The inside story on the Asia tech trends that matter, from Nikkei Asia and the Financial Times
Hi everyone! This is Cheng Ting-Fang, your #techAsia host this week, saying hello from sunny Taipei, the blazing-hot destination for global chip CEOs.
Over the past few days, amid scorching temperatures, I found myself squeezed into crowds waiting for AMD CEO Lisa Su as she outlined the chipmaker's latest artificial intelligence ambitions. Soon after, I was rushing to Nvidia events, where CEO Jensen Huang said his company's spending on Taiwan's AI supply chain has reached as much as $150 billion a year.
And the wave of executives is only growing. Leaders from Intel, Qualcomm, Arm and Marvell, all among the world's most valuable chipmakers, will be arriving in Taiwan in the coming days. They are not here just for the Computex industry event. They are eager to lock in supplies: of chips, packaging resources, substrates, printed circuit boards, cooling systems and power equipment, which are all needed to sustain the global AI buildout.
It's a fascinating time to be a tech reporter. But just as the global AI boom is straining supplies of everything from processors to memory chips, I often feel my own central processing unit -- my brain -- and memory capacity are running critically short as well. The pace of development is so relentless that I sometimes think I need a massive infrastructure upgrade just to keep up.
One of the hottest buzzwords in the chip industry used to be CoWoS, Taiwan Semiconductor Manufacturing Co.'s advanced packaging technology that helps companies such as Nvidia, Google and Amazon combine powerful AI processors with memory chips. In Taiwan, CoWoS became such a mainstream topic that you could hear even retired aunties and uncles casually discussing it over coffee and tea.
Now, a new set of catchy but hard-to-understand phrases is taking over: COUPE, short for compact universal photonic engine, and CPO, or co-packaged optics. COUPE is a TSMC technology that integrates photonic and electronic chips into a single package, while CPO refers to placing these optical engines directly next to the main processor to dramatically speed up communication between chips and servers. CPO could save space, reduce power consumption and enable much faster data transmission. To better understand this new world, I have spent many hours receiving informal masterclasses from industry insiders.
AI infrastructure today still relies heavily on lipstick-sized devices known as pluggable optical transceivers and on extensive amounts of copper wiring, largely because copper remains much cheaper than fiber. But industry leaders from TSMC, Intel and Nvidia to Broadcom increasingly see CPO as a key technology for the next generation of AI infrastructure. Many executives are calling this year the dawn of the CPO era, though adoption will initially be limited to the industry's most advanced AI systems.
But even before CPO becomes mainstream, the growing convergence of optics and electronics, along with expanding fiber usage and surging demand for traditional optical transceivers, is already reshaping supply chains. The shift is creating fresh business opportunities, while also triggering unexpected shortages in some of the industry's most obscure corners.
The global AI boom has triggered an unprecedented surge in demand for optical communications infrastructure, leading to unexpected chokepoints and price increases across the entire supply chain, including fibers, transceivers, lasers and substrates, Nikkei Asia's Lauly Li and Cheng Ting-Fang report.
Optical fiber, for example, has become a strategic resource in the AI data center buildout. Nvidia has signed a long-term agreement with Corning to expand the latter's U.S. fiber capacity tenfold. Laser sources used for optical transmission have also emerged as a critical bottleneck. Supplies of indium phosphide substrates, a key material for making lasers, are increasingly constrained, a shortage few in the industry had anticipated. China's export controls on indium have further complicated the situation.
China plays a key role in the global market for pluggable optical transceivers, with notable players such as Zhongji Innolight, a major supplier for Google, and Eoptolink, a key provider for Nvidia and Amazon. Both companies have seen their profits skyrocket by over 800% and 900%, respectively, since the AI boom began in 2022.
Global buyout firms are exiting China's data center sector as political and regulatory pressures bite, write the Financial Times's Zijing Wu and Arjun Neil Alim.
Princeton Digital Group, backed by Warburg Pincus, is launching a sale of its China assets that could fetch as much as $1 billion, according to people familiar with the process. A sale of PDG, which owns data centers across six Chinese cities, would mark the near end of a decade-long push by global private equity firms to invest directly into China's AI infrastructure.
It follows a wave of similar deals by groups such as Bain Capital and Carlyle, as foreign ownership of critical digital infrastructure becomes increasingly fraught in the country.
From 2017, buyout groups poured billions into the Chinese data center sector, drawn by booming demand from cloud providers owned by Alibaba, Tencent and ByteDance.
But rising AI-driven demand has meanwhile lifted valuations, giving international investors an opportunity to sell to domestic buyers and redeploy capital into politically safer markets elsewhere in Asia where they also see significant growth.
Six years after Huawei was cut off from global semiconductor production, the Chinese tech giant has outlined an ambitious chip comeback strategy starting with new Kirin smartphone processors that it believes can help the company overcome severe U.S. export controls.
The road map was unveiled by Huawei's longtime semiconductor chief, He Tingbo, a low-profile executive who has rarely spoken publicly, Nikkei Asia's Cheng Ting-Fang and Lauly Li write.
She introduced a new design approach and tech road map aimed at reshaping Huawei's chip development around a stark reality: China still lacks access to advanced equipment such as EUV, or extreme ultraviolet, lithography machines needed for cutting-edge chip production. The chip chief also provided rare comments on the challenges Huawei still faces to boost the quality and volume of chip production to meet massive local demand.
