A new US government-linked report says American tariff revenue is being drained by tens of billions of dollars a year through illegal transshipment, with exporters routing China-origin goods through more than 40 third countries — many of them in Asia — to dodge higher US duties. The report, titled "The Great Transshipment Scam," traces the practice to Chinese exporters' response to the Section 301 tariffs imposed in 2018 and estimates current annual illegal transshipment flows at US$40 billion to US$303 billion, depending on methodology.
As physical AI and robotics draw unprecedented market attention, a key bottleneck is clogging the deployment pipeline: quality, real-world physical data. To move applications from the lab to real-world deployment, companies are getting creative in overcoming this obstacle while balancing affordability, stability, and data volumes.
China is accelerating construction of its national computing network, and a growing share of new capacity is likely to use supernodes. With domestic AI accelerators still trailing leading global chips in single-card performance, Chinese vendors are increasingly relying on larger card counts, high-speed interconnects and system-level optimisation to close the gap.
As AI chip value and integration complexity keep rising, King Yuan Electronics (KYEC) President Gauss Chang said semiconductor testing is shifting from being considered "part of the supply chain" to a "part of the process." He said the supply chain is also moving toward a new collaboration model built around four integrated elements: equipment, accessories, testing, and products.
High-bandwidth memory (HBM), a core component in AI semiconductors, has drawn fresh scrutiny after former Intel CEO Pat Gelsinger called it "lousy" at a recent AI summit in Paris. SK Hynix, the market leader in this type of memory, also said it is not the final answer to AI memory bottlenecks, underscoring the industry's push to find a next-generation memory architecture beyond HBM.
Asia Vital Components expects global AI server demand to strengthen in the second half, with liquid cooling adoption becoming more common — a shift carrying implications for data center operators and suppliers worldwide. The company sees higher shipments of ASIC chips and Nvidia's Vera Rubin platform driving growth, and estimates that liquid-cooling penetration in AI servers will reach 50% by 2027.
Connector maker Lotes is deepening its server business, with its server quick disconnect (QD) line set to enter mass production in July, and expansion is expected to accelerate at the same time. A new-generation server platform is expected to begin small-volume shipments in the fourth quarter of 2026, followed by a larger-scale ramp-up in early 2027, making both key growth drivers to watch in the coming quarters.
As physical AI and robotics draw unprecedented market attention, a key bottleneck is clogging the deployment pipeline: quality, real-world physical data. To move applications from the lab to real-world deployment, companies are getting creative in overcoming this obstacle while balancing affordability, stability, and data volumes.
