Despite a recent strong performance by U.S. optical communications stocks, Wall Street giant JPMorgan Chase has downgraded its ratings for two key players in the sector: Corning (GLW.US) and Fabrinet (FN.US). However, investors need not be overly concerned about the sector's prospects. As AI infrastructure development shifts from a competition based on single-point computing power to a contest of cluster efficiency, the role of optical modules is becoming increasingly critical. Optical communications stocks, which have repeatedly hit new highs this year, are likely to maintain their leading position in the U.S. stock market's AI hardware race.

JPMorgan downgraded both Corning and Fabrinet from "Overweight" to "Neutral," citing reasons of excessive valuation and limited earnings visibility, respectively. Consequently, while other optical communications stocks saw broad gains on Thursday—Lumentum (LITE.US) rose over 8%, Applied Optoelectronics (AAOI.US) surged more than 10%, and Coherent (COHR.US) increased over 6%—Corning and Fabrinet closed down 1.30% and 1.92%, respectively. Notably, despite the downgrades, JPMorgan significantly raised its price targets for both stocks. The firm increased its target for Corning from $115 to $175, representing approximately 5% upside from Thursday's closing price of $166.08. For Fabrinet, the target was raised from $530 to $700, implying about 4% upside from its closing price of $672.64.

Specifically, JPMorgan's downgrade of Corning was primarily due to its high valuation. An analyst team led by Samik Chatterjee stated in a client report, "We believe the current valuation sets a challenging execution benchmark compared to the earnings expectations embedded by buy-side investors to justify such a high premium." The analysts added, "We think investors are increasingly shifting their focus to the 2028 outlook, incorporating a degree of idealized scenario forecasting across variables such as fiber optic cable/connector pricing and scaling opportunities. This leaves little room for error regarding capacity risks and the linearity of the adoption curve for optical communications scaling, not to mention that approximately 60% of the company's business remains tied to non-optical communications markets."

Despite the rating cut, JPMorgan raised its revenue forecasts for Corning. The bank increased its full-year 2026 revenue estimate from $18.6 billion to $19.0 billion, its 2027 forecast from $20.9 billion to $21.7 billion, and provided its first 2028 revenue projection of $25.1 billion.

JPMorgan's downgrade of Fabrinet was attributed to "increased volatility" in near-term customer capacity ramp-up progress and limited visibility into the ramp-up pace of new future customers. Analysts noted, "We expect the combined effect of these factors to result in near-term upside for the stock falling below current buy-side expectations, although we remain broadly positive on the company's long-term trajectory, particularly considering its ongoing manufacturing footprint expansion." JPMorgan still raised its revenue and earnings per share (EPS) forecasts for Fabrinet. The bank increased its fiscal 2027 revenue estimate from $5.5 billion to $5.9 billion, its fiscal 2028 forecast from $6.3 billion to $7.1 billion, and issued its first fiscal 2029 revenue projection of $8.5 billion. It also raised its fiscal 2027 EPS estimate from $16.65 to $18.00, its fiscal 2028 forecast from $19.40 to $22.00, and provided its first fiscal 2029 EPS estimate of $26.50. Analysts added, "Our updated forecasts are primarily driven by upside in optical communications revenue, including telecom/DCI and datacom businesses, which we expect to achieve compound annual growth rates of approximately 20% and over 30%, respectively, during the forecast period."

Corning and Fabrinet are among the U.S. optical communications stocks that have demonstrated strong performance recently. JPMorgan's action—downgrading the stocks while raising price targets—indirectly confirms the sector's impressive rally (which some investors may view as overextended) and the market's generally optimistic outlook for the future development of the optical communications industry.

U.S. optical communications stocks have continued to hit new highs, exhibiting a strong upward trend. In early 2026, global capital markets remain focused on artificial intelligence. While many investors debate whether AI computing power has already priced in growth expectations for the next three years and media headlines are still dominated by Nvidia's (NVDA.US) next-generation chip launches, the U.S. optical communications sector has quietly embarked on an independent, robust rally. Data shows that leading optical communications companies like Lumentum, Applied Optoelectronics, Coherent, Corning, and Fabrinet have seen their stock prices repeatedly reach record highs this year, significantly outperforming the Nasdaq Composite Index and other AI giants.

The primary catalyst for the recent surge in optical communications stocks was encouraging comments from Lumentum CEO Michael Hurlston. Hurlston stated last Friday, "The capital expenditure scales of several U.S. hyperscale cloud providers are enormous and show no signs of slowing down. Our capacity is increasingly failing to keep up with demand. If current trends continue, within two quarters, our entire production capacity for 2028 will be completely sold out." Lumentum had previously disclosed that its production through the end of 2027 was fully booked. Therefore, Hurlston's latest remarks reinforced market confidence in the sector's vitality: despite disruptions from Middle East conflicts affecting oil markets and the global economy, demand for data center equipment remains robust. Hurlston added, "This situation cannot last forever; it's unrealistic. But for now, this industry cycle has at least another five years of sustained vitality. When we say capacity is sold out, we mean non-cancellable purchase agreements have been signed. That is crucial."

The strength of optical communications stocks is not merely a rotation of market sentiment but the realization of a long-underestimated industrial logic. Against the backdrop of exponential scaling in large AI models, the true bottleneck has shifted from "computing chips" to "computing interconnectivity." As data traffic between GPUs grows geometrically, optical modules are no longer supporting actors but have become the "blood vessels and nerves" of AI infrastructure. AI infrastructure construction has entered a new deep-water zone—transitioning from competition based on single-point computing power to a contest of cluster efficiency.

