor choice. Faster optical links can improve accelerator utilization, but integrated designs may change maintenance practices and supply dependencies.

Developers rarely purchase optical modules directly. They still experience the consequences through training availability, inference latency, cloud capacity, and the cost of running large models.

Knowledge workers and product teams should watch the infrastructure layer for the same reason. AI services depend on physical networks, and networking constraints can influence product limits long before users see the hardware.

The practical question is no longer whether AI needs more optical bandwidth. Reported shipments, product roadmaps, and cloud spending have already established that direction.

The question is who captures the value as the industry moves from pluggable modules toward integrated photonics. Watch Zhongji Innolight’s margins, Nvidia’s CPO deployments, and hyperscaler orders over the next quarter. Together, those signals will show whether the decline was a valuation reset or the beginning of a deeper change in the optical supply chain.