PERSY at COMPUTEX 2026:
a view from the front row

Taipei, Taiwan  |  June 2026

The PERSY team attended COMPUTEX 2026 in Taipei — the largest edition in the show's 45-year history, with 1,500 exhibitors from 33 countries under the theme "AI Together." As a system integrator with over 35 years of experience building high-performance and high-availability clusters, we don't go to COMPUTEX to browse products. We go to see where the industry is heading two to three years ahead — before the trends reach the catalogues.

Keynote

Intel: x86 Reinvented for the Inference Era

We attended the presentation by Lip-Bu Tan, Intel's CEO — one of the most substantive talks at the show. Intel outlined a strategy across four computing domains: PCs, edge and physical AI, traditional data centres, and the emerging "intelligence centers."

The centrepiece announcement for the enterprise segment was Xeon 6+ — the first data centre processor built on Intel's 18A process node: 288 efficiency cores and 576MB L3 cache, designed for AI workloads while preserving efficiency for mission-critical enterprise applications. IDC data cited on stage: more than 8 in 10 servers will remain x86-based through 2030.

Equally significant is the strategic pivot: through the Rackscale Blueprints initiative and rack-scale AI infrastructure developed with Foxconn and SambaNova, Intel is transitioning from selling processors to integrating complete systems.

The standout moment was a live hybrid inference demonstration with Perplexity: local models on Core Ultra Series 3 autonomously route each task — sensitive data processed on-device, large-scale workloads sent to the cloud — automatically, based on data sensitivity. For regulated industries, this is the architecture that will be in demand.

Market move

NVIDIA: From the Data Centre to the Desktop

NVIDIA made two announcements with long-term implications. First: the RTX Spark Superchip — a Windows on Arm platform with up to 20 Arm cores, a Blackwell GPU with 6,144 CUDA cores, and 128GB unified memory capable of running language models up to 120 billion parameters locally. NVIDIA is entering the personal computer as a platform player, not merely a GPU supplier.

Second: the new NVIDIA Vera CPU is no longer a slide — it is a real product shipping in partner systems. At the GIGABYTE booth we examined the XV25-VCO-AJ1 up close: a server on the Vera CPU platform with up to 3TB LPDDR5X ECC memory, 1.8 TB/s NVLink-C2C interconnect, and PCIe Gen6. The Vera Rubin platform is in full production, with a roadmap extending three generations ahead.

Show floor

Integration Is the New Competitive Territory

Two booths illustrated most clearly where the market is moving:

GIGABYTE

Positioned beyond hardware with the message "AI Infrastructure, Built as One" — an integrated approach to building AI Factories: from consultation and site planning, through rack and server design, cabling, power and thermal management, to deployment, validation and sustainable operations. Component manufacturers are moving onto the system integration playing field.

QNAP + ASUS Ascent GX10

Edge AI in action: the ASUS Ascent GX10 (NVIDIA GB10 Grace Blackwell, 128GB unified memory, up to 1 petaFLOP FP4, Computex 2026 Best Choice Award) integrated with QNAP's NAS ecosystem — AI models and datasets served directly from network storage over 10GbE. The "Ready and Recovery" theme united data sovereignty, AI integration, and cyber resilience — the right priority for European clients with regulatory requirements.

Takeaways

Three Conclusions That Will Define the Next Projects

Ecosystems

Nobody sells a "server" anymore — what's sold is a rack-scale solution: compute, memory, network, cooling, and software stack, designed as one.

Connectivity

In campus-scale clusters, data movement has replaced compute density as the limiting factor. Investment in network architecture is now on par with GPU investment.

Inference

Market focus is shifting from model training to cost-effective model operation — with direct implications for processor, memory and architecture choices.

Planning AI or HPC Infrastructure?

Tomorrow's infrastructure is designed today — network, power and cooling planned together with compute, not after. That is what PERSY has been doing for over 35 years.

Get in touch