Focus on COMPUTEX 2026! Will the entire AI supply chain ignite?
Focus on June 1! Jensen Huang kicks off COMPUTEX 2026—what companies deserve close attention?
COMPUTEX 2026 will be held grandly from June 2 to 5, with this year’s core theme being 'AI Together.'
The keynote speech by NVIDIA CEO Jensen Huang is undoubtedly the most anticipated highlight for global tech enthusiasts and investors. $NVIDIA (NVDA.US)$ CEO Jensen Huang's keynote address, which is expected totake place at 11:00 AM on June 1at the Taipei Pop Music Center, kicking off this year’s AI technology storm!
Morgan Stanley expects Huang to comprehensively showcase NVIDIA’s next-generation AI computing architecture centered on the Rubin GPU and Vera CPU, further solidifying its dominance in AI infrastructure.
By this point, sharp-eyed fellow investors have likely already sensed the next major market theme. So, what are the must-watch, high-value highlights of COMPUTEX 2026? And within the AI supply chain poised to ignite soon, which opportunities deserve our early positioning?
What are the key highlights of COMPUTEX 2026?
COMPUTEX 2026 has fully evolved from a traditional PC exhibition into the global AI supply chain’s 'stress-testing ground.' For industry observers, the most critical signals are no longer about individual product launches, but ratherplatform direction, supply chain collaboration, and capacity planningundergoing comprehensive upgrades.
1. Redefining System Boundaries: From 'AI Chips' to 'AI Factories'
COMPUTEX 2026 has evolved from a traditional PC exhibition into the global AI supply chain’s stress-testing ground.
Citi believes market focus is shifting beyond the specifications of individual GPUs toward complete 'AI infrastructure' or 'AI factories,' encompassing co-optimized CPU, network bandwidth, memory hierarchy, thermal management, and power delivery systems.
Rack-level architecture is replacing standalone servers as the new system boundary, with future deployment models shifting from server-level to rack-level—or even pod-level—integration.
2. NVIDIA’s Next-Generation Core: Rubin GPU and Vera CPU
Rubin GPU Thermal Design and PackagingThe Rubin GPU is expected to natively support a thermal design power (TDP) of 1.8 kW, and through system-level thermal co-design, could even reach 2.3 kW TDP. The Rubin Ultra, expected to adopt HBM4e in 2027, will maintain a single-package, dual-die (2-die) configuration to ensure yield.
Strategic Positioning of the Vera CPUThe Vera CPU is not intended to compete with Intel or AMD in the general-purpose server market; instead, it is specifically designed as a controller for 'Agentic AI' and AI factory orchestration. Supply chain research indicates that Taiwan Semiconductor is allocating additional CoWoS-R capacity and 3nm wafers for the Vera CPU. Based on current capacity planning, Morgan Stanley believes thata reasonable shipment assumption for standalone Vera CPUs is 1.5 million units.
Extreme Rack ConfigurationThis includes the Rubin NVL72 and links eight racks together via copper and optical interconnects to form the massive Rubin Ultra NVL576 compute domain, comprising 576 GPUs.
3. Technical Challenges in Thermal Management, Power Delivery, and Co-Packaged Optics (CPO)
Thermal Management and 800V DC Power DeliveryPower consumption for the Rubin NVL72 rack is estimated to surge beyond 180–220 kW, forcing the industry to shift toward direct liquid cooling and an 800V DC power architecture.
Rise of Networking and CPOSpectrum-X is emerging as a dedicated AI networking architecture. In the Rubin Ultra generation, due to the physical scale of NVL576 exceeding the limits of copper interconnects, co-packaged optics (CPO) and direct optical connectivity will become critical.
4. Advanced Packaging Takes Center Stage Behind the Scenes and Taiwan Semiconductor's Capacity Expansion
According to SemiVision, advanced packaging technologies (such as CoWoS and SoIC) are considered the hidden stars of this year’s COMPUTEX, as AI chips are becoming larger, hotter, and extremely bandwidth-hungry.
To meet strong demand for AI GPUs and CPUs, Taiwan Semiconductor is aggressively expanding its CoWoS capacity. Morgan Stanley estimates its CoWoS capacity assumption for 2027 has been raised from 170,000 wafers per month to 200,000 wafers per month.
Due to prioritizing CoWoS expansion, Taiwan Semiconductor has slightly adjusted its SoIC capacity ramp-up pace, with 2027 and 2028 capacity targets revised downward to 40,000 and 70,000 wafers per month, respectively.
Which segments deserve attention?
According to SemiVision, for industry observers, the most important aspect of COMPUTEX may not be individual product launches. More significant signals will come from platform directions, supply chain collaboration, and capacity planning.
