Zhongji Innolight Says Margins Can Hold, but Supply Is the Real Test

Zhongji Innolight says it can keep gross margins stable despite material shortages, cautious supplier quotes, and intense demand for AI networking equipment. The company made that case during a July 28 investor call summarized by 36Kr and distributed through the RSSHub 36Kr feed.

Management described an optical-module market where both order volume and pricing remain favorable. Yet the strongest demand signals are colliding with constrained supplies of critical components. That conflict makes the margin forecast more consequential than a routine statement of confidence.

The company expects higher-margin 2.4-terabit and near-packaged optics products to support future profitability as shipments increase from 2027. However, those products still face development, qualification, manufacturing, and adoption tests. For now, Zhongji Innolight must protect margins while delivering more 800G and 1.6T modules into a tight supply chain.

This is a contest between product mix and input pressure. Faster modules can command more value, but constrained components can raise costs and limit shipments. The next several quarters will show which force carries more weight.

What Zhongji Innolight Actually Said

The important change is not stronger demand alone. Zhongji Innolight is telling investors that shortages should not erase the profitability created by that demand.

According to the July 28 account, customers are requesting healthy volumes, while market pricing remains favorable. Industry demand is unusually strong, and material availability is creating pressure across the supply base. Vendors are therefore taking a cautious approach when quoting future orders.

The update follows an earlier investor call held on July 12. In that meeting, management reportedly said second-quarter gross margins remained stable despite higher material costs. It also expected component availability to improve during the second half as suppliers expanded production and additional sources entered the market.

The company said it had prepared inventory through long-term agreements and advance payments. Some agreements lock in prices, while others secure supply without fixing the final cost. This distinction matters because supply protection does not automatically guarantee margin protection.

The earlier call also indicated that existing orders covered all of 2026 and part of 2027. Some customers had reportedly supplied quarterly or monthly demand guidance for next year. Management said demand had not fallen, while certain 800G requirements had increased.

These statements remain company guidance rather than independently audited forecasts. Purchase plans can change, especially when customers adjust data-center schedules or redesign network architectures. An order pipeline also does not guarantee that every unit will ship, pass acceptance, or produce the expected margin.

Still, the message is consistent with the company’s recent operating history. Zhongji Innolight reported that optical-transceiver gross margin reached 39.96% in the first half of 2025, up from 33.83% one year earlier. Revenue from that business reached RMB 14.43 billion during the period.

The company attributed that improvement to a larger share of higher-speed products and continuing manufacturing efficiencies. Its interim report said rising customer investment in computing infrastructure increased demand for 800G modules. It also accelerated the move toward 1.6T and faster connections.

The July update extends that argument. Zhongji Innolight is effectively saying the same mix benefit can offset near-term component inflation. It then expects 2.4T and NPO products to provide another lift after 2027.

That is a more demanding claim than saying sales will grow. Revenue can increase while margins fall if input costs, expedited freight, lower yields, or customer concessions consume the added value. The company must show that its supply agreements and manufacturing scale can prevent that outcome.

The RSSHub 36Kr item captured the headline, but the underlying issue is execution. Stable pricing helps only when factories have the right components, yields remain high, and customers accept deliveries on schedule.

Why AI Networks Are Keeping Optical Demand High

AI clusters are forcing data-center operators to buy more bandwidth per accelerator, placing optical connectivity on the critical path of computing deployments.

An optical transceiver converts electrical signals into light and back again, allowing switches and servers to exchange data through fiber. As AI clusters grow, these modules must move larger volumes of data between accelerators with less delay and lower power consumption.

The transition from 400G to 800G and 1.6T increases the amount of data each module can carry. It also raises the technical demands placed on lasers, digital signal processors, packaging, thermal control, and production testing.

Large language models do not run on isolated processors. Training and inference workloads divide calculations among many accelerators, which exchange intermediate results throughout a job. A slow or unreliable network can leave expensive computing hardware waiting for data.

