The more amazing AI is, the faster the data needs to run! Mywell (MRVL.US) opens up a new space for AI infrastructure business with 2 nanometer optical interconnection

The Zhitong Finance App learned that Mewell Technology (MRVL.US), which focuses on AI ASICs and data center optical interconnect chips, is actively showcasing a new generation of 2nm optical interconnect technology to global investors at the 2026 European Optical Communications Conference (ECOC), and its stock trend is receiving a lot of attention as a result. This presentation comes at a time when large AI data center operators are demanding faster connections and more efficient data transmission between AI accelerators, CPU processors, HBM memory, and even data center NAND, while avoiding a sharp increase in connection and transmission power consumption.

Based on the “Jevens Paradox,” the overall demand for AI computing resources will continue to expand in the future due to declining computational power costs on the inference side. Looking at the inference system architecture, more complex tasks often include longer context, multiple rounds of model calls, tool execution, and result verification: Prefill (Prefill) requires processing input, decoding (Decode) continuously generates output, and key value cache (KV Cache) takes up more memory as context and concurrency scale expand, and computational throughput, memory bandwidth, and capacity need to be increased collaboratively. The core inference for the AI computing power industry is that GPUs and TPUs undertake model calculation, CPU tool execution and task orchestration, and HBM, server DRAM, storage and high-performance network infrastructure, and data center optical interconnect devices all support efficient data transportation and state management; ultimately, a complete set of AI computing power server clusters will be needed to deliver sustainable operation services.

The recent stock prices of data center optical product suppliers such as Mywell and Lumentum have continued to be strong. It is also due to market news that the “AI chip hegemon” Nvidia (NVDA.US) is adjusting the next-generation AI rack computing power system architecture. That is, the next generation AI computing power cluster Rubin Ultra may reduce the HBM configuration of each rack and instead use optical interconnection technology architectures to connect multiple AI racks. As the prices of HBM/DRAM memory systems and NAND memory chip products have continued to soar in recent years, the focus of AI computing power infrastructure investment seems to be shifting from GPU/TPU single-point performance and HBM capacity to overall data center architecture efficiency. Among them, high-speed optical interconnection is probably the most critical direction.

This ECOC presentation further strengthened Mywell's optical interconnection growth path: the 2 nm process upgrade and a single channel 400G optical connection aimed at higher bandwidth density and lower energy consumption per bit. Here, 2 nm mainly refers to the manufacturing process of electronic chips such as DSP; calculated at a nominal rate, eight 400G channels can support 3.2T connections. Another 102.4 Tbps co-packaged optical (CPO) platform using single-channel 200G silicon optical technology shortens high-loss electrical channels by making optical connections closer to computing and switching chips.

From the perspective of large-scale data center construction, this helps to carry more communication traffic within the established power supply, cooling, and rack space, and improve the utilization rate of expensive accelerators. For Maywell, its commercial value will be reflected through customer design introduction, high-speed interconnect product upgrades, and subsequent mass production shipments, and is expected to become a new catalyst for improving revenue and profit expectations.

OpenAI Astra's improved capabilities in software engineering, computer operation and multi-step tasks, and suspected Gemini 4 Pro test news — some market news shows that Google is suspected of quietly launching the flagship model Gemini 4 Pro under the name “gemini-3.8-flash” on the AI arena Arena, and is continuously strengthening the market's expectations for a new round of AI application expansion.

Judging from the underlying data center engineering logic, when coding, research, and multi-modal tasks shift to intelligent execution on a large scale, application penetration and concurrency scale will expand, which will promote cross-chip collaboration, context cache access, and cross-rack data exchange in distributed inference, and place higher demands on network throughput, latency, and energy efficiency. This time, at ECOC, Maywell demonstrated single-channel 400G optical interconnection technology based on the 2 nm process, which supports the network upgrade to 3.2T. It is by increasing bandwidth density and reducing power consumption per bit, it helps AI clusters reduce data transmission bottlenecks and improve effective computing power. This is also the consistent logic that Wall Street analysts are bullish on Maywell — that is, application expansion brought about by breakthroughs in AI capabilities, which is expected to simultaneously drive demand for customized ASICs and a new round of high-speed optical interconnection, enabling Maywell to gain business growth at both ends of computing and connectivity.

Maywell enhances AI data center high-speed connection layout with 2nm optical interconnection technology—ECOC highlights at a glance

As AI clusters require faster connections as data center power consumption grows at varying rates, Maywell is driving the evolution of its optical communications product portfolio to more advanced process nodes.

Maywell Technology (MRVL) is showcasing a new generation of 2nm optical interconnect technology at the 2026 European Optical Communications Conference (ECOC), and its stock is drawing attention as a result. This presentation comes at a time when AI data centers require faster connections between accelerators, processors, and memory while avoiding significant increases in power consumption.

ECOC is Europe's largest conference and exhibition dedicated to optical fiber communication and optical networks. Currently being held in Malaga, Spain, the conference attracted more than 340 exhibitors, and featured a special conference on AI infrastructure and advanced technology demonstrations.

Maywell shares rose more than 3% in pre-market trading in US stocks on Monday.

According to the company, Mywell's demonstration included the first single-channel 400G optical PAM4 solution using a 2-nm process, a 2-nm 800G ZR/ZR+ pluggable module with safety features, and a 2-nm 1.6T ZR and lightweight coherent technology.

These products are designed to increase bandwidth density and reduce energy consumption per bit as AI infrastructure grows.

