2026.09.23 | AC | Optical | China Module Maker: Key Client Demand

摘要

·In 2027, Cisco's 800G demand is expected to increase to 2 million units, while 1.6T demand is projected to double. ·Meta will serve as the core incremental client in 2026. ·DSP chips represent the most constrained link, bottlenecked by TSMC's 3nm/5nm capacity, with lead times exceeding 6 months.

The following insights are from a senior industry expert:

Q&A

Could you outline Cig Shanghai's 800G and 1.6T optical module shipments to its primary clients, Cisco and Meta, across the first and second halves of 2026?

Client focus in 2026 is concentrated primarily on Cisco and Meta.

Cooperation with Cisco follows a JDM model, focused mainly on short-reach silicon photonics modules in OSFP form factors, such as DR8 and 2×DR4. In 1H2026, combined 800G optical module shipments to Cisco reached over 700,000 units, comprising roughly 350,000 units in Q1 and approximately 400,000 units in Q2. For 1.6T products, shipments mainly involve 2×DR4 and 2×FR4 modules in OSFP packaging. Small-batch deliveries started in the first half of the year, totaling around 30,000 units.

Meta is a major new client added in 2026. The first half of the year focused on product testing and qualification, with certification secured in April, after which bilateral cooperation kicked off. About 10,000 to 20,000 units of rush orders were fulfilled in Q2. Combined deliveries across Q3 and Q4 are projected to exceed 200,000 units. Specific Q4 volumes will depend on the cadence of the client's order releases. Currently, 1.6T products remain under testing at Meta, and concrete progress will hinge on the clarity of Q4 test results.

What are the current gross margin levels for 800G and 1.6T optical modules across different clients? Are the rumors circulating in the market of a 45% gross margin accurate?

Under the JDM model with Cisco, the PIC and DSP are customer-supplied, while our company self-procures the other electronic components, active components, passive components, and PCBs. Under this structure, gross margins for standard 800G silicon photonics products sit roughly between 30% and 35%. Margins on 1.6T products are relatively higher, largely benefiting from current supply and capacity dynamics.

For Meta, because cooperation only commenced in 2Q2026, some pricing concessions were granted to win orders, so current margin levels have not yet matched those of Cisco.

Regarding the 45% gross margin cited in the market, that level only occurs during early product rollout stages or within select customized projects. Once mass production begins, gross margins cannot be sustained at such elevated levels.

What is the company's current actual optical module capacity layout? Has it reached 1 million units per month?

Currently, our manufacturing footprint is primarily distributed across the Jiashan plant in China, the Malaysia plant in Southeast Asia, and the Mexico facility.

The Jiashan site was developed in two phases: Phase 1 went into operation in 2025, and Phase 2 was commissioned in March

1. As of September 2026, the Jiashan plant can produce approximately 250,000 optical modules per month, dedicated

primarily to delivering 800G and 1.6T products.

As of September 2026, monthly 800G optical module capacity at the Malaysia plant stood at around 150,000 units. By Q4, monthly 1.6T capacity is expected to reach 50,000 units, which will bring the plant's total monthly optical module run-rate to 200,000 units.

The Mexico facility currently manufactures telecom, broadband access, and wireless products, and is gradually pivoting into optical module assembly. By 4Q2026, the site is slated to build an optical module capacity of 30,000 to 50,000 units per month, with actual progress contingent on market conditions.

In aggregate, reaching 1 million units per month is not the current reality, but rather a planned milestone targeted for late 2026 or 1Q2027. Capacity expansion must be synchronized with client orders, staffing availability, and holistic cross-facility planning. What is relatively concrete right now is that the company will possess a monthly capacity of 600,000 optical modules by 4Q2026.

Looking ahead to next year (2027), what is the demand outlook from major clients for 800G and 1.6T products?

Looking to 2027, because of chip shortages across DSPs and PICs alongside manufacturing yield issues, 1.6T penetration may lag expectations; consequently, 800G and 1.6T will share the spotlight as core volume drivers. Looking at major clients specifically:

Cisco: Demand for 800G in 2027 is forecast to rise from this year's 1.6 million units to approximately 2 million units;

1. 6T demand is expected to expand from 200,000 units this year to over 500,000 units.

