Zhongji Innolight Says 2027 Optical Orders Are Still Accelerating

Zhongji Innolight says customer guidance shows another rapid increase in 800G and 1.6T optical-module demand during 2027, despite an already elevated 2026 base. A 36Kr newsflash distributed through RSSHub reported the claim after an August 23 conference call. The company also expects several newer interconnect products to enter volume production from the second half of 2027.

That forecast changes the argument around AI networking. The question is no longer whether hyperscalers need faster optical links. It is whether suppliers can manufacture enough qualified modules without creating shortages, delays, or excessive customer dependence.

The claim also places pressure on Coherent, Lumentum, Eoptolink, and smaller optical suppliers pursuing the same large cloud customers. Their challenge is not simply matching Zhongji Innolight’s advertised speeds. They must secure lasers, digital signal processors, electrical chips, printed circuit boards, and reliable packaging capacity at comparable scale.

The forecast remains a company statement, not independently verified customer-order data. Zhongji Innolight did not identify the customers, disclose unit volumes, or separate firm purchase orders from planning guidance. Those omissions matter because AI infrastructure plans can change before equipment reaches production.

The 2027 Order Picture Is Becoming More Concrete

Zhongji Innolight is presenting 2027 as a capacity-execution year, not a distant demand scenario.

According to the original order update, the company said major customers had provided 2027 guidance and orders. It expects demand for both 1.6T and 800G products to maintain rapid growth compared with 2026.

An optical transceiver converts electrical data into optical signals and back again. The device lets switches, accelerators, and servers exchange data through fiber at much higher bandwidth than conventional electrical connections can support over comparable distances.

The labels 800G and 1.6T describe aggregate transmission capacity. An 800G module carries up to 800 gigabits per second, while a 1.6T module doubles that headline rate to 1.6 terabits per second.

Those figures do not describe an entire network’s capacity. They describe the data rate available through an individual module or port configuration. Actual performance depends on the lane architecture, reach, power consumption, switch platform, and surrounding network.

The reported guidance extends a message Zhongji Innolight delivered in July. The company then said nearly all customer orders covered the full 2026 calendar year, while some orders already extended into 2027.

That earlier customer guidance described 2027 demand for 800G, 1.6T, 2.4T, and near-packaged optics as relatively visible. It also said some customers had offered preliminary guidance for products expected during 2028.

The August update therefore looks less like an isolated prediction. It suggests customers are converting multiyear infrastructure plans into more detailed supplier schedules.

However, “guidance and orders” can cover commitments with different levels of certainty. A firm purchase order, a capacity reservation, and a nonbinding demand forecast do not carry the same financial meaning.

The public summary did not distinguish among those categories. It also omitted expected selling prices, cancellation terms, delivery schedules, and the proportion assigned to each product generation.

That limits what readers can infer. The update supports a strong demand direction, but it does not establish Zhongji Innolight’s future revenue or margins by itself.

The company’s production preparations offer another useful signal. Zhongji Innolight said it was adding capacity while preparing optical chips, electrical chips, and PCBs for a larger 2027 ramp.

Those preparations require capital and supplier coordination before final customer deployment. Manufacturers generally cannot wait for every order to become unconditional before reserving scarce upstream components.

This creates the article’s central tension. Customers want guaranteed supply, while manufacturers must invest before final volumes become certain.

The stronger the demand signal becomes, the more aggressively suppliers must commit capacity. Yet those commitments create exposure if hyperscalers revise architectures, delay data centers, or redirect spending toward another networking design.

Why 800G and 1.6T Demand Keeps Climbing

AI clusters are increasing optical demand through both faster links and a larger number of connections.

Training and serving large AI models requires thousands of accelerators to exchange data with low latency. As clusters expand, networking performance affects how efficiently those expensive processors operate as one system.

The scale-out network connects more servers and accelerator racks across a data center. Scale-up networking links accelerators more tightly within a rack, tray, or compute domain.

Zhongji Innolight said its future products would address both environments. That distinction is important because the two networks can require different reaches, latency profiles, power budgets, and packaging strategies.

The move from 800G to 1.6T does more than raise a specification. It lets network operators increase switch capacity without doubling the physical number of front-panel modules.

Many 1.6T designs use eight electrical and optical lanes operating at 200 gigabits per second. This architecture raises demands on signal quality, thermal control, laser performance, and manufacturing yield.

A module must also pass customer qualification. Large cloud operators test compatibility, reliability, power consumption, firmware behavior, and performance across realistic operating conditions.

Qualification can take months. It creates a gap between demonstrating a module and shipping it in meaningful volume.

That gap favors established suppliers with manufacturing experience and direct customer relationships. It also explains why high market demand does not automatically produce an equal opportunity for every vendor.