The first signal is the competitive balance among AI platforms. NVIDIA remains dominant, but AMD is becoming more aggressive in entering the data center AI market, while hyperscalers’ ASICs are also becoming more visible. The market is unlikely to converge on a single architecture. Instead, we may see a more segmented landscape: GPUs for general-purpose AI training and large-scale inference, custom ASICs for optimized cloud workloads, and Arm-based platforms for high-efficiency computing.
The second signal is the evolution of advanced packaging. Any discussions around CoWoS, HBM integration, chiplet packaging, glass substrates, or larger packaging architectures should be
Watch closely. These are not minor technical details—they directly impact the ability to ship AI systems.
The third signal is the rise of system-level supply chain collaboration. AI server manufacturers, PCB suppliers, thermal solution providers, power module companies, and optical interconnect vendors will become increasingly important. The bottleneck is no longer just GPUs—it's the entire rack.
Fellow investors have also compiled a list of noteworthy concept stocks along the industrial chain for reference:
Core Computing Power and Platform Competition: No Longer Dominated by a Single Architecture
Competition among AI platforms is becoming more balanced. NVIDIA remains the dominant force, but AMD is aggressively expanding into data center AI, while hyperscalers’ custom ASICs are gaining greater visibility.
Key Chip Giants: $NVIDIA (NVDA.US)$ 、 $Advanced Micro Devices (AMD.US)$ 、 $Intel (INTC.US)$ 、 $Arm Holdings (ARM.US)$ 、 $Qualcomm (QCOM.US)$ 。
Massive Memory Demand: The Rubin architecture is driving surging demand for HBM, $Micron Technology (MU.US)$ 、 $CSOP SK Hynix Daily (2x) Leveraged Product (07709.HK)$ 、 $CSOP Samsung Electronics Daily (2x) Leveraged Product (07747.HK)$ which could benefit, in addition $SanDisk (SNDK.US)$ is also worth watching.
Advanced Packaging Evolution: The Lifeline Determining the Ultimate Shipment Capacity of AI Systems
Any discussion around CoWoS, HBM integration, glass substrates, or larger packaging architectures should be closely monitored, as it directly impacts the ceiling of production capacity.
Advanced Packaging Foundry: $Taiwan Semiconductor (TSM.US)$ holds an absolute dominant position, $ASE Technology (ASX.US)$ 、 $Amkor Technology (AMKR.US)$ closely follows.
Core Equipment and Testing: $Applied Materials (AMAT.US)$ 、 $ASML Holding (ASML.US)$ 、 $Lam Research (LRCX.US)$ and the Hong Kong-listed leader $ASMPT (00522.HK)$ provide solid equipment support; the testing segment is led by $Teradyne (TER.US)$ and others.
Optical Communications and DCI Networks: Pushing Beyond Physical Limits Across Rack Boundaries
When cluster interconnect scales reach NVL576, the importance of Data Center Interconnect (DCI) in network and fiber optic supply chains will surge dramatically. This is no longer a simple upgrade—it has become an essential requirement to break through bandwidth bottlenecks.
DCI (Data Center Interconnect): Focusing on the network transmission and fiber optic supply chain $Ciena (CIEN.US)$ And, $Nokia Oyj (NOK.US)$ 。
CPO switches and optical modules: $Broadcom (AVGO.US)$ And, $Marvell Technology (MRVL.US)$ Leading in CPO chips; attention on optical modules and fiber optics $Coherent (COHR.US)$ 、 $Lumentum (LITE.US)$and$Corning (GLW.US)$ 、 $YOFC (06869.HK)$Wait.
System-level supply chain collaboration: The rise of 800V DC power and cooling solutions
The bottleneck is no longer just the GPU, but the entire rack. Cabinet power consumption exceeding 180kW has made power delivery and liquid cooling essential rather than optional.
Industrial power infrastructure and thermal management: $Vertiv Holdings (VRT.US)$ Spanning both power and cooling domains, $Eaton (ETN.US)$ And, $Modine Manufacturing (MOD.US)$ they are also key beneficiaries.
Wide-bandgap semiconductors and power management: $STMicroelectronics (STM.US)$ 、 $ON Semiconductor (ON.US)$ 、 $Wolfspeed (WOLF.US)$ Provide high-voltage components; $Monolithic Power Systems (MPWR.US)$ And, $Texas Instruments (TXN.US)$ Lead the power management system.
Summary
Overall, COMPUTEX 2026 sent the clearest signal yet: the AI race has officially escalated from a 'single-chip showdown' to an all-out competition in 'system-level engineering.' This transformation represents not only a technological upgrade but also a fundamental reshaping of global industry value distribution logic.
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty.Read more
Comments (28)
to post a comment
301
916