Over the past two years, the market widely believed that computing power equated to power, and whoever possessed the most GPUs would hold pricing power in the AI era. Computing power suppliers, represented by Nvidia, pushed single-card performance to physical limits. From the A100 to the H100, and now to the next-generation Blackwell platform, market focus has consistently remained on the core metric of "trillions of operations per second." Investors became accustomed to measuring a tech giant's capability by computing density, as if sufficient computing power would naturally lead to intelligence. However, the industrial reality is that when data centers advance to the ten-thousand-card and hundred-thousand-card cluster stage, computing power no longer depends solely on single-card performance but on the efficiency of data exchange within the cluster.

In large-scale distributed training, thousands of GPUs must work collaboratively, and the communication overhead between them becomes the key determinant of overall efficiency. For a ten-thousand-card AI training cluster, the internal east-west traffic far exceeds that of traditional cloud computing workloads. Data synchronization, gradient backpropagation, and parameter updates between GPUs consume bandwidth voraciously. If GPUs are likened to brains, then optical modules are the neural fibers connecting these brains. When the number of brains increases beyond a certain point, if the transmission speed of the neural fibers cannot keep up, even numerous brains cannot form a cohesive force. This is the well-known "communication wall" problem. In early AI training, communication overhead might have accounted for only 10% of the total time, but in the era of hundred-billion or even trillion-parameter models, this proportion can soar to 30% or more. Once communication is hindered, expensive GPUs sit idle, leading to significant computational waste.

Taking hyperscale data centers from companies like Meta, Microsoft, and Google as examples, internal bandwidth demands within a single AI data center have rapidly escalated from 400G to 800G, with scaled deployment of 1.6T optical modules beginning. Industry calculations indicate that in AI training scenarios, every $1 invested in GPUs often requires nearly $0.50 in supporting network and optical interconnect infrastructure. This means optical modules are no longer ancillary costs but a rigid prerequisite for computing power expansion. In this context, the optical module industry is experiencing a structural inflection point, propelling the optical communications sector to become the standout performer in this year's U.S. stock market AI hardware track.

Unlike the cyclical expansion of traditional cloud computing, AI data centers demand high bandwidth, low latency, and low power consumption with an "exponential upgrade" pattern. Copper cable transmission suffers from excessive loss at high speeds and is distance-limited. Optical communication technology is currently the only solution capable of scaling to handle this deluge of traffic. When computing density increases tenfold, the demand for optical interconnectivity does not grow linearly but non-linearly. This is because the larger the cluster scale, the more complex the inter-node connections become, growing quadratically. Consequently, the growth rate of the optical communications industry is expected to surpass that of computing chips themselves in the coming years.

Historically, optical modules were long regarded as "communications cyclical stocks." Demand fluctuated with telecom operators' capital expenditures, profit margins were constrained by fierce price competition, and valuation multiples consistently hovered in the manufacturing range, typically between 15 to 20 times earnings. However, the advent of the AI era has fundamentally altered the industry's pricing logic.

Firstly, the product structure has leaped forward. The technical barriers for 800G and 1.6T optical modules are far higher than those for earlier 100G/200G products. High speeds impose extreme demands on silicon photonics technology, packaging processes, and thermal management capabilities. For instance, silicon photonic solutions require higher integration levels, and Co-Packaged Optics (CPO) technology necessitates deep collaboration with chip manufacturers. Industry concentration is therefore increasing rapidly, with vertically integrated players possessing strong R&D capabilities beginning to command premiums. Lumentum, leveraging its deep expertise in lasers and optical components, has become a core supplier of high-speed modules for multiple cloud providers. Coherent, with years of experience in photonic chips and advanced packaging, boasts significant technical barriers that are difficult to replicate. Corning has established a monopolistic advantage in optical fiber materials and data center cabling, holding sway over the underlying materials at the physical layer.

More critically, the customer base has undergone a fundamental shift. Previously, optical modules primarily served telecom operators. Now, core demand originates from hyperscale cloud service providers and AI companies. These customers place larger volume orders, demand faster product iteration cycles, and, while possessing strong bargaining power, place an extremely high priority on supply stability. Once a supplier enters the core supply chain, it often leads to long-term cooperative relationships, sometimes involving joint development of next-generation products. This binding relationship significantly reduces customer churn and enhances revenue visibility.

Consequently, the market is reappraising optical modules—they are no longer low-margin "contract manufacturing" items but indispensable, critical components within AI infrastructure. Their business model is shifting from "manufacturing-driven" to "technology-driven." Compared to the often 30+ P/E ratios of leading GPU companies, optical communications firms historically traded at multiples below 20. As their fundamentals and industry standing are reassessed, valuation expansion and earnings growth are creating a double tailwind.

Of course, the market is not without skeptics. Opponents argue that AI investment is ultimately subject to cyclical fluctuations. If large model training demand slows, or if improvements in algorithmic efficiency reduce the need for computing power, leading to a downturn in data center capital expenditures, optical module demand could revisit the old pattern of "oversupply — price cuts — profit compression." However, a fundamental distinction in this investment cycle is that AI computing power construction carries strategic attributes, beyond mere commercial expansion. Whether driven by competition among U.S. tech giants or global investments in sovereign computing power, AI infrastructure is becoming a long-term national and corporate strategic investment. This type of investment is rigid and unlikely to halt easily due to short-term economic fluctuations.

Furthermore, technological generational upgrades are far from over. The progression from 800G to 1.6T, and eventually to future 3.2T, each bandwidth upgrade implies replacement demand for existing equipment. For the first time, the optical module industry possesses a structural growth curve akin to the "continuous iteration" seen in semiconductors. In the traditional telecom era, an iteration cycle might last five years. In the AI era, the iteration cycle for optical modules has shortened to approximately 18 months. This means that even without total market growth, structural upgrades can provide sustained revenue momentum.