Meta: The company spent 1H2026 primarily on site audits, qualification, and testing, and is expected to start capturing order flow beginning in 1Q2027. Capturing a 10% share of Meta's 10-million-plus total 800G requirement would translate into orders of over 1 million units.

On 1.6T products, projected 2027 demand from both Cisco and Meta exceeds 400,000 units with no capped ceiling, depending ultimately on collaboration progress.

Additionally, if validations wrap up smoothly, clients such as Microsoft, Amazon, and Google could contribute incremental orders. Overall, demand for both 800G and 1.6T will maintain upward trajectories in 2027.

Focusing on Cisco as a core client, what is its specific demand growth profile for 800G and 1.6T products next year (2027), and what is the overall growth outlook?

As things stand, Cisco's demand growth for 800G products next year (2027) is expected to top 30%. The growth in 1.6T demand will be considerably more aggressive, doubling year over year. Bilateral collaboration centers predominantly on short-reach silicon photonics modules, a niche where 1.6T adoption is expected to ramp comparatively fast.

Which new clients are currently undergoing qualification, and what is their progress?

New clients currently in validation are primarily large North American CSPs with massive volume requirements, alongside domestic players like Alibaba and Tencent. Validated products are predominantly mainstream OSFP packages such as DR8, 2DR4, and FR4.

Among these accounts, Microsoft is advancing fastest; its 800G products have completed qualification and advanced to the purchasing phase. Currently, our company is penetrating as a secondary or tertiary supplier.

Validations with Amazon and Nvidia are progressing in parallel.

Since Microsoft has already certified the 800G product, why are initial orders only in the tens of thousands of units? Is there potential to land million-unit orders going forward?

Ramping up with a new customer is an incremental process that typically moves through technical validation, factory audits, and framework agreement execution. In our current engagement with Microsoft, 800G products have passed certification, 1.6T products remain in testing, and interoperability evaluations are problem-free. Initial purchase volumes remain modest because we are entering as a secondary or tertiary source; their primary incumbents are Finisar overseas and Innolight domestically. The ultimate scale of subsequent business will depend on deeper mutual planning, encompassing our pricing, capacity, and technical compliance, balanced against client goals around procurement pricing and supply chain diversification.

What is the current qualification status with Amazon and Nvidia, and when is it expected to finish?

Amazon's qualification process entails sample submission followed by interoperability, reliability, signal integrity, power consumption, high- and low-temperature tolerance, optical output power, and receiver sensitivity data logging and verification. Nvidia's timeline runs parallel to Amazon's. Both clients are slated to finish validation work within the fourth quarter of this year (2026), though final execution timelines will depend on specific validation outcomes.

Which entity does the "new cloud" client specifically refer to? What is the cooperation model with Oracle?

Regarding newly entering cloud service providers, there is no explicit indication. Cooperation with Oracle is indirect rather than via direct optical module sales. Specifically, business is conducted through Cisco's integrated solutions, meaning Cisco packages our optical modules into the architectures it delivers to downstream clients like Oracle.

Factoring in demand from key accounts like Cisco and Meta, what are the company's shipment and capacity expansion targets for 800G and 1.6T optical modules next year (2027)?

The most concrete capacity plan right now is an expansion to 6 million units. This footprint breaks down roughly as follows: approximately 2 million 800G units allocated for Cisco, and over 1 million 800G units for Meta, which combined reach over 3 million units. Of the remaining balance, around 1 million units are earmarked for 1.6T lines, while the other 2 million units are reserved to provide capacity buffer for Microsoft, Nvidia, and Amazon. Subsequent expansion will depend on order visibility; for example, if demand from Microsoft, Nvidia, or Amazon breaches 300,000 units, we may evaluate lifting capacity toward 10 million units. This roadmap is formulated on expectations for 1H2027; full-year realities will only solidify once order visibility clears in 1Q and 2Q2027. The company's expansion philosophy closely tracks firm bookings, factoring in depreciation, overhead, and headcount rather than speculatively building out capacity ahead of demand.

