The global artificial intelligence (AI) server market is expected to remain dominated by servers based on graphics processing units (GPUs) this year, but application-specific integrated circuits (ASICs)-based models are set to see faster growth in shipments, the Market Intelligence and Consulting Institute (MIC) said.
As major cloud service providers (CSPs) continue to train large language models for machine learning, outlays on AI data centers and high-density GPU server racks remain a priority, MIC said.
CSPs would continue relying on GPU ecosystems when offering computing resources to customers, since most enterprises and developers remain accustomed to working within GPU environments, MIC information and communications technology section deputy director Chris Wei (魏傳虔) said last week.
Photo courtesy of Hon Hai Precision Industry Co
However, demand for servers powered by ASICs is growing as the market’s focus gradually shifts toward inference, Wei said.
While AI training is akin to preparing materials in a centralized factory — capital-intensive and large in scale — inference resembles bringing those materials back on premises, where use is more distributed and steady.
Such a pattern makes shipment growth for ASIC-based servers more likely to outpace that of GPU-based models this year, Wei said.
Shipments of servers used for AI inference are expected to be about four times higher than those used for training this year, highlighting the role ASICs could play.
ASICs are well-suited to inference demand, as they prioritize efficiency and cost control, Wei said.
Inference workloads are relatively fixed and suitable for specialized designs that can complete specific tasks with lower power consumption, he said.
MIC expects Taiwanese manufacturers who already produce GPU servers to expand into the ASIC segment smoothly, Wei said.
The key to future competition for these manufacturers lies not only in securing orders tied to specific chips, but in whether they can build system integration capabilities and flexible production lines for both GPU and ASIC-based platforms, he said.
Nvidia Corp chief executive officer Jensen Huang (黃仁勳) on Monday introduced the company’s latest supercomputer platform, featuring six new chips made by Taiwan Semiconductor Manufacturing Co (TSMC, 台積電), saying that it is now “in full production.” “If Vera Rubin is going to be in time for this year, it must be in production by now, and so, today I can tell you that Vera Rubin is in full production,” Huang said during his keynote speech at CES in Las Vegas. The rollout of six concurrent chips for Vera Rubin — the company’s next-generation artificial intelligence (AI) computing platform — marks a strategic
REVENUE PERFORMANCE: Cloud and network products, and electronic components saw strong increases, while smart consumer electronics and computing products fell Hon Hai Precision Industry Co (鴻海精密) yesterday posted 26.51 percent quarterly growth in revenue for last quarter to NT$2.6 trillion (US$82.44 billion), the strongest on record for the period and above expectations, but the company forecast a slight revenue dip this quarter due to seasonal factors. On an annual basis, revenue last quarter grew 22.07 percent, the company said. Analysts on average estimated about NT$2.4 trillion increase. Hon Hai, which assembles servers for Nvidia Corp and iPhones for Apple Inc, is expanding its capacity in the US, adding artificial intelligence (AI) server production in Wisconsin and Texas, where it operates established campuses. This
Garment maker Makalot Industrial Co (聚陽) yesterday reported lower-than-expected fourth-quarter revenue of NT$7.93 billion (US$251.44 million), down 9.48 percent from NT$8.76 billion a year earlier. On a quarterly basis, revenue fell 10.83 percent from NT$8.89 billion, company data showed. The figure was also lower than market expectations of NT$8.05 billion, according to data compiled by Yuanta Securities Investment and Consulting Co (元大投顧), which had projected NT$8.22 billion. Makalot’s revenue this quarter would likely increase by a mid-teens percentage as the industry is entering its high season, Yuanta said. Overall, Makalot’s revenue last year totaled NT$34.43 billion, down 3.08 percent from its record NT$35.52
PRECEDENTED TIMES: In news that surely does not shock, AI and tech exports drove a banner for exports last year as Taiwan’s economic growth experienced a flood tide Taiwan’s exports delivered a blockbuster finish to last year with last month’s shipments rising at the second-highest pace on record as demand for artificial intelligence (AI) hardware and advanced computing remained strong, the Ministry of Finance said yesterday. Exports surged 43.4 percent from a year earlier to US$62.48 billion last month, extending growth to 26 consecutive months. Imports climbed 14.9 percent to US$43.04 billion, the second-highest monthly level historically, resulting in a trade surplus of US$19.43 billion — more than double that of the year before. Department of Statistics Director-General Beatrice Tsai (蔡美娜) described the performance as “surprisingly outstanding,” forecasting export growth
RSS