AI computing power competition enters a new stage! Interconnection stabilizes its position in the "C position." JP Morgan downgrades Corning, Fabrinet maintains bullish trend in the optical communication stock market.

Recently, as US stocks in the optical communication sector have shown a strong performance, Wall Street giant JP Morgan has lowered the ratings of Corning Inc (GLW.US) and Fabrinet (FN.US), two optical communication stocks. However, investors need not worry about the prospects of optical communication stocks. Against the backdrop of AI infrastructure construction shifting from single-point computing power competition to cluster efficiency, the role of optical modules is becoming increasingly crucial, and the high-flying optical communication stocks this year may continue to dominate the US stock AI hardware track. JP Morgan Downgrades Corning Inc and Fabrinet Ratings, While Significantly Raises Price Targets JP Morgan lowered the ratings of Corning Inc and Fabrinet from "neutral" to "hold," citing overvaluation and limited earnings visibility, respectively. Despite this, other optical communication stocks saw gains on Thursday Lumentum (LITE.US) rose over 8%, Applied Optoelectronics (AAOI.US) rose over 10%, and Coherent (COHR.US) rose over 6%, while Corning Inc and Fabrinet closed down 1.30% and 1.92%, respectively. It is worth noting that JP Morgan significantly raised the price targets while downgrading the ratings of these two stocks. The firm raised the price target for Corning Inc from $115 to $175, representing about 5% upside from the Thursday closing price of $166.08; and raised the price target for Fabrinet from $530 to $700, representing about 4% upside from the Thursday closing price of $672.64. Specifically, the main reason for JP Morgan's downgrade of Corning Inc was its overvaluation. The analyst team led by Samik Chatterjee stated in a client report, "We believe the valuation has formed a challenging threshold compared to the profit expectations embedded by current buyers to justify such a high premium." The analysts added, "Investors are increasingly focusing on the prospects for 2028 and including some degree of idealized scenario predictions, taking into account multiple variables such as fiber optic cable/connectors pricing and opportunities for scale expansion. This leaves virtually no room for error in production capacity risks and the linearity of optical communication scale adoption process, not to mention the fact that the company still has about 60% of its business related to non-optical communication markets." Despite the downgrade, JP Morgan raised its revenue forecasts for Corning Inc. The firm increased its full-year 2026 revenue forecast from $18.6 billion to $19 billion, raised the 2027 revenue forecast from $20.9 billion to $21.7 billion, and provided a first-ever revenue forecast for 2028 of $25.1 billion. The reason for JP Morgan's downgrade of Fabrinet was the increased fluctuation in short-term customer capacity scale-up progress, and limited visibility on the pace of future customer capacity scale-up. The analysts noted, "We expect the combined impact of these factors to lead to lower short-term upside than current buyer expectations, although we overall remain positive on the company's long-term growth trajectory, especially considering the manufacturing layout expansion the company is currently pursuing." JP Morgan also raised its revenue and earnings forecasts for Fabrinet. The firm increased its fiscal year 2027 revenue forecast from $5.5 billion to $5.9 billion, raised the fiscal year 2028 forecast from $6.3 billion to $7.1 billion, and provided a first-ever revenue forecast for fiscal year 2029 of $8.5 billion. JP Morgan also raised its fiscal year 2027 earnings per share forecast from $16.65 to $18, raised the fiscal year 2028 earnings per share forecast from $19.40 to $22, and provided a first-ever earnings per share forecast for fiscal year 2029 of $26.50. The analysts added, "Our updated forecasts are mainly driven by the revenue upside potential of the optical communication business, including telecom/DCI (data center interconnection) and data communication (Datacom) businesses, with compound annual growth rates of about 20% and over 30%, respectively, during the forecast period." Corning Inc and Fabrinet are among the US optical communication stocks that have shown a strong performance recently. JP Morgan's downgrade of the ratings for these two stocks, along with the significant increase in price targets, indirectly confirms the recent strong performance of optical communication stocks (perhaps seen as overbought by some investors) and the overall optimistic expectations for the future development of the optical communication industry in the market. Continued Highs! US Optical Communication Stocks Show Strong Performance At the beginning of 2026, the global capital markets still focused on artificial intelligence. While most investors were still debating whether AI computing power had already priced in the growth expectations for the next three