COMPUTEX 2025—Taiwan Delivers: AI Workstation Democratization, Maturing AI PC Propositions & Edge AI Systems

Taiwan’s position in the high-end global technology supply chain is unmatched and unlikely to change based on the tariffs or other barriers. Noticeable at COMPUTEX was a more tentative approach by some silicon vendors over last year’s COMPUTEX, as a result of the uncertainty to supply chains resulting from U.S. trade restrictions—especially among those in receipt of government subsidies as part of reshoring efforts. A countervailing force to this pessimism is the flexibility introduced to supply chains during COVID-19, and the learnings from this era, as strategists are better prepared for any radical changes they may need to implement. Nonetheless, the show was packed with announcements. Moreover, NVIDIA expanded its presence over 2024’s show, including a commitment to cement its base on the island with a second flagship office, to be built near the capital, Taipei. Standout developments include:

• Artificial Intelligence (AI) Personal Computer (PC): Competition is heating up in this space as Intel and AMD’s second-generation silicon and developer platforms mature and Qualcomm aggressively pushes into the space with its Arm-based Snapdragon X contender.
• AI Workstations: Intel joins AMD and NVIDIA in the discrete Graphics Processing Unit (GPU) card space targeting AI developers and on-premises inferencing workloads with a low-end challenger from its Arc family in a segment poised for growth.
• Edge AI: Taiwan’s extensive on-premises AI server/box showcased its system innovations with varying levels of maturity—Advantech being a frontrunner with a new containerized approach for deploying AI on diverse hardware.

Edge AI Solutions

Taiwan’s edge AI hardware industry was out in full force, with a clearly improved offering over last year’s event. Pegatron, Advantech, C&T, AAEON, and Portwell were demonstrating edge AI boxes, including fanless solutions, and compact on-premises server offerings in partnership with diverse compute vendors such as DEEPX, Hailo, Axelera AI, Intel, AMD, and NVIDIA, as well as Independent Software Vendors (ISVs) with verticalized applications, such as factory automation and warehouse security. In line with ABI Research’s thinking, vision language models with Generative Artificial Intelligence (Gen AI) capabilities were a popular demo and are seeing a lot of interest and traction for safety and security applications.

AI PCs

AI PCs are slowly maturing and there are huge resources being spent by Qualcomm, Intel and AMD, as well as individual Original Equipment Manufacturers (OEMs) like Acer, ASUS, Lenovo, and Hewlett Packard (HP), on establishing themselves in this area, creating productivity-enhancing AI applications running on heterogenous architectures both within and without the Windows Copilot ecosystem. Intel is the leader and has reaffirmed the shipment target of 100 million AI PCs by the end of 2025, followed by AMD and then Qualcomm. The first frontier is the consumer sector, while the enterprise AI PC is still grappling with a mature value proposition that can address the familiar concerns of businesses deploying PCs at scale, namely security, Total Cost of Ownership (TCO), and hardware lifecycle guarantees.

AI Workstations

The AI workstation market is a significant growth segment, including in the low end, for AI development, which is undergoing a democratization with the help of cheaper hardware and the proliferation of open-source models and their repositories such as Hugging Face. NVIDIA and AMD have been addressing this segment for a longer period with their extensive RTX and Radeon families of GPU cards, respectively. NVIDIA has expanded its high-end portfolio with DGX Spark, which provides headline-grabbing 1,000 Tera Operations per Second (TOPS), albeit in FP4, for desktop AI development. Intel released a new generation of standalone Peripheral Component Interconnect Express (PCIe) GPU cards, the B60 Pro based on the Arc architecture, competitively priced at around US$300 each (and scalable) for the low-end inference workstation segment.

In the rapidly evolving AI landscape, market leadership is no longer determined by hardware performance alone, but by the strength of a vendor's software ecosystem and its commitment to developer enablement. To maintain and expand their competitive edge, companies across the spectrum—from industrial edge providers to AI PC chipmakers—must adopt strategies that address barriers to adoption.

• Edge AI Solutions: Advantech is a leading hardware solutions provider with a rapidly maturing software proposition ahead of its competitors like Axiomtek and ADLINK. To crystallize its lead with the WISE-Edge software platform and become the go-to vendor, enablement is key to broaden AI development to enterprises lacking the in-house AI talent to train and deploy models themselves. Deeper partnerships with other developer platforms such as LandingAI and DeGirum will ensure future-proofing support for a broad range of edge AI semiconductor platforms and enable customers to use their preferred Machine Learning (ML) toolset. A further step would be for Advantech to incentivize partners to build exclusively for the Advantech ecosystem, creating a flywheel effect via ISVs and integrators.
• Enterprise AI PCs: Intel with vPro and, to a lesser extent, AMD with Ryzen Pro, are ahead in this domain, while Qualcomm is aggressively trying to capture consumer AI PC market share and expand support for gaming, which is more established in the x86 arena. This should take a two-pronged approach: 1) establish a clear, separate brand for enterprise-grade PCs with a focus on enhanced security, manageability, and lifecycle support, and TCO; and 2) an ISV ecosystem that should follow the lead of Intel’s AI PC Acceleration Program, which has onboarded over 100 ISVs to develop applications optimized for Intel hardware and goes further by delineating verticalized solutions for, e.g., financial services, content creation, and coding, with a clear emphasis on the productivity enhancing features of applications.
• AI Workstations: Intel’s open ecosystem approach will appeal to many wishing to avoid vendor lock-in with the expensive hardware offered by NVIDIA. Concurrently, AMD has decided to expand its ROCm software stack to include some of its Radeon GPU portfolio, which will address workstation AI development and allow the migration of workloads between cloud and workstation environments, which is already established in NVIDIA’s ecosystem. Intel must ensure that its forthcoming data center AI accelerator offering can enable the same seamless transition between cloud and workstation to capture more of the burgeoning AI developer market as more enterprises look to deploying and developing models across locations. This strategy should be tightly coupled with a university seeding program that funds and supports universities and colleges to build curricula and research around Intel’s hardware and software stack. This long-term strategic play ensures that the next generation of AI talent is native to Intel’s ecosystem.

The key strategic takeaway is that market leadership now depends less on hardware alone and more on creating robust software ecosystems and developer enablement programs to lower the barrier for enterprise AI adoption.