Industry estimates reinforce the direction of Zhongji Innolight’s forecast. TrendForce expects the global AI optical-transceiver market to reach $26 billion in 2026, up from $16.5 billion in 2025.

Its market forecast says 800G-and-faster module shipments should rise from roughly 24 million units to 63 million during that period. It also identifies component shortages as the main constraint on expansion.

These projections are forecasts rather than completed sales. Still, they help explain why suppliers are reserving components before the full 2027 deployment picture becomes public.

The demand comes from more than replacing older modules. New accelerator clusters add ports, while larger networks require additional optical connections between racks and buildings.

Data center interconnect creates another source of demand. Operators increasingly distribute AI workloads across multiple facilities when power, land, or cooling constraints prevent one campus from containing every system.

Those connections can use coherent optics, which encode information through multiple properties of light to carry high data rates over longer distances. Coherent designs differ from the shorter-reach modules commonly used inside a data center.

The market is therefore expanding across several link types at once. Short-reach scale-up connections, rack-to-rack scale-out links, and campus interconnects do not always use identical products.

This diversification can extend the life of 800G even as 1.6T grows. A faster generation does not instantly replace every earlier module because operators optimize speed, reach, cost, and power for each connection.

Zhongji Innolight’s claim that 800G and 1.6T can both grow during 2027 is therefore plausible. A transition can increase demand for the newer generation while the installed base continues absorbing the older one.

The critical issue is the pace of that mix shift. Faster 1.6T adoption could improve revenue per module, but it could also intensify dependence on newer chips and harder manufacturing processes.

Meanwhile, sustained 800G demand offers production stability. It lets suppliers use established lines while customers deploy network platforms that do not yet require 1.6T at every port.

The Real Contest Is Supply Execution

The decisive competition is between promised demand and the industry’s ability to deliver qualified hardware at scale.

Zhongji Innolight competes with Eoptolink, Coherent, Lumentum, Applied Optoelectronics, and other optical-component or module suppliers. Their portfolios overlap, but their levels of vertical integration differ.

Some companies manufacture laser chips and optical components. Others concentrate more heavily on transceiver assembly, design, packaging, or contract manufacturing.

That structure creates interdependence. A module vendor can win customer demand but still miss shipments if a laser, driver, digital signal processor, or PCB supplier cannot expand quickly enough.

High-speed modules combine several scarce capabilities. Manufacturers need suitable optical engines, electrical interfaces, thermal designs, firmware, assembly processes, and acceptable production yields.

Yield describes the share of manufactured units that meet specifications. A small yield decline can reduce effective capacity even when a factory installs more equipment.

Zhongji Innolight’s preparation of optical chips, electrical chips, and PCBs shows that management sees the supply chain as the immediate constraint. Building assembly capacity alone would not solve a shortage elsewhere in the bill of materials.

A June 2026 CMB International analysis described 800G and 1.6T modules as customer-qualified optoelectronic systems rather than simple assemblies. Its industry analysis argued that component access, optical-engine design, yield control, firmware stability, and qualification now converge at the module supplier.

That convergence advantages large vendors. They can negotiate longer supply agreements, operate multiple manufacturing sites, and absorb the expense of parallel customer qualification programs.

It does not eliminate competition. Cloud customers often prefer multiple qualified suppliers to improve negotiating leverage and reduce disruption risk.

A second source can receive a larger allocation when the leading supplier misses delivery or quality targets. That possibility gives challengers a path into an otherwise concentrated market.

Coherent and Lumentum also have substantial upstream photonics capabilities. Their ability to produce or secure lasers can become more valuable when module demand exceeds component supply.

Eoptolink provides another competitive reference. Like Zhongji Innolight, it has benefited from the migration toward faster data-center optics and competes for major international customers.

The outcome will depend on more than who announces the highest-speed prototype. Customers need stable volume, predictable power consumption, low failure rates, and compatible firmware.

Manufacturing geography also matters. Customers seeking resilience may divide orders among suppliers and locations, even when one vendor offers better economics.

That can constrain Zhongji Innolight’s share while preserving overall demand. A customer might accept higher procurement costs to avoid concentrating a critical network component in one company or country.

The company’s broad transceiver portfolio spans products through 1.6T. Its public roadmap shows technical coverage, but production scale and customer acceptance require separate evidence.

This distinction matters for every competitor. A product page can establish availability or intent, but it cannot prove shipment volume, field reliability, or customer allocation.

The supply contest will become harder as 1.6T grows. Each module must handle more data within a limited power and thermal envelope.

Higher speeds increase sensitivity to signal loss and manufacturing variation. They can also raise testing time and capital requirements.

Component suppliers face similar pressure. Laser manufacturers must increase output without sacrificing reliability, while chip suppliers must balance optical demand against other advanced networking programs.

PCBs and connectors also become more demanding at higher lane speeds. Small imperfections can affect signal integrity before the electrical data even reaches the optical engine.