What are the company's R&D initiatives, primary partners, and project milestones in NPO (Near-Package Optics)?

Serving as an intermediate pre-validation phase leading into CPO (Co-Packaged Optics), NPO remains an area of sustained R&D investment for us. Projects such as 3.2T NPO center on core building blocks including optical modems, optical engines, and external laser sources, advancing alongside client milestones.

We currently have two primary partners:

Cooperation with Cisco: Prioritizing ELSFP (External Laser Source Small Form-factor Pluggable) architectures. Working samples are expected to ship to the client for validation in 4Q2026, with production orders potentially materializing in 2Q or 3Q2027 based on test results. This initiative is being verified jointly on Cisco 8000 series switches, targeted for future delivery as an integrated solution. Within this roadmap, external optical engine samples are expected in 4Q2026, whereas integrated laser solutions are not anticipated until 2H2027. Cisco will market these NPO-enabled platforms to end clients like Oracle and xAI.

Cooperation with Nokia: Both parties have executed an R&D agreement and are conducting technical solution reviews. If reviews wrap up cleanly in 4Q2026, formal development will begin, with validation targeted for 2027. Assuming smooth progress, volume orders and concentrated deliveries could ramp in 2H2027. Nokia's 3.2T NPO primarily addresses remote base station backhaul, compute node interconnects, and datacenter cluster optical interconnect scenarios for its own enterprise customers, such as AMD.

How is the company's optical module collaboration with Nvidia progressing, and what products and use cases are in focus?

Our collaboration with Nvidia currently centers on 1.6T optical interconnects, while early-stage technical alignment for

1. 2T products is also underway, albeit at a slower pace. The 1.6T optical modules are predominantly deployed inside

Nvidia rack systems, facilitating internal optical-to-electrical interconnects across compute clusters like GB200 and GB300. At present, primary interconnect demand for GB200 and GB300 architectures is concentrated at 1.6T speeds. Regarding 1.6T validation milestones, conclusions are anticipated in 4Q2026, which may unlock subsequent order allocations. The company does not currently provide 800G products to Nvidia.

In NPO or ELS (External Laser Source) modules, which suppliers provide the primary laser sources?

For ELS (External Laser Source) modules, we currently continue using the EML laser source suppliers established across legacy 800G and 1.6T production. Qualified overseas vendors include Lumentum and Sumitomo. While we are in active contact with domestic vendors, primary sourcing remains anchored with overseas vendors at this stage. For instance, the Cisco collaborative project deploys Lumentum laser sources. External architectures relocate lasers that were historically integrated inside the optical engine to the exterior, effectively resolving heat dissipation and signal interference constraints.

Who supplies the high-power lasers currently used, what are their power ratings, and is the business geared toward domestic or overseas markets?

We currently deploy Lumentum's high-power lasers, typically rated at 300mW. Business-wise, demand is heavily geared toward overseas accounts; domestic demand is presently centered on telecom transmission, exemplified by 400G and 800G deployments across mainstream CSP clients. Among these, 400G sees broader deployment, while 800G remains in an incremental ramp-up phase. Telecom carriers, meanwhile, primarily utilize 100G and below solutions, making them minor contributors to overall revenue today. Although domestic volume is accelerating, demand above 800G remains limited for now.

Across 800G and 1.6T architectures, is the company leaning toward silicon photonics or traditional EML roadmaps?

Current resource allocation and engineering bandwidth are heavily focused on silicon photonics. Traditional EML designs have not been abandoned; we maintain complete technical competencies and know-how. If clients present occasional or small-scale EML orders, we have the manufacturing readiness to deliver, but we do not run large-scale production, keeping activity confined to small-batch sampling or modest shipments.

Which vendors supply the CW laser sources for the company's optical modules? Are there domestic suppliers with core status or sizable allocations?

We engage a diversified supplier base for CW laser sources, working with more than five qualified vendors. Overseas suppliers include Lumentum and Coherent. Domestic partners include Dingxin, Yuanjie, and Z-Block. Among these, our relationship with Dingxin is particularly tight; modules currently delivered to Microsoft and Meta are powered by Dingxin CW chips.

