Highlights from the world’s leading technology trade show
Computex is one of the largest computer and technology trade shows in the world. Held annually for over 40 years, it serves as the global epicenter for the computer hardware industry and supply chain and is renowned as the global stage for tech giants to announce their most exciting launches and roadmaps for the year ahead.
Members of the Third Bridge team headed to Taipei to immerse themselves in the latest intriguing and innovative industry developments in AI computing & infrastructure, robotics & physical AI and next-gen tech. This included hosting a roundtable discussion on Taiwan's advanced packaging and Copper Clad Laminate (CCL) industry updates with sector experts, alongside conducting on-the-ground channel checks across the show floor to track emerging semiconductor trends.
Below are our some of the key observations from the week:
NVIDIA RTX Spark, Vera CPU & ODM Full-Stack Evolution
NVIDIA released its first consumer Windows SoC, the RTX Spark (N1X), built on TSMC 3nm in partnership with MediaTek, integrating 20 Arm Grace CPU cores and a Blackwell GPU with up to 1 PFLOP FP4 AI performance and 128GB unified memory. NVIDIA revealed a multi-generation RTX Spark roadmap, with future iterations based on Vera Rubin (LPDDR6) and Rosa Feynman architectures, signaling a long-term commitment to the Arm-based consumer PC platform.
Alongside the PC announcements, NVIDIA formally unveiled the Vera CPU rack as a dedicated tier within its AI factory architecture, purpose-built for agentic AI orchestration. Key performance claims include 1.8x faster agentic sandbox task completion versus leading x86 CPUs under peak load, 3x higher memory bandwidth per core, and a full rack supporting over 22,500 concurrent sandboxes at 2x the performance-per-watt of an equivalent x86 rack. NVIDIA, Arm, and Qualcomm all emphasized the strategic importance of Arm-based server CPUs in their respective keynote sections, reinforcing a structural challenge to x86's dominance in data center CPU procurement.
On the infrastructure side, NVIDIA confirmed Vera Rubin NVL72 is now in full production, with single rack assembly time reduced to approximately 5 minutes. Most major ODMs — including Foxconn, Wiwynn and Compal— simultaneously showcased both the NVIDIA Vera Rubin NVL72 rack and AMD Instinct MI355X-based platforms side by side, marking the most explicit public demonstration of a dual-GPU vendor strategy across the AI server supply chain to date. ODMs also demonstrated integrated capabilities spanning liquid cooling (DLC, MCL), high-voltage power delivery, and optical interconnects, underscoring their evolution from contract manufacturers to full-stack AI infrastructure solution providers.
Want to learn more? Access relevant Third Bridge transcripts and register interest in upcoming calls:
• 2026/06/17 - AI Server ODMs – 2026 Rubin GPU Demand Outlook, Verification Progress & Competitive Dynamics (Conducted in Mandarin)
• 2026/05/28 - Data Centre CPUs – 2026 Demand Outlook & Agentic Shift
800V Power Rack Acceleration
Delta Electronics debuted its 800VDC In-Row Power Solution with integrated BBU modules, a 2.4MW HVDC CDU, and a prefabricated AI Modular Data Center. Lite-On made its first public showing of an 800VDC liquid-cooled power rack, targeting Q4 mass production, alongside a 110kW Power Shelf supporting Rubin NVL72 already entering volume production. Wiwynn jointly demonstrated a full rack-level 800VDC solution with Delta and TE Connectivity — spanning the power rack, liquid-cooled busbar, and 800V-to-50V PDB.