These constraints explain why customer guidance reaches suppliers years in advance. A hyperscaler’s deployment schedule must propagate through module vendors, component companies, foundries, packaging operations, and equipment providers.

The long chain creates risk on both sides. Too little capacity can delay AI clusters, while too much capacity can trigger price pressure when supply catches up.

Zhongji Innolight is betting that the first risk remains larger through 2027. Competitors must decide whether to make the same bet.

2.4T, NPO, and XPO Expand the Manufacturing Challenge

The next products move optical engines closer to switching silicon, trading easier bandwidth scaling for harder integration.

Zhongji Innolight says it is developing or sampling 2.4T, NPO, and XPO products according to customized requirements from large customers. It expects production to begin in the second half of 2027, followed by larger shipments during 2028.

These remain forward-looking targets. Public reporting has not independently confirmed customer qualification, final architectures, shipment quantities, or exact production dates.

A 2.4T module raises aggregate bandwidth beyond 1.6T. It represents another step in the same race to move more data through a limited number of switch ports.

NPO means near-packaged optics. The architecture places optical engines close to the switch or accelerator package, shortening the electrical path between the processing chip and optical conversion.

XPO generally refers to an external or exchangeable packaged-optics approach. Implementations can differ, but the concept seeks tighter integration than traditional front-panel pluggable modules while preserving some serviceability.

Traditional pluggable optics sit in ports at the front of a switch. Technicians can replace them without removing the switch silicon or an integrated optical assembly.

That serviceability has helped pluggable modules dominate data-center networks. Operators can qualify modules independently, replace failures, and mix suppliers within supported configurations.

However, the electrical connection between the switch chip and front-panel module becomes harder to manage as lane speeds rise. Longer traces consume power and weaken signals.

NPO shortens that path. It can reduce electrical losses and support higher bandwidth density, but it also changes cooling, packaging, repair, and supplier responsibilities.

A failure in a pluggable module usually affects one removable device. A failure in a tightly integrated optical assembly can create a more complex service event.

This is the mechanism behind Zhongji Innolight’s roadmap. Faster AI networks are pushing optics physically closer to compute and switching hardware.

The shift expands the company’s opportunity beyond familiar modules. It also places the company closer to semiconductor packaging, switch design, and system-level thermal engineering.

That proximity creates new competitors and partners. Chip companies, contract manufacturers, optical-engine specialists, and system vendors can claim more of the value chain.

It also reduces standardization. Large customers can request custom designs aligned with their switches, accelerators, rack layouts, and cooling systems.

Custom programs can strengthen supplier relationships. Once qualified, a tightly integrated optical design may be difficult to replace quickly.

Yet customization raises development cost and concentration risk. A product designed around one customer’s architecture may have limited value if that customer changes direction.

The promised second-half 2027 production window should therefore be read as a target. Sampling must be followed by qualification, yield improvement, component availability, and final deployment approval.

There is also no guarantee that every customer will move from pluggable optics at the same pace. Many operators value the operational simplicity of removable modules.

The likely market is not a clean replacement cycle. Pluggable, near-packaged, and more integrated optical systems can coexist across different network layers.

Scale-out networks may retain pluggable modules where reach and serviceability dominate. Scale-up systems may adopt closer optical integration when bandwidth density and power become stricter constraints.

This coexistence could support Zhongji Innolight’s broad roadmap. It can keep supplying 800G and 1.6T modules while developing products for tighter integration.

It also complicates capacity planning. Different architectures require different components, packaging methods, test equipment, and workforce skills.

A factory optimized for high-volume pluggable assembly cannot automatically produce every NPO or XPO design. The company must add capabilities without starving its current products.

Competitors face the same problem. The industry must fund tomorrow’s packaging transition while meeting today’s 800G and 1.6T orders.

What the Bullish Forecast Does Not Establish

Strong customer guidance does not remove policy, concentration, pricing, or technology risk.

The first uncertainty is order quality. Zhongji Innolight has not publicly provided a customer-by-customer breakdown or quantified 2027 units in the reported conference-call summary.

Without that information, investors cannot determine how much demand represents binding orders. They also cannot estimate how easily customers can adjust schedules.

The second uncertainty is customer concentration. The largest AI infrastructure companies account for a substantial share of advanced optical demand.

That concentration gives suppliers volume and visibility, but it also gives buyers negotiating power. A design change by one customer can affect an entire production line.

A supplier can report higher shipments while experiencing lower prices. Revenue and profit depend on product mix, component cost, yield, and customer negotiations, not just unit demand.

The third uncertainty is double ordering. Buyers sometimes reserve more capacity than they ultimately need when they fear shortages.

Suppliers may interpret overlapping reservations as final demand. When component availability improves, customers can reduce or reschedule excess commitments.

