Innolight and Eoptolink Lead Technology News as AI Optics Growth Spreads Across the Supply Chain

Innolight and Eoptolink closed the first half of 2026 with combined net income above RMB 210 billion, despite persistent doubts about AI infrastructure returns. This technology news matters because their growth no longer looks like an isolated windfall for two module vendors. Revenue, profit, and capacity expansion are rising across optical chips, passive components, packaging, and competing transceiver suppliers.

The reporting cycle became complete on August 27, when Huagong Tech published its half-year results. Eoptolink had reported on August 24, while Innolight filed its report on August 22. A September 1 market review then assembled those disclosures into a broader picture of the Chinese optical communications supply chain.

The immediate contest is not simply Innolight versus Eoptolink. The more important test pits concentrated hyperscaler demand against the industry’s attempt to build durable, diversified growth. Amazon, Google, Meta, and Microsoft remain the dominant buyers behind this expansion, giving a small customer group unusual influence over orders and inventories.

The Technology News Behind the Record Half-Year

The headline changed from rising AI expectations to verified earnings across several layers of the optical supply chain.

Innolight, formally Zhongji Innolight, reported first-half revenue of RMB 41.78 billion. That represented year-over-year growth of 182.49 percent. Net income attributable to shareholders reached RMB 13.65 billion, up 241.7 percent, according to its half-year filing.

Those figures put Innolight firmly ahead on absolute revenue and profit. They also show how quickly high-speed products have changed its financial profile. The company’s optical module capacity rose 90.8 percent, while production increased 100.6 percent and sales climbed 109.8 percent.

Eoptolink reported revenue of RMB 20.91 billion for the same period, up 100.34 percent. Net income reached RMB 7.53 billion, an increase of 90.98 percent. Its second-quarter profit also rose sharply from the first quarter, easing concerns created by an earlier sequential decline.

Together, the two companies earned approximately RMB 21.18 billion during the first six months. Their combined results established the “two leaders” framing used in the original market coverage. The distinction reflects their scale, not an assertion that every competitor is losing ground.

The market’s third closely watched supplier, TFC Optical Communication, remained highly profitable. It generated first-half revenue of RMB 2.83 billion, up 15.15 percent. Net income increased 33.92 percent to RMB 1.20 billion, according to its interim filing.

These companies occupy different positions. Innolight and Eoptolink sell complete high-speed transceivers, while TFC supplies optical components and manufacturing services. Their simultaneous growth therefore carries more information than a single vendor’s earnings beat.

An optical transceiver converts electrical data into light, sends it through fiber, and converts the signal back at the destination. AI clusters need these modules because thousands of accelerators must exchange data with limited delay. Faster processors create little value when network links cannot keep them supplied with data.

The industry’s current transition centers on 800G and 1.6T products. Those labels describe each module’s aggregate transmission capacity. A 1.6T module can carry twice the nominal bandwidth of an 800G module, although system performance also depends on switches, fiber, software, and topology.

The strongest demand still comes from data-center networking rather than traditional telecommunications. Innolight said AI data centers have made datacom its central growth engine. Eoptolink similarly connected its performance to high-speed products used in AI clusters and cloud infrastructure.

That is the first major conclusion from the reporting cycle. AI optics have moved beyond forecasts and product demonstrations. The demand is now visible in revenue, margins, shipments, inventories, and factory investment.

Hyperscaler Spending Is Pressuring Every Supplier to Expand

The supply chain must add capacity before it knows how durable today’s concentrated demand will be.

Amazon, Google, Meta, and Microsoft account for more than half of current AI-related optical transceiver spending, according to LightCounting. Their capital programs accelerated demand early in 2026 and pushed suppliers to raise production plans.

LightCounting estimated that optical products used in AI clusters generated about $16.5 billion in 2025. It projected that figure would reach $26 billion during 2026. The research firm later raised its broader Ethernet transceiver growth forecast to 73 percent for the year.

The same research also warned against treating demand as unlimited. Its market outlook tied future orders closely to hyperscaler capital expenditure. A sharp change in spending would travel quickly through module, chip, component, and equipment suppliers.

That concentration creates a demanding operating model. Vendors must secure lasers, digital signal processors, optical engines, connectors, substrates, and manufacturing tools months before completed modules reach customers. Any missing component can delay an otherwise finished product.

Production capacity also differs from usable capacity. High-speed modules require precise assembly, testing, thermal control, and acceptable yields. A new manufacturing line does not contribute its advertised output until those processes stabilize.

