Lead: AI computing power demand is still growing, but optical module companies can no longer prop up their stock prices on order expectations alone.

On November 11, the optical module sector staged a sharp rebound.

The Wind CPO Concept Index climbed 8.25%, with leading stocks such as Zhongji Innolight, Eoptolink, and TFC Optical all posting collective gains. Only a few trading days earlier, these same companies had just been through a notable round of correction.

The market keeps trading the same question back and forth: how many more optical modules will overseas cloud providers continue to buy?

Talk of delays in North American cloud giants' data center projects and cuts to optical module orders has been circulating in the market again recently. Although the companies concerned have come forward to respond one after another, investors' concerns about orders have not entirely dissipated.

On November 3, Zhongji Innolight said during an investor call that the vast majority of customers had not lowered demand or orders, that a small number of customers had made product structure adjustments, and that demand for 1.6T products from some customers had actually risen somewhat.

Cambridge Technology likewise said at its earnings briefing on November 9 that the company had strong order demand for high-speed optical modules and that capacity utilization remained high, adding that market rumors were inconsistent with the company's actual operating conditions.

On November 12, discussion of order cuts in the optical communications industry surfaced once again, and multiple listed companies responded that there had been no adverse changes in recent operating conditions.

Such rumors keep resurfacing not simply because market sentiment is fragile.

Over the past year, the stock prices and valuations of optical module companies have rested to a large extent on expectations of continued growth in overseas AI capital expenditures. Once investors begin to doubt whether these investments can be converted into orders as planned, every change along the industry chain gets magnified.

01 Orders no longer have only one possibility: growth

Over the past two years, optical modules have been one of the few segments of the AI computing power industry chain able to consistently deliver on performance.

NVIDIA keeps rolling out new generations of GPUs, and cloud service providers such as Microsoft, Google, and Amazon are expanding data centers, driving growth in data transmission demand between servers. As a key component for data transmission within data centers and between different devices, optical modules benefit directly from network bandwidth upgrades.

From 400G to 800G and then to 1.6T, product iteration continues to push up optical communications demand for individual data centers.

This growth logic has not fundamentally changed so far.

Amazon's third-quarter AWS revenue reached $46 billion, up 39.3% year on year. Alphabet's third-quarter Google Cloud revenue reached $30.7 billion, up 103% year on year. Microsoft's capital expenditure in the first quarter of fiscal 2027 reached $45.2 billion, and it expects more than $50 billion in the next quarter.

The investment scale of these companies remains enormous, and their cloud businesses are also growing. At least judging from the data disclosed so far, overseas tech giants have not collectively pulled back from AI infrastructure construction.

But the focus of market attention has shifted somewhat.

Data center construction is not an investment that can be completed all at once. Land, power, servers, switches, optical modules, and network architecture need to be deployed gradually according to project progress. A delay in one project may affect the delivery time of a batch of equipment; changes in network architecture may also alter the procurement ratio of different product models.

This means that continued growth in cloud providers' capital expenditures does not mean that every supplier's orders can grow at the original pace.

Zhongji Innolight's response precisely reflects this change.

The company stated that the vast majority of customers have not lowered demand or orders, but a small number of customers do have product structure adjustments. At the same time, demand for 1.6T products from some customers has increased.

This is not the same thing as the broad order cuts described by the market.

If customers reduce 800G purchases while increasing 1.6T purchases, orders for a single product category may decline; but from the perspective of the entire optical communications system, demand may not necessarily decrease.

For optical module companies, what truly deserves attention is total order volume, product structure, and supplier share, rather than procurement changes for a single model.

Zhongji Innolight stated during its third-quarter earnings call that orders for 800G and 1.6T products both achieved quarter-on-quarter growth compared with the second quarter, with revenue from 1.6T products accelerating further. The company also stated on November 6 that orders from major customers had basically covered the full year of 2027, and some key customers had already provided demand guidance for 2028.

Eoptolink also stated in its production and operation announcement on October 23 that 800G and 1.6T are the main products delivered this year, and that 1.6T optical modules have entered a stage of continuous volume ramp-up.

This information shows that leading companies still have strong order support at present.

But orders covering the full year does not mean that revenue and profit for the coming year are already determined. Customer demand, product certification, delivery progress, and procurement prices will still affect the final operating results.

02 1.6T has become the new focus of competition

The optical module industry is undergoing a round of product structure change.

800G was once the main product for high-speed interconnection in AI data centers. As GPU cluster scale expands, more and more data needs to be exchanged between servers, traditional network architecture faces bandwidth pressure, and 1.6T optical modules have therefore begun to enter the stage of large-scale adoption.

For optical module manufacturers, this is not just a matter of doubling product speed; it also involves readjustment of chips, optical components, packaging processes, and production yield.

Different technology routes are also advancing simultaneously.

The traditional EML solution uses indium phosphide lasers to achieve optical-electrical conversion, while the silicon photonics solution integrates some optical functions onto silicon-based chips. The two routes have their own characteristics in terms of cost, power consumption, performance, and manufacturing processes.

As 1.6T products ramp up, the application of silicon photonics technology is also increasing.

This makes competition among optical module companies no longer just about who has more orders, but also about who can complete validation of new products faster, who can stably provide high-yield products, and who can enter the next-generation supply system of core customers.

