COMPUTEX 2026 Complete Guide: Must-See Jensen Huang GTC, Five Major AI Theme Highlights, Time and Venue

COMPUTEX 2026 will take place at the Taipei Nangang Exhibition Center (TaiNEX 1 & 2) from June 2 to 5, 2026. NVIDIA CEO Jensen Huang's GTC Taipei keynote will be held on June 1, followed by CEOs from Qualcomm, Intel, Marvell, and others. (Background: NVIDIA opens nearly 100 positions in Taiwan; at NVIDIA HQ, Jensen Huang teases: "Spoiling it too early ruins the surprise"—major Computex reveals ahead) (Context: AMD announces over US$10 billion investment in Taiwan, racing to secure advanced packaging capacity through partnerships with ASE, Powertech, and Inventec) Under the banner of "AI Together," COMPUTEX 2026 officially declares itself the largest edition in history, with 1,500 exhibitors from 33 countries and over 6,000 booths. Jensen Huang, Lip-Bu Tan, and Cristiano Amon will all stand on Taipei stages within the same week. The officially announced exhibition dates run from Tuesday, June 2 to Friday, June 5, 2026, with the main exhibition area spanning two venues: - Nangang Exhibition Center Halls 1 & 2 (TaiNEX 1 & 2): Main exhibition area, where AI servers, GPUs, robotics, cooling, and major ODM exhibitors are concentrated - Taipei World Trade Center Hall 1 (TWTC Hall 1): Partial exhibition area, including the InnoVEX startup zone - Taipei International Convention Center (TICC): Stage for CEO keynotes and venue for COMPUTEX Forum sessions Additionally, NVIDIA CEO Jensen Huang's GTC Taipei keynote is specially scheduled for Monday, June 1—one day before the show opens—at the Taipei Music Center, with simultaneous online livestream. The full GTC Taipei agenda will run from June 2 to 4 at TICC. Opening hours at a glance: Important reminder: June 2 to 4 are restricted to industry professionals; the general public may purchase tickets to enter only on June 5, and must be at least 18 years old. NVIDIA has announced that Jensen Huang's June 1 keynote at the Taipei Music Center will focus on NVIDIA's latest AI infrastructure roadmap, including the Vera Rubin architecture and progress in deploying its robotics platforms. The key highlights extend beyond product launches—NVIDIA's public stance on the "commercialization of compute" path matters greatly. For cloud providers, ODM manufacturers, and Taiwan stock investors, whatever Jensen Huang says in Taipei is effectively a directional guide for order flow over the next two quarters. Online livestreaming will be available simultaneously. If you cannot attend in person, it is recommended to confirm the official livestream link in advance to capture first-hand information in real time. The COMPUTEX Forum will run concurrently at TICC. Officials have announced six major themes covering Robotics & Physical AI, Generative AI, Data Intelligence & Security, and more. With over 30 speakers from NVIDIA, Microsoft, and Google DeepMind, it is the main battlefield for deep technical sessions. This edition can be broken down into five practical, trackable technology themes: 1. AI Servers and Rack-Scale Supercomputing NVIDIA's Vera Rubin NVL72 is the most-watched hardware exhibit of this edition. The solution integrates 72 Rubin GPUs and 36 Vera CPUs into a single liquid-cooled rack, with Foxconn as the primary contract manufacturer for the compute tray. AMD is also bringing its rack-scale solution Helios, which integrates Instinct MI400 GPUs, 6th-generation EPYC processors, and Pensando networking chips to directly challenge NVIDIA's position in the AI training market. 2. Liquid Cooling: Supporting Infrastructure for Higher Compute Density The higher the GPU compute power, the greater the heat output, and traditional air cooling is approaching physical limits. At this edition, liquid cooling has become a standard showcase project for major ODM manufacturers. Compal, under the theme "The Engine of Intelligence," highlights its liquid cooling solutions and data center infrastructure; Foxconn is showcasing liquid-cooled racks and CPO (Co-Packaged Optics) solutions. 3. Humanoid Robotics and Physical AI NVIDIA's Jetson Thor humanoid robotics computing platform, based on the Blackwell architecture, delivers approximately 2,070 FP4 TFLOPS of compute performance. In simple terms, robots used to rely on remote servers to process their "brain." Jetson Thor moves this brain directly into the robot body, enabling real-time judgment and action even without network connectivity. Both Foxconn and Pegatron are showcasing robotics solutions at this edition, showing that Taiwan's manufacturing supply chain is extending from "assembling AI servers" to "assembling AI robots." 4. AI PC: Compute Moving from Data Centers to End Devices The core concept of AI PC is to run AI inference directly on the chip in your laptop or desktop, without needing to connect to the cloud every time. In simple terms, when you ask AI a question, it computes directly on your computer rather than waiting for a cloud server to respond—faster, cheaper, and your data stays local. Competitors in this theme include the Qualcomm Snapdragon X series, Intel Core Ultra series, and AMD's mobile platform solutions. 