Macquarie said that AI computing is bringing positive changes in sales and average selling prices to semiconductor and hardware suppliers.

(AFP file photo)

[Financial Channel/Comprehensive Report] Macquarie Research said that artificial intelligence (AI) computing is bringing positive changes in sales and average selling prices to AI semiconductor and hardware suppliers.

Macquarie predicts that by 2026, the compound annual growth rate of the AI ​​field, including GPUs (graphics processing units), ASICs (application-specific ICs), and FPGAs (field programmable gate arrays), will reach 17%. Bring tempting opportunities.

Contrary to Moore's law, AI drives up performance and cost. Macquarie estimates that in AI servers, the value of semiconductors is 9-18 times higher than that of ordinary servers.

The inference server (inference server) is about US$33,000 (about NT$1.005 million), the training server (training server) is US$65,000 (about NT$1.98 million), and the general enterprise server is about US$36,000 (about NT$1.98 million). 1.096 million).

Please read on...

Since 2017, the performance of GPUs has increased by 10 times and the cost has been reduced by about 3 times.

Macquarie expects that 3nm and 2nm will continue to drive performance improvements and relative cost reductions.

This will also drive sales growth and average price growth in the AI ​​semiconductor supply chain.

Macquarie and IDC predict that the x86 server CPU (central processing unit) market will maintain a market value of US$25 billion (approximately NT$761.6 billion) in 2022, with a compound growth rate of 3% by 2026; the market value of accelerators in 2022 will be 19 billion US dollars (approximately NT$578.8 billion), with a compound growth rate of 17% by 2026; Macquarie predicts that the annual growth rate of ABF substrates will be 19% by 2025.

The Macquarie report pointed out that training astronomical parameters requires high computing power, but the problem of low bandwidth or memory walls of DRAM (Dynamic Random Access Memory) is more obvious in the AI ​​​​world.

Upgrading the bandwidth of High Bandwidth Memory (HBM) should be challenging and costly.

At the same time, HBM seems to be the only choice for various accelerators from Nvidia to Google TPU (tensor processor) and other ASICs.

Macquarie said that in the Asian technology market, the only industry that sees a steady 15-20% CAGR in revenue over the next five years is hyperscale capital expenditure and AI investment, which can be used as part of a long-term investment portfolio.

In addition to TSMC (2330), which was originally bullish, Macquarie also named 9 other Asian stocks, calling for buying Taiwan stocks: Zhibang (2345), Alchip-KY (3661), Xinhua (5274) ), Wistron (3231), Wiwynn (6669), South Korea’s Samsung (Samsung), SK Hynix (SK Hynix) and Japan’s Ibiden (Ibiden).

Grasp the pulse of the economy with one hand I subscribe to Free Finance Youtube channel

Already added friends, thank you

Welcome to 【Free Finance】

feel good

Already liked it, thank you.

related news