Zhongji Innolight Says Orders Now Reach 2028, but the Optical Boom Faces a New Test

Zhongji Innolight says key customers have provided substantial product guidance for 2028, extending its visibility far beyond a normal hardware order cycle. The statement suggests the AI infrastructure buildout is creating commitments years before the equipment reaches a data center. It also raises a harder question: how much of that demand will become delivered, profitable revenue?

The Chinese optical-module supplier made the remarks during a July 28 conference call, according to a July 28 report. The report said nearly all customer orders cover the rest of 2026. Some extend through 2027, with delivery schedules set by month.

Management also described strong demand for 800G, 1.6T, 2.4T, and near-packaged optics, commonly shortened to NPO. That mix matters because it spans products already shipping at scale and architectures still moving toward broader deployment.

The claim is not an independently audited backlog disclosure. No customer names, contract values, cancellation terms, or product-level volumes accompanied it. The 2028 guidance should therefore be read as a customer planning signal, not guaranteed revenue.

Even with that qualification, the time horizon is unusual. Optical-module orders traditionally move with cloud capital spending, switch upgrades, and periodic inventory corrections. Guidance reaching 2028 implies major AI infrastructure buyers are reserving access before their next network designs are fully deployed.

That pressures competing suppliers such as Eoptolink, Coherent, and Lumentum to secure components, production capacity, and customer qualifications earlier. It also pits two optical architectures against each other. Pluggable transceivers support today’s deployments, while co-packaged optics could change where optical conversion happens inside future switches.

The order horizon has moved from quarters to years

The most important change is not a new module speed. It is the extension of customer planning into 2028.

According to the July 28 account, Zhongji Innolight said almost all customer orders already covered 2026. Certain customers had placed orders extending into 2027, supported by monthly delivery plans. Some key customers had also supplied 2028 guidance for new products.

This progression represents several levels of commercial certainty. A scheduled order for monthly delivery is more concrete than an early product forecast. A product forecast can help a supplier plan capacity, but it does not necessarily carry the same legal or economic weight.

The distinction is especially important because the report did not publish an order value. It characterized the 2028 guidance amount as substantial without defining that term. It also did not explain whether the amount represented expected revenue, requested capacity, or another internal planning measure.

Still, the reported chronology provides a useful view of customer behavior. Buyers are not simply asking whether 1.6T products will be available. They appear to be coordinating delivery windows while discussing the products that follow them.

That behavior fits an AI infrastructure market in which network planning starts long before installation. Cloud operators must align accelerators, switches, optical modules, fiber, power, cooling, and building capacity. A shortage in one category can leave much more expensive computing equipment underused.

Optical transceivers convert electrical data into light and back again. They sit at the ends of fiber links connecting accelerators, switches, racks, and sometimes separate data centers. Faster accelerators create little value if the network cannot move model parameters and intermediate results quickly enough.

The company’s product range gives it exposure to several stages of this upgrade. Zhongji Innolight’s 2025 annual report described products covering 100G through 1.6T for cloud data centers. It also listed both traditional externally modulated laser designs and silicon-photonics approaches for high-speed modules.

The filing reported 2025 revenue of 38.24 billion yuan, up 60.25 percent from 2024. Optical transceiver revenue reached 37.46 billion yuan, representing 97.95 percent of total revenue. Overseas markets generated 34.64 billion yuan, or 90.58 percent of the total.

Those figures show that Zhongji Innolight entered 2026 with considerable exposure to global cloud spending. They do not prove that every reported 2028 forecast will convert into sales. However, they explain why forward signals from its customers deserve attention across the optical supply chain.

A multiyear horizon also changes how the company must operate. It needs to decide which components to reserve, which production lines to expand, and which product architectures will dominate. Each decision requires capital before the related revenue becomes certain.

That creates the central tension behind the announcement. Longer visibility can reduce planning risk, but it can also magnify the cost of betting on the wrong design.

AI clusters are making optical capacity strategic

The reported orders suggest optical networking has become a capacity constraint, not a peripheral accessory to AI computing.

An AI cluster needs many accelerators to operate as one coordinated system. Those processors exchange large volumes of data during training and inference. As cluster sizes rise, the number and speed of network links rise with them.

This demand is pushing 800G modules into broader deployment while 1.6T enters mass production. The labels describe aggregate data rates. An 800G module carries up to 800 gigabits per second, while a 1.6T module doubles that total rate.

Market researchers have connected this transition directly to AI spending. TrendForce estimated that the global AI optical-transceiver market would reach $26 billion in 2026. Its market outlook identified component shortages as a primary constraint on capacity expansion.