5. High-Speed Networking (CPO Optical Communication): The Veins of AI Clusters The performance bottleneck in AI server clusters often lies not in how fast a single GPU is, but in the data transfer speed between GPUs. CPO (Co-Packaged Optics) is currently the most market-favored solution, replacing traditional electrical signal connections with optical signals for faster speed and lower power consumption. Marvell leads the market in this field, which is precisely the logic behind Jensen Huang's decision to personally share the stage with Marvell's CEO during his keynote. The following summarizes data from market institutions and statistics, for reference only and not constituting investment advice. The Big Three ODMs: Does "Old AI" Have Room for a Catch-Up Rally? Market institutions estimate the market share distribution of Taiwan's three major ODMs in AI server rack shipments as follows: - Foxconn (2317): Approximately 3,700 racks, with around 44% market share, ranking first. This edition showcases multiple business lines including Vera Rubin NVL72 compute trays, CPO, robotics, smart cities, and electric vehicles - Quanta (2382): Approximately 2,100 racks, with AI server shipment ratio rapidly climbing - Wistron (3231): Approximately 1,300 to 1,400 racks, with deepening collaboration with NVIDIA Institutional views suggest these three companies' share prices have not yet fully reflected the improved visibility of AI server orders, leaving room for a catch-up rally—but specific operations still require independent risk assessment. Other targets named by institutions: - TSMC (2330): Vera Rubin, Blackwell, and other high-end GPUs are all manufactured using TSMC's advanced processes. The stronger the demand for compute, the more directly TSMC benefits - Gigabyte (2376): A major supplier of GPU server boards and a beneficiary stock in the NVIDIA ecosystem - Wiwynn (6669): An ODM designer and manufacturer of high-end AI servers, and a key partner for the NVIDIA HGX series - Cooling-related sector: Rising demand for liquid cooling brings related manufacturers into institutional focus - Kinsus (3189): An AI server substrate supply chain player, named among the beneficiary targets in market statistics The above compilation reflects market observations only and does not represent BlockTempo's investment advice. Industry Professionals (June 2 to 4) Qualified industry professionals can register for free online in advance via the official portal. Upon completion, present the QR code on-site to exchange for a badge—no need to queue separately to purchase tickets. Must be 18 or older. The COMPUTEX main exhibition area and the InnoVEX startup zone share the same badge, granting access to both areas without separate application. InnoVEX is the hub for startups, early-stage projects, and accelerators—worth scheduling dedicated time for those tracking early-stage technology trends. General Public (June 5 Only) Public admission is limited to June 5; tickets must be purchased, attendees must be 18 or older, and ticket prices are subject to official announcement. On June 5, opening hours at TaiNEX are shortened to 09:30–15:30, closing 90 minutes earlier than previous days. Plan your itinerary carefully to avoid arriving in the afternoon only to find the venue near closing time. Transportation Recommendations The main exhibition venue, Nangang Exhibition Center, is accessible via the Taipei MRT "Taipei Nangang Exhibition Center Station," where the Bannan Line (Blue Line) and Wenhu Line (Brown Line) intersect. As the terminus of the Bannan Line, it takes approximately 20 to 30 minutes from central Taipei with no transfers needed. Crowds will be concentrated during the exhibition period; it is recommended to depart early and avoid the entrance congestion during the first hour after opening. Based on the above information, here are suggested planning directions for readers of different identities: For Industry Professionals (Full 4-Day Schedule): - 6/1: Taipei Music Center, Jensen Huang's GTC Taipei keynote (top priority—arrive early to secure a seat or confirm the livestream link as backup); Qualcomm's Amon keynote the same day at 14:00 at TICC - 6/2 morning: TICC, Marvell CEO + Jensen Huang on the same stage, AI interconnect roadmap; 6/2 afternoon switch to Intel's Lip-Bu Tan session - 6/2 to 6/4: TaiNEX, focusing on the NVIDIA Vera Rubin NVL72 zone, Foxconn AI robotics showcase, and AMD Helios zone; those interested should also schedule time for the InnoVEX startup zone - 6/3: NXP CEO keynote, COMPUTEX Forum deep technical sessions (Robotics and Physical AI, Generative AI theme sessions) For General Public (June 5 Only): - Enter as soon as doors open and prioritize popular booths; TaiNEX closes earlier (15:30), while TWTC stays open until 16:30 - The robotics showcase and liquid-cooled server zones offer the strongest visual impact of this edition—list them as priorities - The InnoVEX startup zone is at TWTC. Those interested in early-stage technology can plan to move there after visiting TaiNEX - Crowds are most concentrated between 11:00 and 13:00; those with flexibility may want to avoid this time slot