Although Zhongji Innolight, Eoptolink, and TFC Optical all occupy important positions in the optical communications industry chain, their business structures are not the same.

Zhongji Innolight is mainly engaged in the research and development, production, and sales of high-speed optical modules and holds an important position in the supply chains of large overseas cloud service providers.

Eoptolink also benefits from demand for 800G and 1.6T products, and its business growth is closely related to overseas data center construction.

TFC Optical is positioned more in optical components and optical engines, providing relevant core components for optical modules.

The opportunities brought by product upgrades will not be distributed evenly among all companies in the industry chain.

Whoever can complete technical validation first and win more orders from core customers will have the opportunity to gain higher market share. Conversely, if the introduction of new products falls short of expectations, even if overall industry demand grows, companies may still face a growth shortfall.

The market's concerns about the optical module industry have extended from simple demand changes to technology routes and the competitive landscape.

03 The other side of order growth

A change that is easier to observe in financial reports than order-cut rumors is that optical module companies are increasing inventory and prepayments.

As of the end of the third quarter of 2026, Zhongji Innolight's prepayments reached 1.567 billion yuan, up 1068.43% from the end of 2025; inventory reached 23.878 billion yuan, up 88.30% year on year.

Eoptolink's changes are even more pronounced.

As of the end of the third quarter, the company's prepayments reached 1.322 billion yuan, up 7695.01% from the beginning of the year; inventory reached 17.413 billion yuan, up 140.70% from the beginning of the year.

Both companies linked the relevant changes to business growth, order stockpiling, and upstream material procurement.

Amid growing demand for high-speed optical modules and tight supply of core materials, paying suppliers in advance and increasing raw material reserves is a way for companies to ensure delivery.

But this also means that companies need to invest more funds to maintain production.

Optical modules cannot be delivered immediately just because customer orders are received. From optical chips and lasers to DSPs, circuit boards, and optical components, insufficient supply in any key link may affect final shipments.

Eoptolink's third-quarter operating performance has already shown some changes.

The company's revenue in the first three quarters was 31.850 billion yuan, up 92.97% year on year; net profit attributable to shareholders was 11.372 billion yuan, up 79.74% year on year.

However, third-quarter revenue was 10.940 billion yuan, and net profit attributable to shareholders was 3.843 billion yuan, down 12.98% and 19.07% quarter on quarter, respectively.

The company's operating statement and related reports mentioned that floods in Thailand affected factory shipments, and tight raw material supply also had a certain impact on output and delivery.

This does not show that customer demand has already declined, but it does show that order growth and revenue recognition are not always synchronized.

Cash flow is also worth noting.

Eoptolink's net cash flow from operating activities in the first three quarters was 1.464 billion yuan, down 68.42% year on year. The company explained that this was mainly related to payments for materials and increases in various taxes and fees.

Zhongji Innolight's net cash flow from operating activities in the first three quarters was 8.008 billion yuan, up 46.81% year on year, but accounts receivable at the end of the third quarter reached 17.309 billion yuan, up 175.76% from the beginning of the year. The company attributed this to an increase in receivables brought by sales growth.

For fast-growing companies, increases in inventory, prepayments, and accounts receivable are not necessarily a bad thing. The key is whether these occupied funds can ultimately be converted into shipments and cash collection.

If orders are executed smoothly, stocking up in advance can help companies expand revenue; if customers delay project progress, companies will need to bear longer inventory turnover cycles and capital occupation costs.

This is also why the market has begun to pay more attention to the cash flow of optical module companies, not just profit growth.

04 Not just an order issue

There is another change in the optical module industry, coming from the investment return pressure on overseas tech giants themselves.

Amazon's third-quarter AWS revenue grew 39.3%, but free cash flow over the past 12 months has turned to a net outflow of $42.4 billion, and the company expects cash capital expenditure in 2026 to be about $220 billion.

Alphabet expects capital expenditure in 2026 to reach $195 billion to $205 billion, and stated that capital expenditure in 2027 will increase significantly further.

These investments are still staggering, but the capital market has begun to focus more on one question: can revenue created by AI businesses gradually cover the rising infrastructure costs?

If cloud providers continue to expand capital expenditure, optical module companies will still have opportunities to obtain new orders; if the investment returns of some projects fall short of expectations, customers may also rearrange construction sequencing and procurement pace.

Such changes do not necessarily mean the end of the AI infrastructure investment cycle, but they may make performance divergence among suppliers more obvious.

For companies such as Zhongji Innolight and Eoptolink that have already entered the supply chains of core overseas customers, the future test is how to maintain market share and convert product upgrades into profit.

For other companies in the industry chain, competition may be more direct. Customer certification, product performance, delivery capability, and cost control will all affect whether they can enter the next round of procurement lists.

The optical module industry still has clear demand for technological upgrading, but industry growth does not mean every company can grow in sync.

The sharp rebound in the sector on November 11 shows that the market has not given up on the AI computing power industry chain. But the previous adjustment also shows that investors are no longer satisfied with hearing news such as cloud providers increasing capital expenditure and optical module companies having full orders.

Next, what the market needs to see is actual shipment volume, the share of 1.6T products, gross margin, and operating cash flow.

For optical module companies, orders are still important, but whether orders can be delivered on time and how much profit can ultimately be retained is becoming a more important issue than orders themselves.