TrendForce separately projected that modules rated at 800G or above would exceed 60 percent of data-center optical-module shipments in 2026. That would be a sharp increase from 19.5 percent in 2024. Forecasts remain estimates, but the direction supports Zhongji Innolight’s description of fast growth across several product generations.

LightCounting has also raised its expectations. Its April forecast said distributed AI clusters would increase demand for high-speed connections between data centers. It expects 800G coherent links to grow in 2026 and 1.6T coherent products to gain ground during 2028 and 2029.

Coherent optics use advanced signal processing to carry high-capacity traffic over longer distances. They serve a different link category from many short-reach modules inside a cluster. Their growth shows that optical demand is spreading beyond individual racks and buildings.

The pressure reaches upstream suppliers as well. High-speed modules depend on lasers, digital signal processors, photonic integrated circuits, connectors, and specialized packaging. Expanding final assembly does not help if one critical component remains unavailable.

Long customer forecasts can address that problem. They give manufacturers evidence for component reservations and new production lines. Suppliers can also negotiate longer agreements with their own vendors.

However, long commitments may move scarcity rather than eliminate it. A large buyer that reserves components early can leave smaller cloud operators and competing module makers facing longer lead times. This creates an advantage for suppliers already qualified by leading hyperscalers.

It also raises concentration risk. Zhongji Innolight’s overseas revenue share demonstrates strong access to international demand, but the annual report does not identify the customers behind that concentration. Investors and buyers cannot independently determine how much of the forward signal comes from one network architecture or one spending program.

The company’s position therefore depends on more than total AI capital spending. It depends on where cloud customers spend that money and how many optical links each computing system requires.

If customers continue building large scale-out networks, in which many racks connect through multiple switch layers, transceiver demand stays high. If network designs flatten or consolidate those layers, the optical content per accelerator can change even while overall computing investment grows.

This is why an order extending through 2028 carries more significance than a quarterly sales increase. It implies customers are planning a sustained buildout. Yet it cannot settle how future network architecture will distribute the resulting revenue.

800G and 1.6T are carrying the current cycle

Zhongji Innolight’s near-term opportunity rests on a transition between two pluggable generations, not on one isolated product launch.

The 800G market provides the installed commercial base. These modules connect high-performance switches and computing systems across fiber. Their deployment expanded as cloud operators built denser clusters around recent accelerator generations.

The 1.6T generation doubles aggregate module bandwidth. It commonly combines multiple electrical and optical lanes to reach that total. The upgrade helps networks carry more traffic through each switch port, but it increases power, thermal, signal-integrity, and manufacturing challenges.

Zhongji Innolight’s annual filing described a 1.6T OSFP family using both EML and silicon-photonics platforms. OSFP is a removable transceiver form factor designed for high-speed networking equipment. The company listed DR8 and 2xFR4 variants, which divide traffic across different lane and wavelength configurations.

The technical diversity matters because no single optical technology wins every link. EML designs use semiconductor lasers that modulate light at high speed. Silicon photonics integrates optical functions on a silicon-based platform, potentially improving scale and integration for certain products.

Competitors are pursuing the same transition. Coherent demonstrated 1.6T designs using silicon photonics, indium phosphide lasers, EMLs, and VCSEL technology at OFC 2026. Its pluggable portfolio illustrates how suppliers are maintaining several technology paths rather than assuming one immediate winner.

This competition is not limited to finished modules. Suppliers can differentiate through laser performance, power consumption, manufacturing yield, packaging, signal processing, and customer qualification. A design that works in a demonstration still must survive production testing and continuous operation inside a data center.

Customer qualification creates another barrier. Hyperscalers evaluate reliability, interoperability, thermal behavior, and manufacturing consistency before approving a module. Once a supplier enters a platform, it can gain large volumes. A failed qualification can exclude it from an entire deployment cycle.

That dynamic helps explain why customers provide guidance years ahead. A 2028 product needs design work and supply planning during 2026. Waiting until the deployment year would leave too little time for engineering, qualification, and capacity expansion.

The reported inclusion of 2.4T products extends the roadmap further. That speed would sit beyond the mainstream 1.6T transition and require another step in lane performance or architecture. Public information about the customers, timing, and exact configurations remains limited.

Therefore, 2.4T should not be treated as a broadly established market today. In the context of the call, it is better viewed as evidence that customers are planning another bandwidth increase.

The sequence also carries execution risk. Zhongji Innolight must continue shipping high volumes of 800G products while scaling 1.6T and developing later generations. Running several generations simultaneously can strain engineering resources and complicate inventory decisions.

A rapid transition might reduce the useful commercial life of some 800G capacity. A slower transition could leave expensive 1.6T investments underused. Long forecasts help manage those possibilities, but they cannot remove them.