Innolight’s capacity, production, and sales growth indicates that it converted much of its expansion into shipments. Its optical transceiver gross margin reached 46.59 percent, up 6.63 percentage points. That improvement suggests the benefits of product mix and scale outweighed the costs of rapid expansion during the half.

Eoptolink reported a 48.46 percent gross margin for optical interconnection products. That was slightly above Innolight’s module margin, although direct comparisons require care. The companies use different reporting categories and may serve different product mixes.

The pressure extends beyond the two leaders. Accelink generated first-half revenue of RMB 6.61 billion, up 26.07 percent. Net income rose 56.34 percent to RMB 582 million, based on its company filing.

Accelink’s overseas revenue increased 73.79 percent, considerably faster than its domestic business. Its transmission product revenue rose 50.51 percent. These figures support the view that data-center interconnection and international demand are lifting suppliers outside the leading pair.

Huagong Tech reported total revenue of RMB 7.82 billion, up 2.53 percent, and net income of RMB 1.19 billion. Profit increased 30.17 percent. Its AI high-speed optical interconnection operation grew faster than the company’s other businesses, and its 800G and 1.6T products entered scaled global delivery.

This broader growth forces suppliers to choose between two risks. Expanding too slowly can surrender customer qualification slots and market share. Expanding too aggressively can leave factories underused if hyperscalers pause purchases or redesign their networks.

The timing also favors early movers. Customers test new modules for compatibility, reliability, temperature control, and power consumption before approving volume shipments. Vendors that complete qualification can secure orders before later entrants finish the same process.

Once approved, however, suppliers still face pricing negotiations. Large cloud buyers have purchasing leverage because their orders are enormous and concentrated. Higher unit volumes do not automatically preserve margins when customers demand lower costs per transmitted bit.

This tension explains why capital spending and order visibility deserve equal attention. Reported revenue describes completed shipments. Capital expenditure reflects management’s expectations about future orders, but it also increases depreciation and execution risk.

Innolight and Eoptolink Lead, but the Mechanism Runs Through 1.6T

The growth mechanism is a simultaneous increase in port count, transmission speed, and optical content per AI cluster.

Traditional cloud networks mainly move traffic between servers, storage, and users. AI training adds intense accelerator-to-accelerator communication. Large models distribute calculations across thousands of processors, which must repeatedly exchange partial results.

This creates scale-out traffic between racks and scale-up traffic within tightly connected computing domains. Both architectures demand more bandwidth, lower latency, and better energy efficiency. Optical links gain importance as electrical connections become harder to extend over distance at higher speeds.

The transition from 800G to 1.6T is therefore more than a routine product refresh. It allows switches and accelerators to exchange more data through each optical port. It also raises technical requirements for lasers, modulation, digital signal processing, packaging, and thermal management.

Cignal AI reported that first-quarter datacom optical component revenue reached a record $7.7 billion. That figure more than doubled from the previous year. The firm also said Innolight led both datacom revenue and high-speed module shipments by a wide margin in its quarterly findings.

The same quarter marked a sharp ramp for 1.6-terabit Ethernet modules. That independent market evidence helps connect the Chinese companies’ results with global deployments. It reduces the likelihood that their reported growth came only from domestic inventory accumulation.

LightCounting expected 800G shipments to more than double during 2026. It also expected 1.6T shipments to reach tens of millions of ports after a limited 2025 base. Sales of 1.6T chipsets were forecast to exceed $2 billion.

This transition helps explain why 800G and 1.6T can grow together. Customers do not replace every network layer simultaneously. New accelerator clusters may adopt 1.6T links while existing facilities continue adding 800G capacity.

That overlap benefits suppliers with broad product portfolios. Innolight sells products ranging from lower-speed modules through 1.6T systems. Eoptolink offers 400G, 800G, and 1.6T products based on several optical architectures.

Silicon photonics is one such architecture. It integrates optical functions onto silicon-based structures, improving manufacturing density and supporting more advanced packaging. Thin-film lithium niobate offers another route for high-speed modulation with favorable optical properties.

Linear pluggable optics, commonly called LPO, removes a full digital signal processor from the module. The design can reduce power and latency, but it demands better signal quality from surrounding equipment. Near-packaged optics moves optical components closer to the switch chip.

Co-packaged optics goes further by placing optical engines beside the switching silicon within one assembly. It promises lower electrical reach and reduced power at very high bandwidth. It also creates challenges involving repairability, thermal design, manufacturing yield, and system integration.

Innolight said it was developing XPO and near-packaged products while refining next-generation optical interconnections. Eoptolink reported products using silicon photonics and thin-film lithium niobate across 400G, 800G, and 1.6T categories.

