Zhongji Innolight’s CPO Boom Has a Cash Flow Problem

Zhongji Innolight more than doubled optical transceiver sales in H1 2026, yet its operating cash flow fell 44.08 percent. That conflict makes the company’s latest results more revealing than another upbeat CPO headline. Demand for AI networking hardware is producing exceptional revenue and profit growth, but the cash conversion behind that growth weakened sharply.

The Chinese optical equipment supplier reported the results on August 21, covering the six months ended June 30. Revenue reached RMB 41.78 billion, up 182.49 percent from the prior-year period. Net profit attributable to shareholders climbed 241.70 percent to RMB 13.65 billion, according to the company’s interim figures.

Those numbers place Zhongji Innolight at the center of the spending race among hyperscale cloud operators and AI computing providers. However, the results also expose tension between reported earnings and cash received from customers. That tension matters as the industry moves from 800G modules toward 1.6T products, silicon photonics, and eventually co-packaged optics.

CPO, or co-packaged optics, places optical engines close to a networking chip instead of relying entirely on removable transceivers. Zhongji Innolight remains primarily a high-speed optical module supplier today. Still, investors often treat it as a major beneficiary of the broader transition toward denser optical connections.

Its immediate growth engine is not mass-market CPO deployment. It is the rapid expansion of conventional 800G and newer 1.6T transceivers inside AI data centers. The distinction matters because those products generate today’s revenue, while CPO represents both an opportunity and a potential architectural threat.

The H1 Numbers Show an Extraordinary Production Ramp

Zhongji Innolight converted AI infrastructure spending into a production increase rarely seen in established hardware manufacturing.

The company produced 18.86 million optical transceivers during the first half, up from 9.40 million one year earlier. Sales reached 18.99 million units, compared with 9.05 million units in H1 2025. Production therefore increased 100.6 percent, while unit sales rose 109.8 percent.

Available production capacity expanded from 11.61 million to 22.15 million units. That 90.8 percent increase shows that the company prepared for more than a temporary order spike. It added enough capacity to support a much larger operating base.

The relationship among production, capacity, and sales also deserves attention. Sales slightly exceeded current-period production, suggesting the company fulfilled part of demand using existing inventory. Production utilization was high even after the capacity expansion.

Revenue grew faster than unit volume. This difference indicates that the product mix shifted toward faster and more valuable modules. Zhongji Innolight attributed the performance to accelerating shipments of 800G and 1.6T products for major overseas customers.

An 800G module can transmit an aggregate 800 gigabits per second across its optical lanes. A 1.6T module doubles that aggregate rate, helping large clusters move more data between accelerators and switches.

These connections have become a critical constraint inside AI systems. Training and inference clusters cannot use expensive processors efficiently when networking delays leave those processors waiting for data. Faster optical modules help reduce that bottleneck across rows and buildings.

The company’s optical transceiver gross margin rose to 46.59 percent from 39.96 percent in the prior-year period. That 6.63 percentage-point improvement helps explain why profit expanded much faster than unit sales.

Higher-speed products generally carry better economics during the early stage of a deployment cycle. Customers pay for greater bandwidth, lower power consumption per transmitted bit, and dependable delivery at scale. Suppliers also benefit when demand exceeds qualified manufacturing capacity.

Zhongji Innolight said its major customers increased capital spending on computing infrastructure. It also reported continued volume deliveries to leading cloud service providers and AI computing solution companies.

The company did not publicly identify each customer in the results materials. That omission limits outside analysis of customer concentration and purchasing schedules. It also prevents readers from assigning specific volumes to individual hyperscalers.

The production numbers nevertheless support the broader demand claim. A manufacturer cannot double physical unit sales through accounting treatment alone. It must acquire components, assemble products, test them, obtain customer acceptance, and deliver them.

The scale also changes the interpretation of Zhongji Innolight’s earlier performance. Its 2025 annual report recorded 21.09 million optical module sales for the full year. H1 2026 sales already reached roughly 90 percent of that annual total.

That comparison captures the speed of the current expansion. Zhongji Innolight shipped almost as many modules in six months as it did throughout the previous year. The business has entered a different operating range.

Revenue and margin gains show that this expansion did not depend on selling twice as many low-value units. The company delivered more modules while increasing its share of higher-speed products. That combination drove operating leverage across manufacturing and overhead.

The H1 results therefore validate strong current demand for high-speed optical interconnects. They do not, however, settle whether the company can turn that demand into equally strong cash generation.

