Micas는 AI 네트워킹 광학 및 실리콘 부품 생산 증대를 위해 말레이시아 제조 시설을 확장합니다.

Micas Networks has expanded its company-operated manufacturing in Malaysia to scale production of co-packaged optics systems for AI networking. The San Jose-based company's expansion adds clean-room manufacturing, optical integration, advanced system assembly, and automated testing and validation for complex Ethernet platforms used in AI infrastructure.

Micas described itself as the only AI infrastructure company in its market with its own factory and dedicated clean-room production environment. The company said direct control over manufacturing is intended to improve quality oversight, product introduction, supply chain management, and delivery scheduling.

The investment focuses on equipment and processes needed to produce co-packaged optics, or CPO, systems. These designs place optical engines alongside the switch application-specific integrated circuit—an approach the industry has pursued as data centre operators seek to manage bandwidth growth and power use in large AI clusters.

At the centre of Micas' current CPO offering is the M2-W6940-128X1-FR4, a 51.2 Tbps Ethernet switch based on Broadcom's Tomahawk 5 Bailly platform. The switch provides 128 ports of 400G FR4 connectivity and is available through Supermicro as part of validated AI infrastructure configurations. Micas said the product was developed with Broadcom and has already entered volume production, which the company presented as evidence that CPO hardware is moving into commercial deployment rather than remaining a laboratory concept.

Micas' operating model splits responsibilities between its California headquarters and Malaysia facility. Product strategy, ecosystem work, and customer engagement run from San Jose, while manufacturing and supply chain execution are handled in Malaysia. This approach differs from many networking vendors, which rely on contract manufacturers for assembly and production. Owning production can give suppliers tighter oversight of system configuration, factory testing, and component handling—particularly for optical systems that require controlled assembly conditions.

Performance data from a validated design involving Supermicro servers, AMD Instinct MI355X graphics processors, and AMD Pensando Pollara 400 AI network interface cards showed Micas' CPO switch setup delivering approximately 38.28% lower switch power consumption than a pluggable-optics setup. The same architecture reported lower latency and more than 4 million cumulative 400G-equivalent port-hours of link-flap-free operation in ongoing company testing. Micas noted these figures were specific to the validated configuration used in the tests.

Suppliers across the AI networking market are working to meet rising bandwidth demands inside large training and inference clusters. Optical interconnect design has become increasingly important as operators seek ways to connect more accelerators while controlling energy use and physical complexity in the network fabric. The company's portfolio covers Ethernet systems from 1G to 800G and includes support for open networking operating systems such as SONiC.

A broader question for networking suppliers is whether they can turn newer optical designs into systems available in volume, rather than limited pilot deployments. CPO has drawn interest for several years, but commercial availability, field validation, and repeatable manufacturing remain key hurdles.

Andrew Qu, Chief Executive Officer of Micas Networks, said the company sees manufacturing control as central to that transition. "Micas is an American company built to turn advanced networking innovation into infrastructure customers can deploy," Qu said. He described the company's operating model further: "Our San Jose headquarters anchors product strategy, ecosystem collaboration, and customer engagement, while our company-operated manufacturing in Malaysia gives us direct control over quality, new-product introduction, supply chain and delivery. That combination is a strategic advantage as AI infrastructure demands higher bandwidth, greater power efficiency and more tightly integrated optical systems."

Qu also linked the expansion to a broader industrial argument about how AI infrastructure expertise will be judged. "American leadership in AI infrastructure will be measured by the ability to execute, not only to invent," he said. "Micas is building both: advanced networking technology led from San Jose and the directly controlled manufacturing capability required to deliver it globally."