years, media headlines were dominated by NVIDIA Corporation's (NVDA.US) new generation chip launch, the US optical communication sector quietly saw a wave of independent strong performance. Data shows that the stock prices of leading optical communication companies like Lumentum, Applied Optoelectronics, Coherent, Corning Inc, and Fabrinet have continued to hit historical highs this year, with their performance clearly outperforming the Nasdaq Composite index and other AI giants. The main reason for the recent surge in optical communication stocks was a statement made by Lumentum CEO Michael Hurlston. Hurlston stated last Friday, "The capital spending scale of several super-scale cloud giants in the United States is extremely large, and it seems that there is no sign of slowing down. Our production capacity is increasingly unable to keep up with demand. If the current trend continues, in another two quarters, we will be completely sold out of capacity for the full year 2028." Lumentum had previously disclosed that its production capacity had been fully booked by the end of 2027. Therefore, Hurlston's latest statement consolidated market confidence in the optimistic outlook for the optical communication industry: even against the backdrop of disruptions in the Middle East affecting the oil market and the global economy, the demand for data center equipment remains strong. Hurlston added, "This situation cannot last forever, it is unrealistic. But for now, this round of industry cycle can at least sustain for about 5 years. When we say the capacity is sold out, it means that we have already signed non-cancelable order agreements. This is crucial." Paradigm Shift From "Chips" to "Interconnection" The strong performance of optical communication stocks is not simply a market sentiment rotation, but the beginning of a long-underestimated industrial logic officially materializing. Against the backdrop of exponential scale expansion, the real bottleneck has shifted from "computing chips" to "computing interconnection." As the data flow between GPUs grows exponentially, optical modules are no longer just supporting actors but have become the "blood vessels and nerves" of AI infrastructure. The construction of AI infrastructure has entered a new deep zone moving from the competition of single-point computing power to the game of cluster efficiency. Over the past two years, the market has generally believed that computing power means power, that whoever has the most GPUs has the pricing power in the AI era. Leading suppliers of computing power such as NVIDIA Corporation have pushed single-card performance to physical limits. From A100 to H100, and then to the next generation Blackwell platform, the market's focus has always been on the core metric of "how many trillion calculations per second." Investors are accustomed to measuring a tech giant's capability based on computing power density, as if as long as the computing power is strong enough, intelligence will naturally emerge. However, the real industrial reality is that as data centers enter the tens of thousands and hundreds of thousands of card cluster stage, computing power is no longer determined by single-card performance but by the data exchange efficiency within the cluster. In large-scale distributed training, thousands of GPUs need to work together, and the communication overhead between them becomes the key to determining overall efficiency. In a ten thousand card-level AI training cluster, internal east-west traffic far exceeds traditional cloud computing loads. Data synchronization, gradient backpropagation, and parameter updates between GPUs consume bandwidth exponentially. If GPUs are likened to the brain, then optical modules are the neural fibers connecting these brains. When the number of brains increases to a certain extent, if the transmission speed of the neural fibers cannot keep up, no matter how many brains there are, they cannot form a united force. This is the famous "communication wall" problem. In early AI training, communication overhead may only account for 10% of total time, but in the era of models with hundreds of billions or even trillions of parameters, this percentage could be as high as 30% or even more. Once communication is hindered, expensive GPUs will be idle, leading to significant waste of computing power. Taking Meta, Microsoft Corporation, Alphabet Inc. Class C, and other super-scale data centers as examples, the internal bandwidth demand of a single AI data center has rapidly increased from 400G to 800G and has begun to deploy 1.6T optical modules on a large scale. Industry estimates show that in an AI training scenario, for every $1 invested in GPUs, there is a need for nearly $0.5 investment in network and optical interconnection infrastructure. This means that optical modules are no longer just auxiliary costs but have become the rigid prerequisite for computing power expansion. In this context, the optical module industry has reached a structural inflection point, driving the optical communication sector to become the "best-looking" sector in the US AI hardware track this year. Different from the cyclical expansion of traditional cloud computing, the demand for high bandwidth, low latency, and low power consumption in AI data centers is experiencing "exponential growth." Copper cable transmission suffers from high losses at high speeds, limited distances. Optimal communication technology is currently the only scalable solution capable of handling the peak flow of traffic. As computing power density increases tenfold, the demand for optical interconnection increases not linearly, but exponentially. This is because as the cluster size grows, the complexity of connections between nodes increases exponentially. Therefore, the growth rate of the optical communication industry is expected to surpass that of computing chips themselves in the coming years. From a cyclic product to a strategic asset, the revaluation of the optical module business model Historically, optical modules have long been seen as "communication cyclic stocks." Demand fluctuates with the capital expenditures of telecom operators, and profit margins are limited by intense price competition. Valuations are typically in the range of 15 to 20 times earnings in the manufacturing sector. However, the arrival of the AI era has fundamentally changed the pricing logic of this industry. Firstly, the product structure has undergone a leap. The technological requirements for 800G and 1.6T optical modules far exceed those of early 100G/200G products. High-speed rates require high demands on silicon photonics technology, packaging processes, and thermal management capabilities. For example, silicon photonics solutions require higher integration, while CPO (co-packaged optics) technology requires deep collaboration with chip manufacturers. As a result, the industry concentration has rapidly increased, with companies possessing vertical integration and R&D capabilities receiving premiums. Lumentum, with its deep accumulation in lasers and optical components, has become a core supplier of high-speed modules for several cloud giants. Coherent, with its long-term focus on photonics chips and advanced packaging, has significant technological barriers and is difficult to be easily replaced. Corning Inc, on the other hand, has formed a monopoly-level advantage in optical fiber materials and data center wiring, holding the dominant power in the physical layer materials. More crucially, the customer structure has fundamentally changed. In the past, optical modules mainly targeted telecom operators, but now the core demand comes from super-scale cloud service providers and AI companies. These customers have larger order volumes and faster product iterations. While they have strong bargaining power, they have high requirements for supply stability. Once they enter the core supply chain, they often form long-term cooperation relationships, and even jointly develop next-generation products. This binding relationship significantly reduces the customer churn rate and enhances revenue visibility. As a result, the market has begun to reprice optical modules are no longer low-margin "contracted parts," but indispensable key components in AI infrastructure, shifting their business model from "manufacturing-driven" to "technology-driven." In comparison to GPU leaders with price-earnings ratios above 30 times, optical communication companies have long been stuck with price-earnings ratios below 20 times. When fundamentals and industry positions are reassessed, valuation expansion and profit growth form a double whammy. Of course, the market is not without skeptics. Opponents argue that AI investments still have cyclic fluctuations. Once the demand for large model training slows down, or if improved algorithm efficiency leads to decreased demand for computing power, data center capital expenditures may fall, and the demand for optical modules may repeat the old cycle of "excess - price reduction - profit compression." However, the fundamental difference between this round of investment and the past is that AI infrastructure construction has strategic attributes, rather than just commercial expansion. Whether it is the competition between US tech giants or the global investment in sovereign computing power, AI infrastructure is becoming a long-term national and corporate strategic investment. This kind of investment is rigid and will not easily stop due to short-term economic fluctuations. More importantly, technological intergenerational upgrades have not ended. From 800G to 1.6T, and then to future 3.2T, each bandwidth upgrade means replacement demand for existing equipment. For the first time, the optical module industry has a structural growth curve similar to that of semiconductors with "continuous iteration." In the traditional telecom era, a iteration cycle may take five years. In the AI era, the iteration cycle of optical modules is shortened to around 18 months. This means that even if the total volume does not increase, structural upgrades can still bring continuous revenue momentum.