Z-Block previously placed a large order for indium phosphide (InP) substrates from AXT. Was this capacity secured exclusively for your company, or part of Z-Block's general supply chain buildout for all customers?

This was primarily driven by Z-Block's own forward-looking capacity planning and delivery commitments. Indium phosphide substrates are critical raw materials for fabricating core components like lasers, photodetectors, and optical amplifiers, and supply remains tight across the market. Z-Block's agreement with AXT was structured to ensure upstream stability. As a customer of Z-Block, our company naturally benefits, as reliable substrate inventory safeguards their capacity to supply us with CW lasers and other chips. Z-Block remains the contracting purchasing entity, while our company acts as an indirect beneficiary, improving long-term visibility for our CW chip supply.

Within the current 800G and 1.6T bill of materials, which component faces the most acute shortage? What is driving it?

Currently, the tightest component is the DSP chip. Silicon photonics chips have at least three viable suppliers, and CW sources have over five, whereas DSP capacity is significantly tighter. There are two core drivers: First, global supply for optical module EDSPs is heavily concentrated, with Marvell and Broadcom together commanding an 80% market share. Second, fabrication technology poses a major bottleneck: 800G DSPs require 5nm processes, while 1.6T DSPs necessitate 3nm nodes. Currently, viable foundry options are restricted to Samsung and TSMC. Samsung's yields remain sub-optimal, funneling wafer demand almost entirely into TSMC. However, TSMC's 3nm capacity is largely consumed by Nvidia GPUs, Apple APs, and Broadcom custom ASICs (switching/accelerator chips), leaving minimal wafer allocation for 1.6T DSPs. These dynamics push DSP procurement lead times past 6 months.

How is the company navigating the DSP supply crunch? What do current inventory buffers look like?

To counter DSP tightness, we are implementing a multi-pronged approach. First, securing capacity and buffering components early; for example, as of September 2026, we have already engaged suppliers regarding 1H2027 DSP requirements, making 30% advance payments per contract terms. Second, establishing a diversified supplier base: while Marvell is our primary supplier, we also maintain a dual-track relationship with Broadcom for 1.6T products. In addition, we are validating products from vendors like Credo to provide downstream alternatives. On the inventory front, we maintain safety buffers: on top of the 500,000 to 600,000 customer-supplied and self-procured DSPs handled monthly, we carry an extra buffer of roughly 100,000 units in stock.

What is the procurement model for DSPs supplied by Broadcom or Marvell—direct purchasing or designated by CSP clients?

Procurement falls into two distinct models. The first applies to JDM/JSP partnerships like Cisco, where materials are consigned by the client. Cisco maintains component buffers aligned with its forecasts and issues them to us through its supply chain infrastructure, maintaining active involvement throughout. The second applies to OTM programs with accounts like Microsoft and Meta, where Cig Shanghai is the legally responsible purchasing entity. Although clients step in to help coordinate when delivery snags arise, commercial ownership and procurement liability rest squarely with Cig Shanghai.

Cisco's quarterly order volume held flat at 400,000 units across Q2, Q3, and Q4 of 2026. Does this signal that subsequent order runs will plateau without meaningful upside?

Yes, Cisco's run-rate each quarter remains broadly predictable. Even when sporadic surge orders of 20,000 to 30,000 units arrive—for instance, pulling for LPO or iSeek products—they create negligible impact against a baseline of several hundred thousand units, keeping aggregate volumes stable without wide swings.

In the Cisco engagement, are client-furnished DSPs and silicon photonics chips booked as company revenue? What percentage of total module pricing do these parts represent?

Under our operating model with Cisco, customer-furnished DSPs and PICs are excluded from recognized top-line revenue. For a transceiver invoiced at $350, the combined value of the DSP and PIC is approximately $100. Consequently, the revenue recognized by the company sits around $250, with gross margins calculated strictly against that $250 base. Conversely, for accounts like Meta and Microsoft where we sell full-assembly modules, the total value of all components is recognized in revenue.