AI Demand Is Pressuring the Entire Optical Supply Chain

The results raise the performance bar for every supplier trying to qualify inside large AI networks.

Optical transceivers are not interchangeable commodities at the highest speeds. Customers test signal integrity, thermal behavior, failure rates, component traceability, and compatibility with networking platforms. A supplier must meet those requirements while shipping millions of units.

Zhongji Innolight’s production ramp increases pressure on rivals such as Eoptolink, Coherent, and Lumentum. Each competes across different product categories, geographies, and customer relationships. All must respond to the same demand for faster and more energy-efficient links.

The pressure extends upstream. High-speed modules need optical chips, digital signal processors, printed circuit boards, connectors, and precision packaging. Doubling output increases procurement needs before the supplier necessarily receives payment for the finished product.

That timing difference sits at the center of the cash flow issue. Manufacturers often pay employees and component vendors before customers settle their invoices. Faster growth can therefore consume cash even when each completed sale remains profitable.

The company’s first-quarter performance shows how quickly conditions changed. Zhongji Innolight reported RMB 19.50 billion in revenue and RMB 5.73 billion in attributable profit during Q1.

Its official Q1 filing also showed RMB 3.37 billion in operating cash flow. That figure was 55.58 percent higher than the prior-year quarter.

Subtracting Q1 from the first-half totals produces a striking second-quarter picture. Revenue was approximately RMB 22.28 billion, while attributable profit was about RMB 7.92 billion. Both improved sequentially.

However, first-half operating cash flow was only RMB 1.80 billion. Since Q1 generated RMB 3.37 billion, the reported totals imply negative operating cash flow of roughly RMB 1.57 billion during Q2.

This calculation is an inference from the two reported periods, rather than a separate company disclosure. It shows that cash conversion deteriorated while quarterly profit accelerated.

Several normal working-capital mechanisms can produce such a pattern. A company can build materials for future orders, extend payment terms to important customers, or pay suppliers before collecting receivables. Tax and bonus payment schedules can also shift cash between quarters.

None of these explanations automatically signals a broken business. Rapidly expanding hardware companies frequently experience uneven cash flow. The issue becomes serious when receivables or inventory keep growing faster than sales across several reporting periods.

The scale of the divergence remains difficult to ignore. H1 attributable profit reached RMB 13.65 billion, yet operating activities generated RMB 1.80 billion in net cash. Cash conversion was therefore far below the level implied by accounting earnings.

The comparison with 2025 makes the reversal clearer. Zhongji Innolight generated RMB 10.90 billion in full-year operating cash flow during 2025, up 244.31 percent. That amount closely matched its RMB 10.80 billion attributable profit.

A business that converted profit effectively last year converted much less during the latest half. The change does not erase the growth story. It adds a test that the next financial report must answer.

Competitors face the same underlying challenge. Winning a larger allocation from a hyperscaler can require capacity investments and substantial component purchases. Suppliers with weaker balance sheets may struggle to finance that ramp.

At the same time, customers have incentives to maintain multiple qualified sources. Dependence on one vendor creates supply and pricing risks. Strong results from Zhongji Innolight can therefore encourage cloud operators to increase orders elsewhere, not only consolidate them.

The company’s scale gives it leverage in procurement, manufacturing, and customer qualification. Yet scale also makes every working-capital movement larger. A small change in collection timing can move billions of renminbi across reporting periods.

That is why the cash decline matters beyond accounting. It measures how much of the AI infrastructure boom has already entered the company’s bank accounts.

Zhongji Innolight’s CPO Story Still Depends on Pluggable Modules

The central reversal is that Zhongji Innolight’s CPO valuation rests on booming products that CPO could eventually replace.

Most of the company’s current revenue comes from optical transceivers that plug into the front of network switches. These modules provide flexibility because operators can replace, repair, or upgrade them independently.

CPO uses a different physical arrangement. Optical engines sit beside the switching silicon inside one package or assembly. Shorter electrical connections can reduce power use and signal losses at very high bandwidths.

This architecture becomes more attractive as switch capacity rises. Moving electrical signals across a board grows harder when lane speeds increase. Signal conditioning consumes more power, while heat complicates system design.

That mechanism explains why Zhongji Innolight invests in CPO-related technology. The company has experience in optical design, coupling, packaging, testing, and high-volume production. Those capabilities remain useful even if the product’s physical location changes.

