Agent-based AI will drive a new wave of computing demand. AI agents are rapidly entering personal computers, enterprise software, and work systems, creating enormous computing demand. This not only presents opportunities for GPUs but also provides new growth opportunities for CPUs in reinforcement learning and AI orchestration. NVIDIA will likely showcase the development of physical AI, including driverless taxis, autonomous vehicles, and humanoid robots.

Taiwan is becoming a key hub for the global AI industry. NVIDIA believes that AI will spawn a completely new industry, just like steam power, electricity, and the internet. Leveraging its semiconductor and supply chain advantages, Taiwan now holds a significant position. (GTC Taipei 2026 Highlights Expectations)

CPU to Become the Backbone of Agentic AI Infrastructure:

Last year, the focus was on core counts, node transitions, and ASICs. This year, we expect the focus to shift toward the CPU, as it is now the backbone of agentic AI infrastructure. The CPU is no longer a head node attached to the GPU cluster. It is the orchestration layer that determines GPU utilization, tool-call throughput, and agent parallelism. We are going to see CPU vendors competing on cores-per-watt for action-oriented workloads (AMD Venice Dense, 256 cores / 512 threads with SMT) and on CPU-to-XPU interconnect bandwidth for reasoning-heavy workloads (NVIDIA Vera, NVLink-C2C at 1.8 TB/s bidirectional).

Further, we will see acceleration toward ARM-based CPUs. Google's Axion is now the default head node for TPU v8. AWS Graviton5 is in its fifth generation, with more than half of AWS instances now running on Graviton. NVIDIA is going all-in on Vera-based racks for the agentic era, with around $20B in CPU-only revenue for 2026, making it the biggest CPU vendor for the data center (DC). ARM's own AGI CPU further highlights that the ISA compatibility barrier is eroding faster than previously thought. Counterpoint forecasts that the global DC CPU market will expand from around $25B in 2025 to roughly $130B by 2030 (see Global Data Center x86 CPU Analysis and Insights). This is driven by rising CPU-to-GPU attach ratios. ARM is expected to capture around half of the total market by 2030 and all three architectures — x86, captive ARM (Refer to Arm-based CPUs to Capture 90% of AI ASIC Server CPU Market by 2029), and merchant ARM (Arm 2.0: Going Beyond Licensing to Capture Silicon Value with AGI AI CPU) — will compete simultaneously for the same hyperscale infrastructure budget.

Breaking the Copper Wall: the Next Big Focus, and Silicon Photonics Will Be the Key

Rollout of 1.6T Co-Packaged Optics (CPO) - high-performance scale-out networks are quickly transitioning from 800G to 1.6T architectures. However, conventional pluggables such as QSFP and OSFP face mounting challenges at these data speeds with respect to electrical channel loss, power consumption, front-panel density, and thermal constraints. To overcome these challenges, the industry will transition to CPO and NPO over the next two years (Embedded Optical Transmission Tech, Arrival of CPO Set to Deliver AI Supercomputing at Scale). We expect to see a few product and partner/customer announcements from optics vendors, particularly with respect to NPO.

Explosion of Inter-Rack Optical Pods - Another trend fast gathering pace in 2026 is scaling up multi racks via CPO/NPO optics. As AI clusters continually increase in size, copper will be confined to ultra-short, intra-rack connections and there will be a surge in CPO/NPO-based  scale-up switch deployments to enable multi-rack scale-up pods that allow hundreds of clustered GPUs across separate racks to share the same memory fabric.

Multi-Wavelength External Laser Sources (ELS) - Most external lasers used in data centers right now are single wavelength engines. However,  as data rates climb to 1.6T and 3.2T, laser engines will pivot towards Continuous-Wave (CW) architectures capable of delivering 8, 16 or 32 distinct wavelengths of unmodulated O-band light down a single strand of fiber.  I expect to see announcements about this plus also some news updates from vendors developing alternative integrated-laser optics solutions.

The AI PC Market Is Becoming Polarized:

Last year, the focus and conversations centered on NPUs and TOPS performance. This year, we expect the conversations and demonstrations to continue to shift toward practical user benefits, such as improved productivity, local AI processing, agent-based computing, and enhanced device connectivity. At the same time, higher memory and component costs are driving up notebook production costs. As a result, many brands and OEMs are focusing on premium SKUs that highlight AI features, OLED screens, and unique user experiences. Global Quarterly AI Laptop Tracker and Forecast, Counterpoint Conversations: Intel Explains the Role of Displays in Next-Gen AI PC Era, Global PC Shipments Grow 3.2% YoY in Q1 2026…  

At the same time, Apple — with the $499 MacBook Neo (MacBook Neo Apple MacBook Neo Set to Drive 7x Surge in Sub $700 Market, Macbook Neo: Targeting Higher Education) — has started a mini disruption at the lower end of the market, which is suffering from memory and CPU price hikes. This segment is ripe for a mini-war that cuts against the tide, and it will be telling to see how the non-Apple camp reacts and responds to Apple eating their lunch. The market is becoming polarized at both the low end and the premium end.

Physical AI Will Add a New Dimension to Computex:

We expect players such as NVIDIA, Qualcomm, Synaptics, Intel, TI, and NXP to showcase and expand their focus on Edge and Physical AI developments at the show. The intersection of Agentic and Physical AI kicks off this new wave of Embodied AI, with full-stack physical architectures where voice, vision, reasoning, and real-time motion control operate entirely at the edge. (Cumulative Physical AI Device Shipments to Reach 145 Million Units by 2035)

Humanoids and autonomous systems such as AGVs will gain major mindshare and showcase space at the show this year. Beyond compute, perception, sensor stacks, and precision machinery for Physical AI applications will also be a bigger part of the discussion and exhibits.

Some companies to watch:

Precision Machinery & Industrial Robotics: HIWIN, Solomon

Motion & Flexibility: ADLINK, Aetina, Cambrian Robotics, and Universal Robots

Vision & Perception Systems: Turing, oToBrite, Ability Enterprise, YUAN High-Tech, Lanner Electronics