What does the pricing landscape look like for 1.6T optical modules? Taking your core short-reach silicon photonics transceivers as a benchmark, what is the ASP, and does revenue recognition track the 800G model?

Pricing on 1.6T modules varies significantly across architectures and use cases. For example, coherent modules engineered for long-haul transport top the pricing curve at over $1,000, whereas simplified LPO transceivers price lower, in the $700 to $800 range. Our volume is concentrated in short-reach silicon photonics, where ASPs sit slightly above $800. In terms of top-line accounting, the treatment mirrors 800G: customer-supplied DSPs and silicon photonics dies are not recognized in our reported revenue.

Who are the company's primary suppliers for silicon photonics chips today? What is the current adoption status for domestic suppliers Xihe Technology and SiFotonics?

We source silicon photonics chips from several key players. For Cisco programs, we rely primarily on their subsidiary, Acacia. Other foreign vendors include Broadcom. Domestically, engagements center on Xihe Technology and SiFotonics. Currently, SiFotonics commands a slightly larger share than Xihe Technology, but allocations remain tied to module interoperability, optical performance, and price-to-performance metrics; we run ongoing compatibility testing across all vendors, prioritizing components that deliver superior performance and cost efficiency.

How did legacy business perform in the first half, and did it weigh on overall corporate profitability?

Legacy business revenue declined year over year in the first half of the year. This was driven by two headwinds: first, legacy business inherently carries lower gross margins; second, price surges in commodities like Flash and MLCCs lifted input costs, squeezing pricing flexibility and overall market traction. Consequently, in an already soft market, rising bill-of-materials costs further compressed operating margins. However, the legacy segment did not slip into operating losses; it simply operated with thin gross margins. To navigate market softness and input inflation, we adopted a strategy of dialing back shipments to avoid loss-making volume. Strategically, it was precisely due to the low margins and sluggish growth in legacy product lines that the company launched its pivot in 2018 through overseas acquisitions (such as Japan's Oclaro Japan and a Lumentum subsidiary), realigning our core focus toward optical transceivers from 2019 onward. Therefore, while legacy business still delivers positive gross revenue, it has undeniably created a drag on blended gross profitability.

Out of the company's roughly RMB 300 million net profit in the first half of the year, what proportion was contributed by the optical module business?

The optical module division contributed roughly 70% of 1H net profit. Looking at margin profiles, standard optical modules generate gross margins between 30% and 35%, while 1.6T margins are expected to track between 35% and 40%.

Regarding competitor Source Photonics, what does their operational progress, technical positioning, and capacity layout look like this year?

Source Photonics has moved aggressively this year across commercial expansion, capacity scaling, and tier-one client wins. Formerly based in California, following its acquisition by DSBJ in 2025, the firm established an operational model with overseas R&D headquarters and domestic corporate headquarters. Its competitive strengths center on a few key areas: First, as an IDM player, it commands integrated chip-and-module capabilities supported by captive chip technology, including 100G and 200G technical know-how in EML chips. Second, it possesses dedicated in-house DSP design capabilities. Finally, it operates a distributed manufacturing footprint spanning Chengdu, Changzhou, Taiwan, and Thailand. On balance, Source Photonics exhibits distinct competitive advantages across technology R&D and global manufacturing.

What are the specifics regarding Source Photonics' DSP design capabilities, including their development model and fabrication channels?

To our knowledge, Source Photonics possesses 800G 5nm DSP design capabilities. Operationally, this silicon development is typically carried out jointly with hyper-scale CSP clients (such as Meta, Microsoft, and Google), followed by collaborative tape-outs and contract manufacturing at TSMC for verified designs. Globally, leading-edge 3nm and 5nm foundry capacity is concentrated almost exclusively between TSMC and Samsung. Their active silicon initiatives encompass both DSP and EML chips.

How are Source Photonics' order intake and shipping runs tracking with Meta across this year (2026) and next year (2027)? What is their core edge in capturing volume?