Scale-out connections link servers and racks across a data center. Scale-up connections create tighter links among accelerators working as one system. Zhongji Innolight expects optical technology to capture a larger role in both environments as electrical connections encounter distance, power, and signal-integrity limits.

That expectation explains the attention given to future scale-up optical solutions. Shorter links have historically relied more heavily on copper because it is familiar and economical. However, copper becomes harder to use as bandwidth rises and systems pack more accelerators into each domain.

The change creates an opening for module vendors, but it also raises the performance bar. A scale-up link must deliver low latency and high reliability because communication failures can disrupt tightly coordinated computing work. Operators also care about power consumption because networking competes with processors and cooling systems for a limited energy budget.

Zhongji Innolight’s existing 800G and 1.6T business gives it manufacturing experience relevant to this transition. Its 2025 interim filing listed 200G, 400G, 800G, and 1.6T products for cloud data-center customers. The company said it produced 9.4 million optical modules and sold 9.05 million during that half-year period.

Annual results suggest the mix continued moving upward. The company’s optical-transceiver gross margin reportedly rose from 34.65% in 2024 to 42.61% in 2025. Optical-module sales increased from 14.59 million to 21.09 million units.

The annual-results analysis linked that margin expansion to a larger contribution from 800G and 1.6T products. Optical-transceiver revenue rose faster than the associated cost of sales, supporting the mix argument presented by management.

Those figures provide useful historical evidence, but they do not settle the 2026 question. The economics of an early product ramp often differ from those of a mature one. Initial pricing can be stronger, while later competition, customer negotiations, and improved industry supply can reduce average selling prices.

Demand also depends on a concentrated group of large buyers. In 2025, overseas sales reportedly represented 90.58% of company revenue. The five largest customers accounted for 75.98% of annual sales, and the largest customer contributed 24.06%.

Customer concentration can help a supplier plan production against clear forecasts. It can also magnify the consequences of one buyer delaying a cluster, changing its network design, or qualifying another vendor. Strong aggregate AI spending does not remove that account-level risk.

This is why the margin outlook matters across the optical industry. Peers such as Eoptolink, Coherent, Lumentum, and Applied Optoelectronics face versions of the same challenge. Each must secure components, qualify faster products, and preserve enough pricing power to fund the next development cycle.

The pressure extends upstream. Optical-module makers need lasers, continuous-wave light sources, driver chips, transimpedance amplifiers, digital signal processors, connectors, and specialized packaging. Capacity in one category cannot compensate for a missing part elsewhere.

High demand therefore creates an unusual operating problem. Suppliers have attractive orders, but their revenue depends on the least available component in each design. The resulting bottleneck can delay complete modules even when most materials are ready.

The Margin Battle Is Product Mix Versus Material Costs

Zhongji Innolight’s margin defense rests on higher-value products, better yields, and purchasing discipline overcoming the cost of scarce components.

Product mix is the first mechanism. Faster modules generally contain more advanced components and solve a more valuable customer problem. During a transition, qualified suppliers can also face less competition because only a limited group can deliver at scale.

Zhongji Innolight previously said 1.6T demand would increase as more customers adopted the format. It also expected silicon photonics to represent a growing share of 800G and 1.6T shipments.

Silicon photonics integrates optical functions onto silicon-based structures, reducing reliance on some discrete optical components. The company has said this approach can lower bill-of-materials costs while increasing integration. Those benefits depend on design choices, manufacturing yields, and customer requirements.

The company does not use one technical route for every order. Some customers require externally modulated laser designs, while others accept silicon-photonics solutions. Maintaining both options can broaden the customer base, but it also complicates procurement and production planning.

Yield improvement forms the second margin mechanism. Yield measures how many manufactured units meet specification. A small change can materially affect profit when modules use costly chips and require extensive testing.

Higher yields spread material and factory costs across more saleable units. They also reduce rework, scrap, and delayed delivery. During a shortage, yield becomes even more important because a defective unit consumes components that may be difficult to replace.

Scale is the third mechanism. Large purchase commitments can improve access to constrained materials and justify supplier expansion. Zhongji Innolight says it uses advance payments and longer agreements to protect availability.