However, CPO also alters who controls the finished system. Switch-chip vendors, foundries, packaging specialists, and optical suppliers must coordinate much more closely. Value can move away from a standalone module vendor toward integrated platform owners.

The practical contest is therefore not Zhongji Innolight against one named rival. It is the promise of sustained high-margin growth against the reality of a changing interconnect architecture.

For now, pluggable modules retain strong advantages. Data center technicians know how to deploy and replace them. Operators can mix qualified vendors, manage failures at the module level, and upgrade links without replacing an entire switch assembly.

CPO creates serviceability questions. A failed optical engine located beside an expensive switch chip can be harder to replace. Thermal management, fiber attachment, manufacturing yield, and field maintenance all require new operational procedures.

Those obstacles have kept CPO from displacing pluggable optics across mainstream deployments. They also leave room for intermediate designs, including near-packaged optics and improved pluggable modules.

Zhongji Innolight’s near-term growth therefore comes from the continued strength of the architecture that CPO aims to improve. Its 800G and 1.6T shipments show that operators still need enormous volumes of removable optical modules.

Silicon photonics strengthens that bridge. The technology integrates optical components using semiconductor manufacturing methods, allowing denser and potentially more efficient designs. It can support both pluggable modules and future co-packaged systems.

The company said its 1.6T silicon-photonics modules had entered a volume ramp. That claim aligns with its broader explanation for faster revenue growth and margin expansion. Independent product-level shipment data remains limited.

Zhongji Innolight is also developing XPO and NPO products. These labels generally describe optical engines placed closer to the switching chip than conventional front-panel modules. They represent possible steps toward tighter integration.

The presence of several architectures is important. Data center operators do not need to switch from pluggable optics to full CPO in one move. Different designs can coexist based on bandwidth, reliability, cost, and maintenance requirements.

That coexistence benefits an experienced optical manufacturer. Zhongji Innolight can supply current modules while preparing packaging and optical-engine capabilities for future systems. It does not need full CPO adoption to grow during 2026.

The longer-term risk concerns value capture. If integrated switch platforms commoditize external optics, unit demand for traditional modules could slow. If CPO deployment remains difficult, suppliers may spend heavily on technologies that produce limited near-term revenue.

Both outcomes make execution more important than labels. Calling Zhongji Innolight a CPO leader does not reveal how much revenue comes from CPO products. The disclosed H1 growth is overwhelmingly an optical transceiver story.

That distinction should shape how readers interpret the company’s performance. The earnings validate the AI networking demand cycle. They do not yet validate a large commercial market for fully co-packaged optics.

Investors should also resist treating every high-speed optical product as CPO. A 1.6T pluggable module can use silicon photonics without being co-packaged with a switch chip. The technologies overlap, but they are not synonyms.

Zhongji Innolight’s strategic position remains attractive because it participates in both phases. It sells the products customers require now and develops technologies that customers might require later. The risk comes from managing the transition without weakening returns.

What the Profit Surge Does Not Prove

Exceptional earnings do not prove that current margins, cash conversion, or customer demand will remain at H1 levels.

The most immediate uncertainty is working capital. Operating cash flow fell to RMB 1.80 billion from roughly RMB 3.22 billion one year earlier. The 44.08 percent decline moved in the opposite direction from revenue and profit.

The published production data confirms that the company doubled output and sales. It does not fully explain when customers must pay or how much material the company secured for future production.

Customer concentration adds another uncertainty. Large cloud and AI companies can place enormous orders, but they also possess purchasing leverage. Their qualification decisions can shift volume among suppliers or pressure prices after supply expands.

Zhongji Innolight reported serving most leading global cloud providers and AI computing solution companies. That breadth is encouraging, although the statement does not disclose revenue shares by customer.

Geographic exposure deserves similar caution. Overseas demand drives much of the high-speed module business. That creates exposure to trade restrictions, tariffs, export controls, currency movements, and customer efforts to diversify supply.

The company has expanded manufacturing outside mainland China, including capacity in Thailand. International production can improve supply resilience, but additional facilities require capital, trained employees, and stable component logistics.

Margins are another source of uncertainty. A 46.59 percent gross margin for optical transceivers reflects favorable demand and product mix. Competitors will work to qualify more capacity while customers seek lower unit costs.

Hardware prices usually decline as products mature. Suppliers must offset that decline through better yields, lower component costs, improved designs, and a shift toward the next speed generation.

The current 1.6T ramp can support margins because qualified supply remains limited. That benefit will narrow if production catches up with demand. The timing depends on component availability and customer qualification cycles.