No public evidence currently establishes that double ordering is driving Zhongji Innolight’s guidance. It remains a risk worth testing against future inventory and delivery data.

The fourth uncertainty is technology selection. NPO and XPO promise better electrical efficiency, but customers have not converged on one universal architecture.

Pluggable modules continue improving. If they remain economical at higher speeds, adoption of tightly integrated alternatives could develop more slowly than supplier roadmaps suggest.

The fifth uncertainty is manufacturing yield. New speeds and packaging methods require production learning before gross output becomes dependable customer supply.

A company can install nominal capacity without achieving the yield needed for profitable shipments. Investors should avoid treating equipment purchases as completed production.

The sixth uncertainty is geographic policy. Advanced optical components have become part of broader debates about telecommunications security and supply-chain dependence.

Cignal AI has examined the potential effect of tighter United States restrictions on Chinese optical equipment. Its policy assessment identifies Zhongji Innolight as especially exposed because of its position in data-center optics.

That analysis describes a policy scenario, not a settled conclusion about every product or customer. Regulatory scope, implementation, and enforcement can differ across telecom and private data-center markets.

Still, customers may diversify suppliers before a formal restriction takes effect. Procurement teams often respond to policy uncertainty through additional qualification and geographic redundancy.

This creates an unusual market structure. Overall optical demand can remain strong while one supplier’s accessible market becomes less certain.

Policy pressure can also benefit competitors with manufacturing footprints that customers consider less exposed. Coherent, Lumentum, Applied Optoelectronics, and contract manufacturers could receive additional qualification opportunities.

However, replacing a major supplier’s volume would not be immediate. Competitors would need components, equipment, trained workers, and customer approval.

That replacement difficulty can preserve Zhongji Innolight’s position in the short term. It can also encourage customers to accelerate alternative sourcing over several years.

The final uncertainty concerns timing. AI infrastructure spending can be strong while individual campuses face power, permitting, cooling, or construction delays.

Optical modules sit late in the deployment chain. A delayed data hall can push delivery even when the underlying computing plan remains intact.

For this reason, company guidance should be compared with actual shipments and customer capital expenditures. One data point cannot prove the entire 2027 cycle.

Three Signals Will Test the 2027 Thesis

The next evidence must come from purchase commitments, production results, and customer qualification rather than another broad demand statement.

The first signal is the conversion of 2027 guidance into disclosed backlog and shipments. Future company updates should clarify whether orders are binding, how far schedules extend, and which product generations drive growth.

A continued increase in delivered 1.6T modules would strengthen Zhongji Innolight’s thesis. It would show that customer plans are passing through qualification and into deployed networks.

Repeated references to guidance without shipment detail would weaken confidence. It would leave open the possibility that customers are reserving capacity without final deployment commitments.

The second signal is component availability and manufacturing yield. The company should demonstrate that capacity additions translate into qualified output rather than unfinished inventory.

Investors should watch inventory, capital expenditure, gross margin, and supplier commentary together. Rising inventory with delayed revenue can indicate bottlenecks or schedule changes.

Stable margins during a rapid ramp would suggest that yields and component costs remain controlled. Falling margins could reflect price pressure, expensive components, or inefficient new lines.

Competitor results provide a useful cross-check. Coherent, Lumentum, Eoptolink, and Applied Optoelectronics should report similar demand if the entire market is expanding at Zhongji Innolight’s suggested pace.

A single supplier can gain share, but broad AI network growth should appear across laser, chip, module, and manufacturing providers. Contradictory commentary would require closer examination of customer mix.

The third signal is qualification of 2.4T, NPO, and XPO products before the second half of 2027. Samples alone do not establish that these architectures will reach mass deployment.

Customers must complete technical validation, finalize system designs, and assign production volumes. Evidence of multiple qualified customers would strengthen the roadmap.

A delay into 2028 would not necessarily invalidate the long-term shift toward integrated optics. It would show that the transition requires more engineering or operational change than the current schedule assumes.

Readers should also distinguish a product announcement from a completed qualification. The most useful disclosures will identify production status, deployment context, and customer acceptance without relying on vague labels.

Zhongji Innolight’s latest message is therefore significant, but narrower than a guaranteed boom. The company says 800G and 1.6T demand will keep growing rapidly during 2027, while future architectures approach production.

The market now needs evidence that suppliers can convert that demand into dependable output. Optical chips, electrical chips, PCBs, packaging yield, and policy exposure all stand between a forecast and a shipped module.

For enterprises planning AI infrastructure, the practical question is not which speed has the most impressive label. It is whether network vendors can deliver qualified capacity when accelerator systems arrive.

Watch the next order disclosures, production metrics, and qualification milestones. Together, they will show whether Zhongji Innolight’s 2027 forecast represents durable deployment demand or an unusually confident reservation cycle.