Yet long commitments introduce risk. A contract can secure components at an unfavorable cost if market prices later decline. It can also leave a company with excess inventory if customers change their schedules or migrate to another product generation.

That danger is visible on the balance sheet. Zhongji Innolight’s optical-module inventory reportedly reached 5.21 million units at the end of 2025, an increase of 105.12% from the prior year. Total inventory had a book balance of RMB 12.98 billion, with RMB 298 million reserved for impairment.

Management linked the inventory increase to stronger customer orders and additional preparation. That explanation is plausible during rapid growth. However, investors should distinguish inventory built for scheduled production from inventory that cannot move because a required part or customer approval is missing.

Technology cycles make the distinction important. An 800G component remains valuable while customers continue ordering 800G designs. Its value can decline if a buyer accelerates its transition to 1.6T or adopts a different architecture.

The fourth mechanism is pricing. Management described current prices as healthy and suggested suppliers are cautious about future quotes. Tight delivery capacity gives vendors a reason to resist aggressive reductions.

However, major cloud customers have substantial negotiating leverage. They qualify multiple suppliers when possible and seek annual price reductions as products mature. They can also redesign systems to use a different module type or technical approach.

Stable gross margin does not require every product price to remain unchanged. Manufacturing efficiency and a richer mix can offset normal price declines. The problem arises when price erosion and material inflation occur simultaneously.

Currency movements add another variable. Zhongji Innolight earns most of its revenue overseas and purchases important components from international suppliers. Its 2025 exchange loss reportedly reached RMB 318 million, compared with an exchange gain of RMB 123 million in 2024.

Gross margin excludes many financing effects, but currency movements can still influence input costs and contract economics. The company’s annual sensitivity analysis indicated that a 4% currency move could change pretax profit by about RMB 645 million, assuming other factors remained constant.

The July 28 statement therefore represents a combined operational forecast. Zhongji Innolight is betting that pricing, product mix, supply agreements, and yield gains will collectively absorb shortages. None of those levers works independently.

For technology teams tracking infrastructure suppliers, the practical lesson is to preserve source material and distinguish management guidance from reported results. A searchable technical knowledge base can connect call summaries with later filings without treating each claim as settled fact.

2.4T and NPO Offer More Margin, but Not Yet

The next product cycle offers a credible path to higher profitability, but Zhongji Innolight still has to turn engineering programs into qualified volume shipments.

A 2.4T optical module carries up to 2.4 terabits per second through one interface. That capacity makes it a potential successor to 1.6T products in systems that require denser and faster connections.

Zhongji Innolight said during its earlier July call that 2.4T products remained in development and customer testing. It expected initial shipments during the second half of 2027, followed by more meaningful volume in 2028.

That schedule places an important limit on the July 28 margin thesis. The products expected to lift future profitability will not solve every cost issue during 2026. The current margin defense still depends primarily on 800G, 1.6T, silicon photonics, and procurement execution.

NPO, or near-packaged optics, moves optical components closer to the switching chip while keeping them more serviceable than fully integrated co-packaged optics. Shorter electrical paths can reduce signal loss and power consumption at very high bandwidth.

The design attempts to occupy a middle ground. Conventional pluggable modules sit at the edge of a switch and are easy to replace. Co-packaged optics place optical engines beside the switching silicon, improving electrical reach but making cooling, manufacturing, and servicing more complex.

NPO brings optics near the switch chip without necessarily combining everything in one package. In principle, that can preserve some modularity while improving performance. In practice, the industry still needs agreement around packaging, connectors, thermal designs, testing, reliability, and field replacement.

Zhongji Innolight reportedly saw customer interest in NPO accelerate from March 2026. Some customers had supplied demand expectations for 2027 and 2028, while others had started internal projects. The company said it was developing customized products against those requirements.

Customer projects are meaningful early signals, but they are not equivalent to production orders. A data-center operator can evaluate several architectures before selecting one. Qualification can also expose reliability or integration issues that change the schedule.