It is understood that Meta reallocated substantial order volume to Source Photonics this year (2026), with their 800G order run-rate projected to surpass 4 million units, alongside modest 1.6T allocations. Full-year 2026 shipments into Meta are expected to reach 3 to 4 million units. In addition, Source Photonics has locked in incremental wins at Microsoft and Google. Their competitive edge is driven not by aggressive pricing, but rather by their captive chip technology and offshore manufacturing footprint. Looking ahead to next year (2027), Source Photonics continues to expand its offshore facilities, such as in Thailand, and its order allocation with Meta is anticipated to surpass last year's level.

Market speculation suggests Meta's combined demand for 800G and 1.6T optical modules could top 10 million units next year. What market share do you anticipate Source Photonics securing?

Meta's total requirement next year is anticipated to exceed 10 million units. However, because hyper-scalers systematically enforce multi-sourcing, Meta is unlikely to award the entirety of its procurement to Source Photonics. Source Photonics is expected to capture roughly 40% to 50% of the volume, translating into roughly 6 million units.

What is the current vendor hierarchy across Meta's 800G optical module supply chain?

In 1H2026, Innolight served as Meta's primary supplier, but a portion of order allocations rotated toward Source Photonics in the second half of the year. Currently, Source Photonics has stepped into the lead supplier position, followed by Innolight in second place. Finisar and Eoptolink hold the third and fourth slots, while other vendors like Cig Shanghai are making inroads into the vendor matrix.

What is the status of Source Photonics' collaboration with Amazon on 800G modules, and what share are they modeled to capture across 2026 and 2027?

Source Photonics supplies 800G modules to Amazon. Amazon's total 800G module demand in 2026 is roughly 5 to 6 million units, with Source Photonics modeled to secure at least a 10% share. For 2027, their allocation remains fluid. If their product specifications and co-development align closely with the client, their share could expand; conversely, if their competitive position in pricing or other factors weakens, their share could see volatility. Under conservative assumptions, their 2027 allocation will likely hold around 10%. Amazon maintains a diversified vendor ecosystem that includes domestic suppliers like Eoptolink, Innolight, and TFC alongside Source Photonics.

What is Amazon's projected blended demand across 800G and 1.6T optical modules for 2027?

Amazon's 2026 demand run-rate stands at 5 to 6 million 800G units and roughly 1 million 1.6T transceivers. Extending from that base, its blended 800G and 1.6T requirement in 2027 is forecast to hit roughly 7 million units, nearing the 10-million threshold, though remaining below Meta in total volume.

Where does Source Photonics stand in Nvidia's supply chain, particularly regarding 1.6T product qualification?

Source Photonics' engagement with Nvidia appears to remain in early technical qualification, and they have yet to secure certification for mass-production orders. Nvidia's volume demand is concentrated in 1.6T optical modules deployed within liquid-cooled optical fabric architectures across GB200 and GB300 racks, which enforce exceptionally strict IC performance thresholds. Because Source Photonics' engineering and manufacturing resources remain heavily anchored to 800G, progress with Nvidia has moved comparatively slowly, with transceivers still working through qualification.

If Source Photonics secures entry into Nvidia's 1.6T supply chain in 2027, what market share could they realistically secure?

Within Nvidia's current 1.6T supplier landscape, overseas vendors Lumentum and Finisar collectively account for roughly 40%, while domestic suppliers Innolight and Eoptolink together hold another 40%. The remaining 20% pool is contested among players like TFC, Source Photonics, and Linktel. Should Source Photonics succeed in winning commercial volume in 2027, they are projected to take roughly 10% of that third-tier allocation.

What is the status of Linktel's collaboration with Nvidia, and what are their technical advantages and operational bottlenecks?

Linktel's partnership with Nvidia is aimed primarily at next-generation architectures, specifically the optical engines required for NPO and CPO. Linktel has notable engineering depth; structured as an IDM player similar to Source Photonics, it holds extensive capabilities in optical component packaging and optical IC design. For example, their thin-film lithium niobate modulator technology accelerates development of high-speed 400G transceivers. Additionally, Linktel possesses solid technical reserves in long-haul transceivers (such as 10km LRS). However, their primary constraint today is overall manufacturing capacity, particularly their limited offshore footprint, which could limit their ability to absorb large-scale allocations later; that said, capacity can be expanded as the partnership matures.

