NVIDIA: From GPU Vendor to End-to-End AI Platform

NVIDIA CEO Jensen Huang delivered the de facto opening keynote with two highlights that defined the session. The first was the formal introduction of the RTX Spark, NVIDIA's new ARM-based SoC co-developed with MediaTek, the company's first direct entry into the consumer laptop processor market. The second was Vera Rubin, designed for agentic AI, that earned three honours at COMPUTEX Taipei this year. It is reportedly set to raise the bar for scalability, resiliency and sustainable AI infrastructure.

Brady Wang, Associate Director at Counterpoint, noted, "From energy and infrastructure to chips, models, and applications, the entire AI industry is growing very quickly. As the successor to Blackwell, Vera Rubin is designed for the agentic AI era and can scale from rack-level systems to larger data center deployments. NVIDIA is building a complete AI platform, not just selling GPUs."

Commenting on the RTX Spark launch, Neil Shah, Vice President of Research at Counterpoint, said, "This is going to be the 'RTX Spark' moment for the personal computing segment in the AI era, the same way the iPhone, ChatGPT, or DeepSeek were for their respective eras in years to come. The traditional app-centric PC is evolving into a genuinely useful Agentic AI Personal Computer."

Shah added that the success of this vision hinges on one pivotal partnership: “The biggest success driver for this vision will be how Microsoft transforms itself to bring it to life. Qualcomm tried hard with Microsoft, but it is still a work in progress to drive real on-device AI stickiness to the mainstream. To make this happen at scale, it is going to require a truly special, committed partnership."

Jason Park, Research Associate at Counterpoint, highlighted what sets the Vera CPU apart at the infrastructure level: "NVIDIA has demonstrated its technical prowess with the new Vera CPU. It features 1.5TB/s bandwidth based on LPDDR5X with added ECC for enhanced security. Furthermore, the first-ever implementation of PCIe Gen 6.0 and support for CXL 3.1 underscores NVIDIA's focus on maximizing bandwidth."

Qualcomm: Owning the Power-constrained Edge

Qualcomm CEO Cristiano Amon delivered the official COMPUTEX Taipei opening keynote, centring his remarks on how AI agents will replace devices as the primary end point of digital experience. Qualcomm announced the Snapdragon C platform, targeting $300 Windows laptops.

Drawing a sharp contrast between NVIDIA and Qualcomm's approaches, Counterpoint Senior Analyst Parv Sharma said, "NVIDIA has a heavy focus on AI training, high-fidelity simulation environments, and high-performance compute platforms, while Qualcomm focuses on highly integrated, power-efficient edge AI compute combined with robust on-chip connectivity platforms for physical AI."

ARM’s Compute Platform Ambitions

ARM CEO Rene Haas took the stage during COMPUTEX Taipei 2026 to talk about the shift to agentic AI and how ARM is powering the next era of AI compute across cloud, AI PCs, and the intelligent edge and also to share more about how ARM-based products are shaping the future of AI.

Neil Shah noted that ARM moving into its own silicon underscores how rapidly the ISA compatibility barrier is eroding. "This is a great win for ARM as it extends its share in the PC segment. This is also a successful play for MediaTek to indirectly expand into the PC segment at the highest end."

Intel: CPUs at the Centre of the Agentic Era

Intel's keynote at COMPUTEX demonstrated a "Disaggregated Inference" architecture jointly with SambaNova and Vista Equity Partners, assigning different AI tasks to the processors best suited for them and delivering 2x-3x higher performance than a GPU-only approach. Intel confirmed that its 18A process node has entered full-scale production. It also introduced Project Falcon, an approach to running confidential AI workloads locally while using the cloud only for public information.

Brady Wang, Associate Director at Counterpoint, attended the Intel keynote and noted the significance of the CPU's resurgence in the agentic era: "AI is moving from simple 'Prompt and Response' systems to 'Agentic AI,' where models can use tools, check their own work, and complete multi-step tasks. As these workloads rely more on logical and sequential processing, Intel believes CPUs will play a larger role, with the CPU-to-GPU ratio moving from 1:8 toward nearly 1:1 in future AI systems."

Commenting on Intel's privacy-first approach to enterprise AI, Wang said, "Many companies want to use AI but do not want sensitive data sent to the cloud. Intel's Project Falcon solves this by running confidential AI workloads locally while using the cloud only for public information, helping businesses balance privacy and performance."

Marvell at its Optics Inflection Point

Marvell made a pointed declaration at its COMPUTEX session – 200G per lane will be the final generation where standard copper wiring is viable for the rack distances. To scale AI beyond a single rack or data center cluster without massive latency and power spikes, optics must move directly to the rack level.

The company showcased its new Teralink T100 switch with integrated Co-packaged Optics (CPO), claiming a drastically reduced board footprint while maximising data throughput. Notably, the Teralink T100 can handle both scale-up and scale-out workloads. Jensen Huang joined Marvell CEO Matt Murphy on stage and called Marvell "the next trillion-dollar industry," describing it as a critical partner enabling NVIDIA to scale its networks.

Lightmatter: Photonics Moves Closer to Market

Lightmatter showcased its lineup at COMPUTEX Taipei 2026, including the Passage L20 Near-packaged Optics (NPO) module, the L200 Series Co-packaged Optics (CPO) module, eClick and vClick fibre optic connectors, and the Guide VLSP laser engine, with several products already sampling and commercial volumes for some expected in H2 2026.

Two technology differentiators stood out. First, Lightmatter is currently the only company offering bi-directional (BiDi) multiplexing, which halves the number of fibre optic cables required in a rack, critical for high-density deployments. Second, its Guide laser engine is the only product currently capable of operating at 16 wavelengths.

AI PCs: A Market Polarizing at Both Ends

Counterpoint Research's visit to ASUS headquarters during COMPUTEX Taipei provided a ground-level view of how OEMs are navigating the rapidly shifting landscape. The session, attended by Ritesh Bendre, Global Content Manager at Counterpoint, included a roundtable with Peter Chang, General Manager of Asia Pacific, and Arnold Su, VP of Consumer Business at ASUS India.

ASUS has outlined a deliberate chip partner strategy – Intel for common usage, AMD for high-end AI and gaming, and Qualcomm SoCs for thin-and-light, battery-optimised AI PCs, while actively exploring agentic AI solutions and collaborations including with Google for Googlebook.

Addressing concerns about rising laptop prices, ASUS said it is working with financing partners in India to offer better EMI schemes and saw an increase in EMI adoption from 20% at the end of 2025 to around 35% in Q1 2026.

Conclusion

COMPUTEX Taipei 2026 showed that the AI era has moved decisively from the data centre to the device and now, to the physical world.

Counterpoint Research will continue to track developments across semiconductors, AI PCs, physical AI, and the broader computing ecosystem. For our latest data and analysis, visit counterpointresearch.com.