Want to learn more? Access relevant Third Bridge transcripts and register interest in upcoming calls:
• 2026/06/11 - Delta Electronics – HVDC Power Rack Demand Outlook & Supply Chain Dynamics (Conducted in Mandarin)
• 2026/03/10 - Delta Electronics – HVDC Power Rack Validation Progress & SST Commercialisation (Conducted in Mandarin)
CPO Acceleration
In partnership with Ayar Labs and GUC (Global Unichip Corporation), Wiwynn showcased its first scale-up CPO (Co-Packaged Optics) solution and targets mass production in 2028. Rather than being locked into a single chip supplier, Wiwynn’s system design accommodates a broad ecosystem of compute options — spanning NVIDIA GPUs, AMD Instinct accelerators, custom hyperscaler ASICs, and Arm-based CPUs. This open platform strategy reflects Wiwynn’s deliberate effort to serve hyperscale customers who are increasingly pursuing procurement diversification across chip vendors, reducing single-source dependency on NVIDIA and gaining negotiating leverage across their AI infrastructure buildout.
MediaTek showcased a Co-Packaged Optics (CPO) evaluation board featuring an integrated EIC (Electronic Integrated Circuit) + PIC (Photonic Integrated Circuit) co-packaged module, achieving 400Gbps per fiber optical bandwidth. The demonstration marks MediaTek’s entry into the CPO silicon supply chain, leveraging its TSMC process relationships to develop a vertically integrated EIC+PIC solution.
BizLink demonstrated a complete rack-scale CPO interconnect solution, showcasing a full AI server rack integrating Switch, Server, (RMC) Rack Management Controller, Shuffle Box, (BBU) Battery Backup Shelf, Power for Server, (PDU) Power Shelf, High Density AI Server, and Application Demo Server — all within a unified CPO-enabled chassis. On the optical interconnect side, BizLink presented CPO and transceiver module solutions supporting 1.6T and beyond bandwidth requirements, integrating 1×64 optical switch, SN-MT/MMC multi-fiber connectors, FAU (Fiber Array Units), fiber crossover optical paths, and silicon photonics IC integration. Based on our channel checks at the event, BizLink is currently shipping FAU (Fiber Array Units) and Shuffle Box to key customers including Senko and NVIDIA.
Taken together, Computex 2026 marked a clear inflection point for the CPO ecosystem— transitioning from concept demonstrations to early customer qualification and initial production ramp. With industry consensus converging on a 2027–2028 timeline for CPO mass production, aligned with the Rubin Ultra deployment window, CPO became one of the highest-conviction structural growth themes emerging from this year’s show.
Want to learn more? Access relevant Third Bridge transcripts:
• 2026/05/07 - AI Interconnect – Long-term TCO Drivers, CPO Competitive Landscape & NPO Readiness
• 2026/04/13 - CPO Passive Optical Interconnect – Architecture Evolution, Volume Usage & Competitive Landscape (Conducted in Mandarin)
Liquid Cooling: From DLC to MCL
Jentech (TPE:3653) and Wiwynn (TPE:6669 ) showcased Micro-Channel Lid (MCL) demo — a packaging-integrated liquid cooling solution that is widely expected to be adopted for NVIDIA’s Rubin Ultra GPU platform. The MCL represents a fundamental architectural shift from the current Direct-to-Chip (DLC) cold plate approach, integrating microfluidic channels directly into the chip package lid itself, bringing coolant within just a few hundred microns of the heat source. Key advantages include lower thermal resistance and higher cooling capacity(>1kW TDP). Per Jentech, the bottleneck for micro channel lid now lies in packaging and testing.
Wiwynn showcased its new 6kW dual-sided liquid cooling plate, featuring liquid metal thermal interface material (TIM) and integrating a Cavity PCB architecture with Vertical Power Delivery (VPD). It also demonstrated diamond composite material in their microchannel cold plate configuration, with explicit positioning for future IC package-level cooling.
Want to learn more? Access relevant Third Bridge transcripts and register interest in upcoming calls:
• 2026/06/11 - Delta Electronics – Liquid Cooling Growth Outlook & Competition – Competitor Assessment (Conducted in Mandarin)
• 2026/05/14 - ASIC Liquid Cooling Market – 2026 Demand Outlook, Market Share Trends & Verification Progress (Conducted in Mandarin)
Our Sector Analysts are continually unlocking high-quality expert insights on the latest emerging trends in tech. Request a trial to discover more of the Third Bridge Library.