What is the industry consensus regarding Meta's 800G optical module requirements for 2026?

Prevailing industry consensus pegs Meta's 2026 800G optical module procurement at over 10 million units. Earlier figures suggesting full-year demand of only 3.8 to 4.6 million units—with 800G representing just 2.7 to 3.2 million—materially undercounted the real requirement and conflicted with industry consensus.

What commercial framework governs engagements with clients like Meta, and what is the typical procurement cadence?

Engagements with Meta follow an ODM framework, where production volumes are formalized through ad-hoc bidding tenders or structured quarterly business reviews. Allocation across other tier-one peers also follows competitive bidding, where customers comprehensively grade bill of materials, pricing, technical compliance, and downstream capacity support. Clients typically adopt a "select-three-from-five" or "select-two-from-three" vendor allocation structure. Ordering rhythms track quarterly cycles, though rapid-turnaround or expedited batches are confirmed with the supply chain on an ad-hoc basis.

Why did Source Photonics emerge as Meta's lead supplier starting in 2H2026, and will vendor market shares at Meta remain dynamic going forward?

All supply allocations are awarded via competitive bidding; Source Photonics' expanded share reflects the client's holistic scoring. Key drivers included: first, captive silicon assets co-developed with Meta; second, their established offshore plant network. These assets aligned closely with Meta's strategic criteria, driving order reallocation. Looking forward, vendor market share remains dynamic and entirely subject to change. Every allocation is decided through competitive tender, evaluated against cost, price, technical support, module performance compliance, and manufacturing scalability. If an incumbent like Innolight enters 2027 with vast material stockpiles and unconstrained manufacturing capacity, it could very well leverage aggressive pricing and volume scale to reclaim the top bidding spot.

Market rumors suggest Cig Shanghai won orders from Meta in June 2026 at low prices of $300 to $320. Is this pricing accurate?

In 2Q2026, the company completed site audits, factory certifications, and product qualification. Early on, to secure an initial foothold, we indeed offered pricing concessions across conventional optical modules and silicon photonics LPO (Linear Pluggable Optics) products. We do not anticipate additional price concessions in the third quarter. Going forward, pricing on standard products will be systematically recalibrated based on underlying bill of materials, direct labor costs, and overall market variables.

What are your demand projections for 800G and 1.6T transceivers at Microsoft, Oracle, and xAI in 2027?

Microsoft's 2027 procurement is projected at roughly 10 million 800G optical modules and 1 to 2 million 1.6T transceivers. As for Oracle and xAI, because our engagement is routed entirely through Cisco without direct client contact, we do not possess standalone procurement projections.

Regarding the 3.2T optical transceiver roadmap, will conventional EML or silicon photonics paired with thin-film lithium niobate become the dominant architecture?

The company is positioned across both technology tracks. During early stages, priority validation is centered on external laser source configurations utilizing LPO. By 2H2027, we plan to roll out thin-film lithium niobate solutions built on an 8-channel, single-lane 400G/λ architecture, packaged in an NPO (Near-Package Optics) form factor with integrated lasers.

Is NPO (Near-Package Optics) a stopgap solution or an enduring, iterable architectural roadmap?

NPO's lifecycle depends directly on how CPO (Co-Packaged Optics) matures. If vendors like Nvidia successfully validate and scale CPO while resolving thermal dissipation and field serviceability challenges, NPO may remain an intermediate bridging step. Conversely, if CPO runs into delays around industry standardization, thermal management, or field repairability, NPO will enjoy an extended commercial runway. Today, CPO specifications remain fluid, and its long-term viability depends on the evolution of next-generation silicon architectures.

What is the demand outlook for NPO and CPO in 2027?

We have initiated NPO collaborations with accounts like Cisco and Nokia; volume demand in 2027 will hinge on sample validation and pilot deployments, meaning definitive volume forecasts are premature. Regarding CPO, commercial deployments are not anticipated in 2027, with the earliest viable production push arriving in 2028.