The route to higher margins has several gates. First, a product must meet bandwidth, latency, power, and error-rate targets. Second, it must work with the customer’s switch, accelerator, fiber, and software environment.

Third, the supplier must manufacture it at acceptable yield. Fourth, the customer must deploy the associated system at scale. Finally, the commercial terms must preserve enough value for the vendor after normal price negotiations.

This sequence explains why higher bandwidth does not guarantee higher margins. More capable products can also contain more expensive materials, require additional testing, and suffer lower early yields. Their economics improve only when price and production efficiency compensate for those costs.

Competition will intensify before 2.4T reaches volume. Established optical vendors are developing faster pluggable modules, silicon-photonics platforms, NPO designs, and co-packaged systems. Switch and accelerator suppliers can also influence which architecture becomes standard.

The main contest is therefore not Zhongji Innolight against one named rival. It is the company’s margin promise against the operational reality of a technology transition. Competitors matter because they affect pricing and qualification, but execution remains the decisive variable.

One risk is that traditional pluggable modules remain effective for longer than expected. That would delay NPO adoption while extending demand for mature products. The outcome could support near-term volume but postpone the premium product mix anticipated for 2027.

The opposite risk is a faster architectural transition. If major customers adopt co-packaged optics or another integrated design, vendors need the right technology and manufacturing partnerships. Zhongji Innolight’s annual reporting acknowledges that an incorrect research direction can expose products to substitution risk.

Another uncertainty involves 3.2T. Management said that product was also under development and expected wider industry maturity around 2028. Development timelines can overlap, raising the possibility that customers compare multiple generations before committing large volumes.

That does not make the 2.4T opportunity artificial. Network bandwidth must continue rising if accelerator density and model workloads keep increasing. It does mean that the financial result depends on timing, architecture, and supplier position, not bandwidth alone.

The July statement uses future products to support a margin-growth narrative. Investors should treat them as staged options rather than booked profit. Each qualification, production ramp, and customer deployment will provide a better measure of their value.

What the Margin Forecast Does Not Prove

Management confidence does not prove that shortages will ease, orders will convert without delay, or future products will carry premium margins.

The first uncertainty is component supply. Zhongji Innolight said shortages were pressuring availability and making suppliers cautious about quotes. That condition can support pricing for finished modules, but it can also raise the cost of lasers and electronic chips.

Supply constraints are rarely uniform. One component may improve while another becomes the limiting factor. Companies can qualify additional suppliers, but validation takes time because optical modules must meet strict performance and reliability requirements.

The second uncertainty is customer concentration. A small number of cloud and technology companies account for much of the industry’s demand. Their spending plans are large, but individual network decisions can shift orders among vendors and product formats.

The third uncertainty is inventory quality. More inventory can protect production during shortages, yet it also ties up capital. The key question is whether that inventory matches firm customer schedules and current technical designs.

The fourth uncertainty is pricing after supply catches up. Tight capacity supports cautious quotes today. New factories, better yields, and more qualified suppliers can change that balance.

Customers often expect lower prices as a product matures. Zhongji Innolight must introduce faster modules and improve manufacturing efficiency quickly enough to stay ahead of those reductions.

The fifth uncertainty is the difference between companywide and product-level margin. A higher-margin new product can lift the average even if older products face pressure. Conversely, a difficult ramp can reduce the average despite strong demand.

Investors should therefore seek a margin bridge. That means separating the effects of product mix, unit pricing, material costs, yields, freight, and currency. A single gross-margin figure cannot show which force produced the result.

The 2025 numbers demonstrate that a favorable mix can lift profitability. Optical-transceiver gross margin reportedly improved by 7.96 percentage points from 2024. However, repeating such a gain becomes harder after the base has already moved higher.

The sixth uncertainty is deployment timing. Customers can give suppliers detailed demand guidance before final installation schedules are fixed. Data-center construction, power availability, cooling capacity, accelerator supply, and network validation can all affect timing.

