01. Intel Unveils High-Memory GPU
On May 20, Intel CEO Pat Gelsinger visited Taiwan and launched multiple GPU and AI products, including the Intel Arc Pro B50/B60 GPUs, the Gaudi 3 AI accelerator, and the Intel AI Assistant Builder. This move showcases Intel's latest strategic deployment in AI and professional visual computing. The two GPUs are based on the Xe2 architecture and are specifically designed for AI inference and graphics workstations. The Intel Arc Pro B60 features 24GB of video memory and an AI peak computing power of 197 TOPS, while the Intel Arc Pro B50 has 16GB of video memory and an AI peak computing power of 170 TOPS. Both GPUs are equipped with Intel's XMX AI core and ray tracing units, and support multi-GPU expansion, offering high-cost-performance solutions for creators and AI developers. The Intel Arc Pro B50 has a retail price of $299, highlighting its excellent cost-performance ratio, while the Intel Arc Pro B60 is set to collaborate with system integrators and OEM vendors through a B2B model.
02. TrendForce: New HBM4 Specifications Raise Manufacturing Barriers, Expected Premium Exceeds 30%
On May 22, TrendForce reported that HBM4 technology, driven by the demand for AI servers, is being actively advanced by the three major manufacturers. Due to the doubling of I/O numbers to 2,048 in HBM4 and the adoption of logic chip architecture by some suppliers, costs have significantly increased, with the expected premium exceeding 30%. The data transfer rate of HBM4 remains above 8.0 Gbps, doubling the transmission volume compared to HBM3e. SK Hynix and Samsung's HBM4, which use logic chip architecture, offer higher stability and transfer efficiency. It is projected that by 2026, the total shipment volume of the HBM market will exceed 30 billion GB, with HBM4's market share increasing quarter by quarter, and it is expected to surpass HBM3e as the mainstream in the second half of the year. SK Hynix is set to maintain its market leadership, while Samsung and Micron need to improve yield and capacity to catch up.
03. AI-Driven, FSTs Quarterly Growth Is Expected
FST's earnings call on May 19 revealed that the company is actively expanding into the US and Singapore markets to diversify risks from Sino-US competition, and has achieved significant cooperation results in the HBM field. It is expected that the company's operations will grow quarter by quarter in 2025, but it needs to deal with uncertainties such as exchange rates and tariffs. The first-quarter financial report showed a revenue of NT$29.84 billion, a gross margin of 17.77%, a net profit margin of 7.99%, and an earnings per share (EPS) of NT$0.61. The main reasons for the decline were the global semiconductor inventory adjustment and the weak demand for 8-inch wafers. Qiu Shaoxun said that the company, with long-term orders and stable demand for 12-inch advanced process wafers, maintains a robust cash flow and profitability. The demand for 12-inch wafers, driven by AI, remains strong, while the demand for 8-inch wafers is sluggish, with only a slight increase due to short-term policy factors. Looking ahead to the second quarter, the demand for 12-inch wafers is expected to remain strong, driven by AI, and there are signs of recovery in memory applications; the demand for 8-inch wafers is expected to remain sluggish.
04. Wingtechs Semiconductor Business Grows Strongly, with Over 200 Analog Chips to Achieve Mass Production by Year-End
On May 21, Wingtech disclosed that its semiconductor business is accelerating the launch of new products this year. Its GaN products cover voltage specifications from 40V to 700V and have achieved mass production and delivery. In terms of SiC products, the company has launched a 1,200V automotive-grade SiC MOSFET suitable for applications such as onboard chargers, which is currently undergoing customer qualification. In the analog chip field, the company expects to achieve mass production of over 200 analog chips by the end of 2025. In the market, Wingtech's automotive-grade products have a significant advantage, with over 60% of its revenue coming from the automotive sector. The inventory in the consumer sector is healthy, and it is expected to enter the peak season in the second and third quarters. In the second quarter, the company's semiconductor business is expected to achieve year-over-year and quarter-over-quarter growth, with sufficient orders and near full-capacity production. Looking ahead to the second half of the year, the company is expected to expand its high-voltage power devices and analog chip offerings, with revenue growth expected to accelerate from 2026 onwards.
05. SEMI Chairman: Quantum Computing to Drive Semiconductor Market to $2 Trillion by 2040
On May 20, Ajit Manocha, President of the International Semiconductor Industry Association (SEMI), said at the Southeast Asia International Semiconductor Exhibition that the development of quantum computing is expected to double the global semiconductor industry's market size to $2 trillion by 2040. He pointed out that the growth of the semiconductor industry from $600 billion to $1 trillion mainly benefited from artificial intelligence, and that quantum computing will become the new growth driver in the future. Manocha also announced that SEMI will establish its second headquarters in Singapore this month to support the Asian market, reduce operational complexity, and increase flexibility. He emphasized that the semiconductor industry still faces supply chain resilience issues, and despite increased inventory, the chip shortage crisis has not been completely resolved.
06. Counterpoint: AI Applications Enter a New Stage, Covering Cloud, Edge, and Connectivity
At the Computex 2025, Counterpoint Research pointed out that AI technology has been widely applied in the fields of cloud, edge, and connectivity, marking a new stage of industrial application. AI servers and cooling technologies became the focus of the exhibition. Foxconn announced a collaboration with NVIDIA and TSMC to launch an AI supercomputer center, initially with a scale of 20MW, which will be expanded to 100MW in the future, with support from the Taiwanese government. Foxconn has already applied AI in electric vehicles, smart manufacturing, and digital health, and is transforming from a hardware OEM to an integrator of AI and digital platforms. MediaTek has become one of the first partners of NVLink Fusion, collaborating with NVIDIA to develop the GB10 Grace Blackwell superchip. It also showcased a home gateway combining 5G and generative AI, as well as the Dimensity Auto series of automotive chips. Through these collaborations, MediaTek has strengthened its presence in the fields of AI data centers, automotive, and edge computing.