The company said its advanced optical and silicon chip designs utilize leading-edge manufacturing processes to achieve improvements in performance, bandwidth density, and power consumption, which are critical to efficiently scaling AI infrastructure.

Faster AI chips can only work if massive amounts of data can be transmitted between chips fast enough. This is why the AI data center optical interconnect/optical communication architecture led by Maywell Technology is moving from 1.6T connections to 3.2T and higher speeds, increasing the demand for faster and more energy-efficient optical links. Mywell's single-channel 400G demo is designed to support this transformation, and its 1.6T technology targets high-bandwidth connections within and between data center facilities.

The company is also showcasing its 102.4 terabits per second co-packaged optical (or CPO) platform. As traditional copper connections face distance and power consumption limitations, the platform integrates optical connections more closely with AI accelerators and switching chips.

At ECOC, Maywell plans to present 38 technology demonstrations, covering 224G serializer/deserializer (SerDes), 1.6T optical connectivity, coherent interconnect, and optical memory interconnect networks, further advancing its overall layout to solve the connectivity bottlenecks caused by the expansion of AI systems.

Additionally, Lumentum (LITE.US), Qualcomm (QCOM.US), and Corning (GLW.US) will demonstrate a high-density die-to-chip optical interconnect solution for AI vertical scaling systems at ECOC.

The solution combines Qualcomm's interface technology, Lumentum's 1060nm vertical cavity surface-emitting laser (VCSEL) optical platform, and Corning's multi-mode optical fiber to provide a higher bandwidth and lower power consumption connection as AI systems break through the actual limits of electrical interconnection.

However, on the Stocktwits platform, retail sentiment surrounding Myswell shares is still partial. Since this year, the stock has accumulated a cumulative increase of more than 173%. In contrast, iShares Semiconductor ETF (SOXX), which holds Mewell shares, rose nearly 70% over the same period.

The more capable AI agents are, the more unblocking the data center network and high-speed data transmission!

The core logic behind Maywell's sharp rise in stock prices this year is to open up revenue space at the same time as customizing the AI chip volume and upgrading high-speed optical interconnection. Cloud vendors are expanding the deployment of self-developed accelerators, bringing growth to their customized ASIC (ASIC) business; the expansion of accelerator clusters is also driving growth in demand for optical communication digital signal processors (DSPs), high-speed interfaces, and switching chips. Financial reports already reflect this trend: for the second quarter of fiscal year 2027 ending August 1, 2026, data center revenue reached US$2.171.5 billion, an increase of 46% over the previous year, accounting for 79% of total revenue.

In September, Maywell's management further projected revenue of US$12 billion and US$18 billion for the 2027 and 2028 fiscal years, respectively, and expected revenue from the customized business to double in the 2028 fiscal year. Judging from investment logic, these growth expectations provide a strong and specific basis for the market to reevaluate its AI computing power infrastructure construction business layout. This ECOC presentation can be described as further strengthening the strong growth path of optical communication product upgrades and volume.

The joint demonstration by Qualcomm, Lumentum, and Corning also confirmed the trend of extending high-speed optical interconnection technology in data centers to the vicinity of core AI computing chips from another technology route: the three companies combined D2D interface, 1060 nanometer wavelength VCSEL, and high-density multi-mode optical fiber to expand the connection between bare chips to tens of meters for future CPO, near-package optics (NPO), and uCie interfaces. This solution targets vertical scaling (scale-up), that is, expanding the computational domain of tightly coupled accelerators. The investment implication behind it is that the demand for optical connectivity is expanding further from traditional network ports to the vicinity of computing chips and advanced packaging; Mewell also covers custom chips, optical DSP, CPO, and optical memory interconnections, so it has the opportunity to participate in value creation in more systems. ASIC expands the scale of computing, and optical interconnection improves operating efficiency after scale expansion. Together, the two enhance its long-term growth space.

The effective throughput of AI systems is constrained by computational power, memory bandwidth, and cross-chip communication efficiency: distributed training requires synchronous gradients and exchange of intermediate results, mixed expert models (MoE) require routing data between different experts, and long context reasoning increases the capacity and access requirements of the KV Cache (KV Cache). Customized ASICs can optimize chip design around the customer's operators, numerical accuracy, storage levels, and data flow to improve energy efficiency for specific workloads; optical interconnections ease the distance, bandwidth density, and power consumption constraints faced by high-speed data transmission, and reduce the time the accelerator waits for data.

As a result, the expansion of global AI deployment not only increases the demand for computing chips, but also increases the requirements for connectivity and memory systems. Mywell's unique advantage is that it also participates in “improving unit computing efficiency” and “improving cluster collaboration efficiency” to help customers obtain more effective computing power and reduce processing costs for each token within limited power and capital budgets.

Wall Street financial giant Bank of America maintains a target price of $365 for Mywell, and places particular emphasis on expanding the revenue share of each AI system through customized AI accelerators (or AI ASICs) and supporting optical interconnect chips that focus on the inference needs of heavyweight AI agents. The highest target price is $400. From KeyBanc analyst John Vinh, the target price was raised from $385 to $400, and the “gain” rating was reiterated on August 28. The target price of $400 corresponds to a potential upside of around 63.77%. Analyst Vinh sees Maywell as one of the main long-term beneficiaries of AI network expansion. His team also previously emphasized that the growth opportunities for data center optical interconnection services are becoming stronger and more sustainable — this bullish logic complements ASIC: cloud giants are increasing the deployment of customized self-developed AI chips, and they also need to expand high-speed optical connectors connecting these chips. As a result, Maywell can participate in capital expenditure growth at both ends of computing and interconnection.