What is the company's current monthly production run-rate for high-speed optical transceivers, and what does the forward expansion roadmap look like?

Our manufacturing operations are situated across Jiashan, Malaysia, and Mexico. As of September 2026, the Jiashan site supports a monthly capacity of 250,000 units, primarily supplying global regions outside North America. The Malaysia site runs an 800G monthly capacity of 150,000 units focused on North America, and will bring 50,000 units of monthly

1. 6T capacity online in the fourth quarter. The Mexico plant currently focuses on legacy broadband access and wireless

hardware, with plans to initiate optical transceiver assembly in Q4 covering 800G, 1.6T, and future NPO lines, followed by phased expansion through 1Q2027 based on progress. Our formal plan targets an annualized capacity of 6 million units across the three plants by year-end 2026. Higher capacity targets—such as monthly runs of 1 million, 1.5 million, or beyond—will be determined dynamically based on firm client purchase orders, equipment lead times, and staffing ramps.

Regarding NPO (Near-Packaged Optics), how does the industry evaluate external versus internal laser source architectures, and what is the trajectory moving forward?

The architectural divergence between external and internal laser sources hinges on whether the laser emitter is physically decoupled from the primary module chassis. For example, in a 3.2T transmission design, integrating the laser, photodetectors, and optical modems onto a unified module constitutes an internal architecture. Currently, given the delicate trade-offs between thermal dissipation and power density, the architecture we are co-developing and deploying with clients is an external ELSFP (External Laser Source Small Form-factor Pluggable) form factor. This decouples the laser from the optical engine to the front faceplate of the chassis, while leaving the photodetectors and modems seated on the host board. In parallel, the company is developing integrated internal laser solutions targeted for release in 2H2027. Both architectures are currently progressing concurrently, and it remains premature to declare a definitive winner. Over the long run, however, if integrated internal modules successfully resolve heat dissipation and power consumption constraints, their cost and integration advantages could make them the primary architecture.

Is it accurate to assume that in NPO designs, if thermal and power hurdles are cleared, the industry will favor internal laser designs, falling back to external configurations only if those challenges persist?

That is an accurate assessment. The core engineering calculus determining whether to use external or internal configurations centers squarely on power consumption and thermal dissipation.

It has been noted that internal architectures can curb power consumption by leveraging lower-power CW silicon photonics chips (e.g., 120–150mW). Is this technical approach viable?

Yes, driving down power draw is the prerequisite for deploying internal laser sources. This necessitates core material breakthroughs, such as swapping standard silicon wafers for novel substrates like indium phosphide; if such technologies mature and successfully lower overall thermal dissipation, the commercial feasibility of internal architectures will rise substantially.

Given that current NPO projects with North American clients deploy external laser sources, how are competing suppliers approaching this architectural choice?

NPO technology remains in an early stage of development, and the industry has yet to coalesce around a unified standard. Consequently, individual vendors are executing based on their internal R&D capabilities and client-specific co-design requests, developing external and internal architectures simultaneously. Vendors are actively shipping samples to clients, and which architecture ultimately dominates will become clear only after transceivers clear qualification and translate into volume production orders, guided by client adoption. This contrasts with 1.6T transceivers, where the architectural baseline is largely established, despite ongoing segmentation between standard versus silicon photonics designs and DSP-based versus LPO architectures.

Across the optical module PCB supply chain, which fabricators supply 1.6T mSAP substrates, and are there capacity bottlenecks?

The company maintains established relationships with several top-tier PCB fabricators, including Shennan Circuits, Wus Printed Circuit, and Kinwong Electronic. At present, domestic PCB manufacturing capacity is more than adequate. Although transceivers impose exacting technical specifications on PCBs, the physical panel sizes are compact, consuming a negligible fraction of overall fabricator capacity. So long as fabricators possess the required process know-how, production capacity is not a constraint. Accordingly, the PCB supply chain faces zero bottlenecks.

Beyond DSP chips, are there other latent material bottlenecks across the optical transceiver supply chain?

Beyond DSP chips, silicon photonics dies also carry supply chain risk. If silicon photonics foundry capacity remains constrained, shortages could readily emerge.