CONNECT WITH US
Sep 14
CPO nears mass production — next battle about interfaces, yield and maintenance
As AI computing bandwidth keeps rising, optical interconnect architectures are moving beyond pluggable optical modules toward near-packaged optics (NPO) and co-packaged optics (CPO). But large-scale deployment will depend on more than optical-engine performance. Precision alignment, serviceability, manufacturing yield, and whether the supply chains for light sources, packaging, testing, and materials can advance in step are becoming equally critical.

At this year's China International Optoelectronic Expo (CIOE), Sivers Semiconductor brought two products to its booth that trace the trajectory of the optical networking industry: a 1.6T pluggable optical transceiver currently ramping toward mainstream adoption, and a prototype module built for external light source (ELS) architectures, a technology that could define the next generation of data center interconnects.

Taiwan's copper-clad laminate (CCL) makers significantly outpaced printed circuit board (PCB) and IC substrate suppliers in August, as stronger server demand, pricing and product-mix upgrades lifted revenue across the board supply chain.
TSMC's 3nm, 2nm, and CoWoS advanced packaging capacity remained in short supply in late 2026's third quarter, as changes in major customer orders and front-end and back-end supply models prompted the foundry to adjust its advanced capacity allocation. Supply chain sources said Nvidia and Apple continue to dominate TSMC's advanced process and packaging resources, while AWS has recently expanded AI chip tape-outs, giving Annapurna more 3nm capacity.
Chinese-made semiconductor equipment accounted for roughly 35% of tools installed at fabs in China in 2025, up from about 25% a year earlier, as domestic suppliers gain ground in key process-equipment markets.
Last week, from September 9 to 11, thousands of vendors gathered for an enormous trade show in the Chinese tech hub of Shenzhen. At the China International Optoelectronic Expo (CIOE), DIGITIMES reporters got a peek into the trends going on within this booming industry.
XPeng's Turing chip expands beyond cars
Sep 15, 10:25
XPeng Motors is extending its self-developed Turing AI chip from intelligent electric vehicles (EVs) into humanoid robots, broadening the processor's role in the company's Physical AI strategy. Originally designed for intelligent driving, the chip is now also used in XPeng's IRON humanoid robot. On September 8, the company launched its robot production line, with the first IRON completing automated final assembly.
US President Donald Trump joined Nvidia CEO Jensen Huang by phone on stage, using the call to place data center construction at the center of his economic agenda and to dismiss warnings about machine intelligence.

Chinese RISC-V chip designer Nuclei System Technology has formally begun its IPO process, seeking fresh capital while heavy R&D spending continues to outweigh revenue despite gross margins above 80%.

In 2026, major smartphone system-on-chip (SoC) vendors began moving selected flagship processors to 2nm technology, seeking not only to secure sufficient capacity but also to deliver meaningful upgrades in technical specifications.

Nvidia, MediaTek deepen 20-year chip ties
Sep 15, 07:38

Nvidia and MediaTek have recently drawn market attention after Nvidia invested US$3.5 billion in MediaTek convertible bonds, expanding cooperation into custom AI chips, AI PCs, automotive, and data centers. But their overlap began more than 20 years ago, during competition in the PC chipset market, when MediaTek, ALi, ULi, and Nvidia were already connected.

PwC and Oxford Economics jointly released their Global Data Centre Outlook 2026–50 in early September, projecting cumulative global data center capex of US$31.6 trillion by 2050 under the baseline scenario. If AI adoption accelerates, investment could approach US$50 trillion under an upside scenario.