The existence of several architectures matters. No supplier can assume that today’s dominant module design will remain unchanged throughout the next upgrade cycle. The leading companies must scale current products while funding alternatives that might eventually reduce demand for traditional pluggable modules.

TrendForce expects the global AI optical transceiver market to reach $26 billion in 2026. It describes 2026 and 2027 as crucial years for obtaining 1.6T customer qualifications. Its industry forecast also identifies component shortages as the main capacity constraint.

That constraint turns upstream suppliers into part of the central story. A transceiver maker cannot satisfy demand without enough lasers, modulators, photodetectors, optical connectors, and testing capacity. The product transition raises specifications across each category.

For technology news readers, the mechanism is straightforward. More AI processors require more network connections. Faster connections increase optical content and manufacturing difficulty. Those requirements distribute spending from cloud companies through module vendors and into the component base.

The Boom Has Reached Chips, Components, and Packaging

The half-year reports show demand spreading upstream, but growth rates vary sharply by product exposure and production readiness.

TFC Optical Communication provides precision components, optical engines, packaging services, and manufacturing support. Its 33.92 percent profit growth exceeded its 15.15 percent revenue growth. That gap indicates favorable product mix and operating leverage, although it does not guarantee the same pattern will continue.

Source Photonics chip supplier Yuanjie Semiconductor offered a more dramatic example. Its first-half net income reached RMB 607 million, up 1,212.2 percent. Data-center products accounted for 65.39 percent of revenue, making AI-related demand the central earnings driver.

Yuanjie develops and manufactures laser chips used in optical modules. These chips generate the light that carries data through fiber. Higher transmission speeds impose tighter requirements on output power, modulation, reliability, and manufacturing consistency.

The company also increased purchases of wafer-processing and chip-production equipment during the half. Its expansion reflects both current demand and expectations for wider 1.6T deployment. Those investments will matter only if production yields and customer approvals follow.

Shijia Photons reported first-half net income of RMB 315 million, up 45.3 percent. Its portfolio includes optical chips, arrayed waveguide gratings, components, and cable materials. That mixture gives it exposure to several parts of the network buildout.

Accelink presents a different model because it spans devices and completed modules. Its data and access products generated RMB 4.31 billion, while transmission products contributed RMB 2.28 billion. The company therefore benefits from data-center demand without relying entirely on one module category.

Its cash-flow figures also show the burden of expansion. Accelink reported operating cash flow of negative RMB 1.24 billion, compared with negative RMB 107 million one year earlier. Working capital can absorb cash when inventories and receivables grow faster than collections.

The leading trio provides another warning. Innolight, Eoptolink, and TFC generated combined net income of RMB 22.38 billion. Yet their combined operating cash flow was only RMB 4.42 billion, according to the aggregated half-year review.

The difference does not invalidate reported profit. It shows that fast growth requires cash for inventory, production, and customer credit. Investors and suppliers should track whether those working-capital investments turn into collections during subsequent quarters.

Inventory can represent either preparation or risk. Suppliers need buffers when key lasers and other components remain constrained. However, excess stock becomes costly if customers delay deployments, switch architectures, or revise orders.

Customer concentration increases that uncertainty. Overseas revenue represented 97.85 percent of sales at the most internationally exposed company within the leading group. A concentrated customer base can drive extraordinary growth, but it can also magnify negotiation, geopolitical, and scheduling risks.

The upstream results still support the “full supply chain” thesis. Module competitors grew, component specialists expanded, and chipmakers reported stronger demand. The pattern is broader than a temporary increase in one company’s market share.

It is not fully uniform. TFC’s revenue growth remained below the two module leaders’ rates. Huagong Tech’s consolidated revenue increased only 2.53 percent because its other operations diluted the optical business’s faster expansion.

Telecommunications demand also follows a different cycle. Carrier networks depend on operator budgets, broadband upgrades, and regional deployments. Those programs do not necessarily accelerate alongside hyperscaler AI spending.

Calling the entire optical industry equally strong would therefore overstate the evidence. The strongest results cluster around high-speed datacom products and related components. Vendors centered on slower telecom products face a less direct benefit.

What the Numbers Do Not Settle

The earnings confirm exceptional demand, but they do not prove that every factory expansion or valuation assumption will survive the next spending cycle.

The first uncertainty is the return on hyperscaler AI investment. Cloud companies are committing enormous capital to accelerators, data centers, power, and networking. Their optical suppliers benefit before the final economic return becomes clear.