This is the practical contest among Zhongji Innolight, Eoptolink, Coherent, Lumentum, and other optical suppliers. The winner will not simply announce the highest speed. It must deliver qualified products at volume while keeping yields, power, and reliability within customer requirements.

Co-packaged optics changes the competitive equation

The largest architectural risk is that future switches may move optics away from removable modules and closer to the switching chip.

Traditional pluggable transceivers sit in ports at the front of a switch. Electrical signals travel from the switch chip across a circuit board before reaching each module. At higher speeds, that electrical path consumes more power and becomes harder to maintain.

Co-packaged optics, or CPO, place optical engines beside the switch application-specific integrated circuit. The shorter electrical path can improve power efficiency and signal integrity. However, the architecture also changes manufacturing, cooling, fiber management, and equipment servicing.

NPO represents an intermediate approach. Near-packaged optics place optical engines close to the switch chip while preserving more physical separation than a fully integrated CPO design. The reported demand for NPO shows customers are evaluating more than one path beyond standard pluggables.

For Zhongji Innolight, this transition is both an opportunity and a threat. Its expertise in optical engines, silicon photonics, lasers, and module manufacturing can remain relevant. Yet the value chain may shift toward switch vendors, advanced packaging providers, and deeply integrated system partnerships.

Nvidia has already moved CPO from a roadmap concept into production. In June 2026, the company said Spectrum-X Ethernet Photonics had entered full production for Vera Rubin systems. Its production announcement identified TSMC, SPIL, TFC, and Foxconn among the manufacturing partners.

Nvidia says the platform places optical technology with the switch package and supports scale-out and scale-across networks. Scale-out connects more computing systems within a cluster. Scale-across links separate AI facilities into a broader computing fabric.

This does not mean pluggable modules disappear. Data centers contain many link types, distances, replacement requirements, and cost targets. Pluggables retain advantages in serviceability, standardization, and supplier flexibility.

A removable module can be replaced without discarding a switch package. Operators can also source compatible products from multiple approved vendors. CPO creates tighter integration, which can improve performance but increase dependency on a smaller technology ecosystem.

The two approaches will therefore coexist across different network layers. The relevant question is how quickly CPO captures high-value switch connections that otherwise would use premium pluggable modules.

Zhongji Innolight’s reported 2028 guidance offers one indirect clue. If the new products include significant NPO or integrated optical-engine demand, the company may be participating in the architectural shift. If guidance remains concentrated in conventional pluggables, its exposure to substitution deserves closer attention.

The public report does not provide that product split. It groups 800G, 1.6T, 2.4T, and NPO demand into a broad statement of confidence. Readers should avoid converting that statement into a precise market-share forecast.

Competition will also occur through partnerships. Nvidia and Coherent announced a multiyear agreement in March 2026 covering advanced optics, manufacturing capacity, and research. The arrangement included a large purchase commitment and a $2 billion Nvidia investment in Coherent.

That optics partnership demonstrates how leading buyers are securing technology and capacity directly. It also shows why a module supplier’s reported customer guidance matters even when customer identities remain undisclosed.

The optical contest is becoming a competition between supply networks, not just component specifications. Module design, laser capacity, advanced packaging, switch integration, and cloud deployment schedules increasingly move together.

Zhongji Innolight’s long visibility suggests it has a seat in those planning cycles. The unresolved issue is how much value it retains when optical and computing architectures become more tightly integrated.

A long forecast is not the same as guaranteed revenue

The bullish signal deserves attention, but the disclosure leaves major questions about enforceability, concentration, and profitability.

The July 28 report uses several commercial terms without defining them. Orders, delivery plans, product guidance, and indicated amounts can describe different levels of customer commitment. Treating them as one firm backlog would overstate what the available evidence supports.

An order may include rescheduling or cancellation provisions. A delivery forecast may guide capacity without requiring the customer to purchase every unit. A product roadmap can change before qualification or mass production.

The absence of disclosed customer names adds another limitation. Hyperscalers operate different accelerator platforms and network designs. Guidance from several independent buyers would be more resilient than guidance concentrated in one architecture.

Customer concentration can amplify both growth and volatility. A large program can fill factories quickly, but a design change can affect several quarters of expected deliveries. Optical suppliers have experienced inventory corrections before when buyers ordered ahead and later adjusted deployment schedules.

Capital spending also remains a dependency. Zhongji Innolight reportedly expects major downstream customers to sustain their investment. That expectation aligns with current AI infrastructure activity, but it is still a management judgment about future budgets.

Cloud operators can redirect capital among accelerators, networking, power, buildings, and proprietary silicon. A growing total budget does not guarantee the same share for every optical product.

