Hi, this is Yifan from Silicon Valley, your #techAsia host this week. It’s getting a bit chilly in the San Francisco Bay Area. The smell of fall reminds me we are now in the season of Halloween, pumpkin spice lattes — and of course third-quarter tech earnings.
While the ultimate barometer of artificial intelligence demand, Nvidia, won’t report its earnings until next month, we’re already seeing some pretty telling numbers from other chip companies about where the AI boom is heading.
ASML, the Dutch chipmaking equipment supplier, surprised the market (and everyone else) last week with a pessimistic sales outlook that dragged down all major chip stocks as investors question whether the AI-driven chip craze is over. Samsung’s operating profit fell almost 13 per cent in the third quarter as it struggled to supply its latest AI memory chip.
But TSMC soon came out saying the “insane” AI demand is real and will continue to benefit the world’s biggest contract chipmaker. And SK Hynix, Samsung’s main rival on the high-performance AI memory front, just reported a record quarterly profit.
It is not hard to see that not all chipmakers are benefiting equally from the generative AI boom, much like Nvidia has reaped most of the benefit from demand for AI chips. The question is, will those falling behind eventually catch up, or will the gap between the market leader and others only grow?
Blacklisted companies unite
Chinese artificial intelligence giant SenseTime Group is using “a lot” of domestic chips, including from Huawei Technologies, to train its large language models, as part of efforts to rapidly develop a homegrown AI computing ecosystem.
SenseTime’s vice-president of Asia-Pacific operations told Nikkei Asia’s Cissy Zhou, Cheng Ting-Fang, and Lauly Li that the company’s next-generation Artificial Intelligence Data Centre in Shanghai is equipped with chips from all the main domestic manufacturers, including Huawei and Biren Technology.
All three of these companies have been put on trade blacklists by Washington, meaning they have no access to the most advanced AI chips or chipmaking tools from the US.
Huawei, meanwhile, is not only working to hone its semiconductor prowess. The company announced this week that airlines Emirates and ride-hailer Grab have launched apps for its HarmonyOS, a significant step forward for the Chinese company’s bid to crack overseas markets after US trade restrictions cut off its access to American operating systems.
AI on a budget
Chinese artificial intelligence groups are finding creative ways to drive down the cost of building competitive models, leveraging their strengths to lessen the impact of US chip restrictions.
Tech groups 01.ai, DeepSeek, MiniMax and Stepfun are all choosing a technically challenging but cheaper way to build AI models using small data sets, writes the Financial Times’ Eleanor Olcott. This allows them to offer developers more affordable models to build AI apps.
At the same time, there is a vicious price war in China to slash inference prices — the price of calling upon large language models to generate a response — to a fraction of that offered by US counterparts.
This has made it easier for Chinese start-ups seeking to replicate the viral success of ByteDance’s TikTok video-sharing app to launch new AI products by leveraging the country’s vast pool of talented and cheap engineers.
Not for everyone
Only a few top chip developers will be able to continue investing in semiconductors for AI computing given the massive amounts of money needed to stay competitive, the chief technology officer of US chipmaker Marvell told Nikkei Asia’s Cheng Ting-Fang in this exclusive interview.
Noam Mizrahi, Marvell’s CTO, said the market for customised artificial intelligence chips will grow sharply, however, even if “not many” players can afford to stay in the game.
Marvell is the world’s second-largest networking chip provider after Broadcom, with products including connectivity chips for data centres, telecom and networking gear.
While off-the-shelf products from Nvidia and Advanced Micro Devices remain popular, leading cloud service providers such as Google, Microsoft, Amazon and Meta are investing heavily to develop their own chips for data centres, which creates big tailwinds for Marvell.
But Mizrahi said the AI chip game is not for everyone. “It’s a play [ground] for those who have the scale and who can invest over a long period of time.”
A good day for Tesla
Tesla has had a rough month. After its highly anticipated robotaxi, unveiled on October 10, failed to impress investors, the company’s shares entered a sharp and protracted fall. But it had a surprising comeback on Wednesday after reporting an 8 per cent year-on-year net income growth in July-September, Nikkei Asia’s Yifan Yu reported.
Profitability has been a black eye for Tesla earnings this year as the US electric vehicle maker faces weakening EV demand in key markets, including China, where heated competition has pushed local players like BYD to cut prices. Tariffs on China-made EVs, meanwhile, threaten the future of its main export hub and most cost-effective plant, the Shanghai Gigafactory.
But on Wednesday, Tesla said the Shanghai factory exported its millionth vehicle in September and achieved the lowest per-vehicle production cost during the third quarter.
Suggested reads
-
TSMC says it alerted US to potential violation of China AI chip controls (FT)
-
Carmakers double down on hybrids as Vietnam sales slump (Nikkei Asia)
-
KKR and Bain in all-out $4bn fight for Japan’s Fuji Soft (FT)
-
Panic spreads in India over ‘digital arrest’ scam (Nikkei Asia)
-
TSMC profits jump 54% on back of AI chip boom (FT)
-
China’s ecommerce industry pushes back on ‘no-return’ refunds (Nikkei Asia)
-
Nikon seeks to diversify beyond cameras and chips amid geopolitical tensions (FT)
-
AI diagnosis tech to help Thailand, Brazil navigate doctor shortages (Nikkei Asia)
-
Japan Display suspends OLED fab talks in China, clouding recovery (Nikkei Asia)
-
China asks: what is an e-bike? (FT)
Read the full article here