If AI services generate sufficient revenue, customers can keep expanding clusters. If returns disappoint, capital budgets can change faster than suppliers can reduce capacity. Optical demand would then face inventory corrections and pricing pressure.

LightCounting has already described the tension between strong supply-chain conditions and investor concern about AI returns. Its optimistic forecast still assumes continued spending by a small number of cloud companies. The firm warns that a sharper capital-expenditure change would produce a deeper optical correction.

The second uncertainty is component availability. TrendForce identified lasers and related parts as important bottlenecks for capacity expansion. Shortages can restrict shipments, but easing shortages can later create excess inventory across several supply-chain levels.

Companies often respond to shortages by placing larger or earlier orders. Suppliers may interpret those orders as end-market demand even when customers are building safety stock. This bullwhip effect can exaggerate apparent demand upstream.

The third uncertainty is architecture. Pluggable transceivers remain central to current deployment, but linear optics and co-packaged designs aim to reduce power consumption. A faster transition could move value away from some established module components.

CPO adoption is not automatic. Operators must solve cooling, repair, interoperability, and manufacturing problems. Large deployments require reliable economics across the entire switch system, not only a successful optical engine demonstration.

The fourth uncertainty involves trade policy and geographic exposure. The leading Chinese suppliers earn much of their revenue overseas. Export controls, customer sourcing rules, tariffs, and localization requirements can alter access to components or markets.

The fifth uncertainty is cash conversion. Rising receivables and inventory are normal during rapid expansion, but they deserve scrutiny. Revenue quality becomes clearer when earnings turn into operating cash without continued working-capital acceleration.

Gross margins provide another signal. The current shift toward 800G and 1.6T products supported favorable margins for major vendors. Competition and customer negotiations can narrow that benefit as production becomes more standardized.

Capacity utilization will become decisive after expansion projects begin operating. High utilization spreads equipment and labor costs across more units. Low utilization produces the opposite effect, even if long-term demand remains intact.

These risks do not erase the first-half performance. They define the conditions needed to sustain it. The correct reading is strong verified demand combined with unusually high execution and concentration risk.

That distinction matters for anyone following technology news. A strong market does not make every supplier equally attractive or every forecast dependable. Product exposure, customer qualifications, cash flow, yield, and architecture determine which revenue survives.

Three Signals Will Test the AI Optics Expansion

The next stage depends on order durability, 1.6T execution, and proof that profits are converting into cash.

The first signal is hyperscaler capital expenditure and optical order visibility. Upcoming results from Amazon, Google, Meta, and Microsoft should reveal whether elevated infrastructure budgets remain intact. Sustained spending would reinforce the view that first-half demand reflects continuing deployment.

A spending reduction would weaken that conclusion. It would not immediately stop shipments because suppliers operate against existing orders and schedules. However, weaker forecasts would raise the risk of inventory corrections during 2027.

The second signal is the volume ramp for 1.6T products. Investors should look beyond announcements and track customer qualifications, actual shipments, production yields, and revenue contribution. Tens of millions of ports would validate forecasts from LightCounting and other research firms.

Delays would shift more demand back toward 800G and extend the mixed-generation period. That outcome would still support optical revenue, but it could change margins and competitive positions. Suppliers prepared for both generations would have greater flexibility.

The third signal is operating cash flow across Innolight, Eoptolink, TFC, Accelink, and upstream component companies. Stronger collections alongside stable inventory would support the quality of reported growth. Continued divergence between profit and cash would increase concern about working-capital pressure.

These signals should be evaluated together. High capital expenditure without successful 1.6T production can produce component shortages. Successful production without sustained customer budgets can create excess capacity.

For enterprise technology buyers, the reporting cycle also offers a practical lesson. AI system performance depends on networking as much as accelerator specifications. Procurement teams should evaluate topology, optical availability, power consumption, and upgrade timing before treating compute capacity as an isolated purchase.

Engineers face a similar challenge. Pluggable optics, linear designs, and co-packaged systems create different maintenance and integration requirements. Technical teams need a durable record of vendor claims, qualification results, and architecture decisions.

A searchable technical knowledge base can help teams preserve that evidence as specifications change. The goal is not to predict one winning architecture. It is to keep decisions connected to tests, constraints, and deployment results.

The half-year reports establish a clear starting point. Innolight and Eoptolink lead on scale, while demand has spread into competing modules, optical components, chips, and packaging. The expansion is real, but its durability still rests on a narrow group of cloud customers.

Watch the next hyperscaler budgets, verified 1.6T shipments, and cash conversion. Together, those measures will show whether this technology news marks a durable optical infrastructure cycle or the most profitable stage of a concentrated buildout.