Manufacturing execution presents another risk. High-speed modules require consistent yields across sensitive optical and electronic components. Rapid expansion can create quality issues, while component shortages can delay finished products despite strong customer demand.

Margins matter as much as shipment volume. Competition can reduce prices as products mature. New generations may initially carry attractive value, but their development, qualification, and yield costs can offset that benefit.

Architecture can change unit demand as well. Faster ports move more data through each module. Depending on network topology, that can reduce the number of ports required for a given traffic target. Conversely, larger clusters and denser connections can more than offset that reduction.

CPO introduces an additional uncertainty. If integrated optics replace more front-panel modules, traditional transceiver volumes may fall in specific switch layers. Optical suppliers can recover value by providing engines, lasers, or packaging, but their economics will differ.

Geopolitical exposure should not be ignored. Zhongji Innolight generates most of its revenue outside China, according to its annual filing. Cross-border technology restrictions, customer procurement rules, or supply-chain policy changes could affect access to components and markets.

None of these risks invalidates the reported customer signal. They define what must happen before that signal becomes durable financial performance.

The strongest confirmation would come from audited disclosures showing a growing contractual backlog, sustained revenue, stable margins, and rising cash generation. Product-level shipments would provide further evidence that 1.6T demand is converting from qualification into volume.

Until then, the 2028 guidance is best understood as evidence of customer planning intensity. It supports the case for a prolonged optical cycle, but it does not remove the cyclicality of hardware manufacturing.

This distinction is valuable for enterprise buyers too. Long supplier visibility can improve availability, yet it may also encourage customers to secure capacity earlier than they need. That behavior can distort apparent demand if several buyers reserve the same contingency volume.

Procurement teams should separate installed requirements from precautionary reservations. They should also assess alternate suppliers, interoperability, and upgrade paths between pluggable, NPO, and CPO designs.

The market has moved beyond asking whether AI needs faster optics. The difficult questions now concern timing, architecture, and who captures the value.

Three signals will test the 2028 claim

The next evidence should come from shipments, product mix, and customer architecture rather than another broad demand statement.

The first signal is Zhongji Innolight’s 2026 financial performance. Revenue growth alone will not be enough. Investors should compare shipment growth, gross margin, inventory, receivables, and operating cash flow.

If revenue and cash generation rise together, the reported orders are more likely to represent delivered demand. If inventory or receivables grow much faster than sales, execution or collection risk deserves attention.

The mix between 800G and 1.6T will be especially useful. Rising 1.6T shipments would confirm that the next product cycle is moving beyond sampling and qualification. Continued dependence on 800G would not be inherently negative, but it would make the 2028 roadmap less visible.

The second signal is the pace of CPO and NPO deployment. Nvidia’s production ramp provides a concrete test of integrated optics at scale. Adoption by cloud providers beyond early systems would strengthen the case for a structural architecture shift.

That outcome would not automatically weaken Zhongji Innolight. The effect depends on whether the company supplies optical engines or other content inside those systems. More detailed product disclosure would clarify its position.

Slow CPO deployment would support a longer commercial life for high-speed pluggables. Strong reliability, standardized form factors, and straightforward replacement could keep removable modules central across many network layers.

The third signal is evidence of 2027 and 2028 commitment quality. Future disclosures should distinguish scheduled purchase orders from forecasts and capacity reservations. Customer diversification would also make the outlook more credible.

Product qualifications offer another form of confirmation. A named design win, mass-production milestone, or measurable shipment ramp would carry more weight than an unspecified guidance amount.

These signals should be considered together. Strong near-term shipments without progress on new architectures could indicate a lucrative but aging cycle. Architecture wins without manufacturing execution would show technical relevance but limited financial conversion.

Zhongji Innolight’s July remarks point toward something larger than another strong quarter. They suggest leading AI infrastructure buyers are planning optical capacity across several product generations and multiple years.

That is a meaningful change in the industry’s planning horizon. It reflects the growing difficulty of coordinating computing systems whose value depends on thousands of high-speed connections.

However, the statement does not settle the contest between pluggable modules and integrated optics. Nor does it prove the amount, timing, or profitability of 2028 revenue.

Developers and AI users should care because network availability affects when new computing capacity becomes usable. Enterprise buyers should care because architecture choices made now can shape availability, interoperability, and operating costs for years.

The question to track is no longer whether AI clusters consume more optics. It is whether suppliers can turn early forecasts into qualified products before network architecture changes again. Watch the next financial report, the 1.6T shipment mix, and real CPO deployments. Together, those indicators will show whether Zhongji Innolight’s long-range customer signal represents a durable infrastructure cycle or an unusually early reservation of uncertain demand.