The interim dividend adds another dimension. Zhongji Innolight proposed RMB 12 for every ten shares, with an estimated aggregate distribution of about RMB 1.40 billion.

The distribution signals confidence, but it also approaches the RMB 1.80 billion generated by operating activities during H1. Dividend capacity ultimately depends on sustained cash generation, not only reported profit.

Research spending is essential as architectures change. The company must support current 800G production, scale 1.6T products, and prepare optical engines for NPO or CPO systems. Each path demands engineering resources and testing.

The results announcement said ongoing product optimization and operating efficiency supported profitability. Readers should treat those statements as management explanations until later reports show whether the improvements persist.

The same caution applies to the CPO narrative. The company possesses relevant manufacturing capabilities, but industry adoption depends on system-level decisions made by switch vendors and hyperscalers.

No single supplier controls serviceability standards, packaging interfaces, thermal design, or deployment schedules. CPO progress requires coordination across the networking supply chain.

The current results can support two ideas at once. Zhongji Innolight is benefiting substantially from AI infrastructure spending, and its cash conversion weakened at a critical moment.

Neither fact cancels the other. The first describes demand and profitability. The second identifies the financial pressure created while serving that demand.

The bearish interpretation would treat the cash decline as evidence that earnings quality has already deteriorated. That conclusion is premature after one half, especially during an extreme production ramp.

The overly bullish interpretation would dismiss cash flow as harmless timing. That conclusion is also premature because Q2 appears to have consumed operating cash while delivering record profit.

A stronger judgment requires the next balance-sheet and cash-flow disclosures. If collections normalize, the H1 divergence will look temporary. If the gap widens, growth will require closer scrutiny.

Three Signals Will Decide Whether the CPO Boom Is Sustainable

The next report must show that Zhongji Innolight can preserve growth while restoring cash conversion and preparing for architectural change.

The first signal is operating cash flow. It should be compared with attributable profit, rather than viewed only as a year-over-year growth rate.

A rebound would support the explanation that H1 reflected temporary payment and procurement timing. Another weak period would increase concern about receivables, inventory, supplier payments, or customer terms.

Quarterly inference will remain useful. The latest results summary indicates Q2 profit rose approximately 38 percent from Q1. Future readers should compare that earnings momentum with the cash generated during the same period.

The second signal is the 1.6T product mix. Zhongji Innolight needs rising high-speed shipments to offset normal price pressure on older modules.

Sustained gross margins near the H1 level would suggest that product mix, manufacturing yields, and supply conditions remain favorable. A sharp margin decline would indicate that competition or customer pricing pressure is arriving faster.

Volume alone will not answer this question. Unit sales can increase while economics weaken. Revenue per unit, gross margin, and inventory movements provide a more complete picture.

The third signal is evidence of commercial progress beyond conventional pluggable optics. Readers should look for customer qualifications, production milestones, or disclosed revenue linked to optical engines, NPO, or full CPO systems.

Product demonstrations are useful technical indicators, but they are not equivalent to volume orders. A credible transition requires qualified deployments and repeatable manufacturing.

Progress would strengthen the argument that Zhongji Innolight can carry its position into the next networking architecture. Delays would leave the company more dependent on the current pluggable-module cycle.

These three signals belong together. Strong cash generation funds research and capacity. A favorable 1.6T mix protects returns during expansion. Commercial optical-engine progress prepares the company for what follows.

Zhongji Innolight’s H1 performance shows why optical networking has become central to AI infrastructure. Processor demand receives more attention, but accelerators need high-bandwidth connections to operate as useful clusters.

The company sold 18.99 million optical modules in six months and lifted attributable profit by 241.70 percent. Those achievements make the results difficult to dismiss.

Yet the operating cash decline prevents a simple victory narrative. The company generated less cash from operations while producing, selling, and earning far more.

That is the real financial test behind the CPO story. Can Zhongji Innolight finance an extraordinary manufacturing ramp, preserve high margins, collect cash efficiently, and navigate the move toward integrated optics?

The answer will not come from another CPO demonstration or a larger headline profit figure. It will come from cash conversion, 1.6T economics, and evidence that next-generation optical designs are becoming commercial products.

Readers following AI infrastructure should track all three, not only shipment growth. If cash rebounds while advanced products scale, the H1 divergence will look like the cost of rapid expansion. If cash remains weak, the market will need to reconsider how much of Zhongji Innolight’s reported boom is reaching the business in spendable form.