Pat Gelsinger leads six startups to Taiwan as Playground connects AI computing supply chain 2026.10.27 10:19 Commercial Times, Chang Chia-jui As AI computing demand continues to expand, bottlenecks in power, memory, and data transmission are emerging, and Silicon Valley deep-tech venture capital firm Playground Global is expanding its collaboration with Taiwan's supply chain. Former Intel CEO and Playground partner Pat Gelsinger, together with partner Peter Barrett, brought six portfolio companies to Taiwan on the 27th to present their latest developments, covering AI memory, power management, optical communications, and advanced packaging, seeking to combine Taiwan's manufacturing strength to accelerate commercialization of new technologies. Gelsinger said Taiwan has precision manufacturing, technological innovation, and rapid execution capabilities, making it a key hub for moving forward-looking technologies from R&D to scaled production. Playground is visiting Taiwan for the second consecutive year and will continue to deepen cooperation between its startups and Taiwan's technology ecosystem. During the trip, it also co-hosted an industry exchange dinner with Hon Hai Chairman Young Liu to strengthen long-term cooperation between the two sides. Peter Barrett noted that AI, quantum computing, and cryogenic computing are driving changes in computing architectures, and Taiwan's deep technology and manufacturing foundation will play a key role in the development of next-generation computing. In optical communications, Ayar Labs opened an office in Hsinchu to deepen cooperation with Taiwan manufacturing and packaging partners. It has raised a total of US$650 million this year, with strategic investors including MediaTek and Wiwynn; the funds will be used for product development, validation, and production preparation to advance deployment of co-packaged optics (CPO) technology. Ayar Labs' cooperation with Taiwanese companies has also produced concrete results. Wiwynn, Global Unichip Corp. (GUC), and Ayar Labs demonstrated an integrated optical computing system at this month's OCP Global Summit, combining GUC's ASIC, Ayar Labs' TeraPHY optical engine, and SuperNova light source, and presented real-time monitoring data. The collaboration has advanced to the system integration and functional validation stage, laying the foundation for subsequent rack-level deployment. Power management startup Powerlattice showcased a single-chip integrated voltage regulator (IVR) chiplet that moves the final power delivery stage closer to the processor, integrating the controller, power transistors, and on-chip magnetic thin-film inductors to reduce transmission losses and meet the rapidly changing load demands of AI accelerators. The company said that by more tightly integrating power delivery and computing, its technology can reduce compute power consumption by more than 50%, increasing available compute within existing data center power budgets. In advanced packaging, Syenta showcased for the first time the application of localized electrochemical manufacturing (LEM) technology across an entire wafer, using a stamp to perform localized copper deposition that takes only a few minutes per deposition. This step does not require traditional lithography and targets high-resolution copper interconnect needs. Its platform is also being designed for large-area panel manufacturing, with subsequent efforts to advance process scaling and mass-production applications. Nagare Labs, making its first public appearance, was co-founded by Georgia Tech scholars Suman Datta and Shimeng Yu and targets AI systems' demand for high-density, high-performance memory. The company emphasized that its new technology is compatible with existing silicon processes and does not rely on special materials, and can be combined with Taiwan's semiconductor manufacturing system to advance further development. Another optical interconnect startup, PicoJool, completed a US$27.5 million Series A round, bringing total funding to US$39.5 million, and will expand its micro vertical-cavity surface-emitting laser (microVCSEL) product lineup. In June this year, the company launched 200G VCSEL and MicroVCSEL product families; the 200G products have a bandwidth exceeding 37 GHz, providing a technology option for AI data centers to build 1.6T optical interconnects. Snowcap is targeting longer-term superconducting computing and has demonstrated working superconducting devices on 300 mm wafers manufactured using a foundry's CMOS process, advancing toward high-density, large-scale digital circuit development. The company aims to reduce AI computing energy consumption and hopes in the future to lower the power required for large-scale AI computing from the GW level to the MW level; this remains a technology development goal.

PowerLattice raised $25M led by Playground Global and Celesta. The company’s chiplet cuts compute power needs by >50%, breaking the "AI power wall."