Delays do not necessarily eliminate demand. They can push revenue into a later quarter and create temporary inventory growth. That distinction matters when judging whether a margin miss reflects weaker economics or a scheduling mismatch.

The original 36Kr newsflash offers a concise summary, not a full transcript with quantified guidance. The RSSHub 36Kr distribution preserves the event lead, but it cannot replace an exchange filing or audited report. Readers should avoid assigning unsupported precision to the company’s statement.

There is also no verified figure in the July 28 summary for the expected gross margin of 2.4T or NPO products. Calling them higher-margin expresses management’s anticipated product economics. It does not establish the eventual margin after yields, competition, and customer pricing.

A cautious reading still recognizes the positive evidence. Demand is reportedly strong, prices have not collapsed, and the company has a recent record of margin expansion. Its products serve a part of AI infrastructure where bandwidth requirements continue to rise.

The cautious conclusion is narrower than the bullish one. Zhongji Innolight has a plausible route to stable margins, but it has not removed the supply, concentration, inventory, and architecture risks surrounding that route.

Three Signals to Watch Next

The margin thesis now depends on three observable signals: reported profitability, supply conversion, and customer validation of next-generation products.

The first signal is the next reported gross-margin figure. Investors should compare it with the company’s recent results and examine whether the explanation emphasizes mix, pricing, yields, or material costs.

A stable figure would support management’s claim that product mix and operating improvements can absorb supply pressure. A decline would not automatically invalidate the long-term thesis, but it would reveal that shortages or pricing had greater influence than expected.

The most useful disclosure would separate current products from early ramps. Growth in 1.6T shipments should improve the mix only if production yields and customer pricing remain favorable. Reported revenue growth without comparable margin performance would indicate weaker conversion.

The second signal is whether supply agreements translate into completed deliveries. Zhongji Innolight expects shipment volume to rise sequentially as component availability improves. Investors should compare that expectation with revenue, inventory, receivables, and operating cash flow.

Inventory that rises alongside shipments can be normal during expansion. Inventory that outpaces deliveries for several quarters deserves closer examination, especially when products are changing quickly.

Supplier diversification is another part of this signal. Additional approved sources would reduce dependence on constrained components and strengthen purchasing leverage. Yet simply naming more suppliers is insufficient if customers have not qualified their parts.

The third signal is formal customer progress for 2.4T and NPO. Useful milestones include completed testing, design wins, production orders, and a confirmed shipping schedule.

Samples and development agreements strengthen the technical case but provide limited evidence about volume economics. A customer qualification followed by repeat orders would offer stronger support for the margin-growth forecast.

The order of these signals matters. Near-term reported margin tests the current defense. Delivery and inventory data test supply execution. Product qualifications test the 2027 growth argument.

Competitor disclosures can provide additional context, but they should remain secondary. Similar comments from Coherent, Lumentum, Eoptolink, or Applied Optoelectronics would support an industrywide demand and supply picture. Divergent comments might reveal differences in customer mix or component access.

Readers should also watch the balance between 800G and 1.6T. Continued 800G growth can extend a profitable product cycle, while faster 1.6T adoption can raise value per connection. Either outcome can help if Zhongji Innolight manages price and cost effectively.

The harder scenario combines weaker 800G demand, a delayed 1.6T ramp, and persistent component inflation. That would place both volume and margin under pressure before 2.4T or NPO contribute meaningfully.

The favorable scenario looks different. Component availability improves, 1.6T volume rises at acceptable yields, and customers convert NPO development work into production commitments. That sequence would support both stable near-term margins and higher profitability after 2027.

For now, Zhongji Innolight has presented a coherent mechanism rather than a guaranteed result. Strong demand creates pricing support, advanced products improve mix, and scale helps secure materials. The next filings must show that those parts are working together.

Anyone following the RSSHub 36Kr headline should return to the operating evidence as it arrives. Watch gross margin first, delivery conversion second, and next-generation qualifications third. Those measurements will reveal whether management’s confidence reflects durable economics or the temporary leverage of a constrained market.