AI/ML News & Innovations Hub

Hello from Yifan, your #techAsia host this week. I'm sending this newsletter from New York after spending an afternoon in a cold ballroom where Qualcomm executives laid out the company's data center chip ambitions.

It was a busy day for semiconductors. After Nvidia hosted its annual shareholder meeting, where Jensen Huang said AI infrastructure demand will continue for "decades," memory chipmaker Micron reported blockbuster earnings. After a scary few days that saw SK Hynix and Samsung shares briefly plunge, this came as a reassurance for investors that the feared "AI bubble" is not about to burst.

But arguably Qualcomm, the company known for its mobile chips, was the star of this busy day.

At its annual investor day in New York, Qualcomm unveiled more details about its data center chip roadmap, from a new AI accelerator platform to CPUs, ASICs and connectivity chips. Apart from the air conditioning in the ballroom being out of control -- CEO Cristiano Amon welcomed reporters to "Siberia" -- it was a great day for the company, as its stock shot up more than 14% at one point following the investor meeting.

I was particularly intrigued by a new chip architecture Qualcomm calls high-bandwidth compute (HBC). It's supposedly a better design than the current GPU-plus-high-bandwidth memory (HBM) setup in AI racks because it places memory closer to computing power, an approach known as near-memory compute. This reduces how far data must travel between the two, saving time and energy.

Memory chips have become the latest bottleneck in AI development, as we've reported extensively at Nikkei Asia. Amon said a new design like HBC will help ease the global memory crunch that some are predicting will last till 2030.

This made me wonder: Both SK Hynix and Samsung have ridden the AI chip boom high, especially with the growing need for HBM for AI data centers. Will a new technology like HBC, that supposedly reduces the need for HBM, further accelerate their rise in the AI era, or are the good days for memory chip makers coming to an end?

Qualcomm in China

Qualcomm unveiled its data center chip lineup on Wednesday, becoming the latest chipmaker to enter the AI processor race in an attempt to challenge market leader Nvidia.

CEO Cristiano Amon told Nikkei Asia's Yifan Yu that the company is eyeing the China market for its data center products, including designing chips specifically for Chinese customers that are in compliance with U.S. export controls.

China accounted for 46% of Qualcomm's revenue in 2025, mostly from smartphone chips.

Amon said as the company diversifies its business, its partnerships with customers in China are also expanding, while its existing relationship with Chinese smartphone makers and OEMs is a "strength" it can naturally bring to the data center business in the country.

Nvidia in demand

Nvidia's AI chips have more than doubled in price on China's black market as a U.S. crackdown on illicit exports collides with strong demand from Chinese companies, writes the Financial Times' Eleanor Olcott.

The price of the U.S. tech giant's flagship DGX B300 server has soared to more than 8 million yuan ($1.1 million) over the past six months, up from 4 million yuan, according to FT interviews with multiple Chinese chip traders.

The system, which contains eight Blackwell graphics processing units, typically retails in the U.S. for about $400,000. Nvidia's RTX 6000 Pro workstation chip, a popular option for startups deploying large language models, has risen in cost from about 50,000 yuan at the start of the year to as much as 130,000 yuan, according to traders. The RTX 6000 and DGX B300 are barred from export to China under Washington's chip controls.

The price rises underline how Washington's latest clampdown has squeezed the black-market routes used to move restricted chips into China, while also showing that demand for Nvidia's processors remains strong despite Beijing's push to replace them with domestic alternatives.

"The loopholes have shrunk. It is becoming more and more risky for intermediaries to trade these chips as prices have surged," said one trader who sells to large data center clients.

Man's new best friend?

Australia has more than 270,000 livestock dogs, and collectively they contribute around 3 billion Australian dollars ($2.1 billion) in value to the country's economy. But is man's best friend about to be replaced by drones?

As a new generation of farmers takes the reins, famous breeds like kelpies, collies and cattle dogs are no longer the only option to help move livestock more efficiently, particularly out in the paddocks and pastures, where drones, artificial intelligence and other technologies are offering new ways of mustering -- or herding -- sheep and cattle across vast distances, Nikkei Asia's Shaun Turton writes.

A 2024 Roy Morgan survey of 1,001 farmers found that 72% were using ag-tech in their operations. Farm management software was the leading technology, followed by electronic tags for animals, satellite analysis and GPS-guided farming. Drones were the fifth-most common, followed by remote sensors.

Going big in Japan

Blackstone plans to invest $30 billion in artificial intelligence data centers in Japan over the next three to five years, the asset manager's President and Chief Operating Officer Jonathan Gray told Nikkei's Haruki Kitagawa in an interview.

So far, Blackstone has developed data centers in Japan with a combined capacity exceeding 500 megawatts. Gray told Nikkei that the company is in discussions to develop facilities exceeding 1 gigawatt, equivalent to the capacity of a nuclear reactor, adding that the risk of not building enough computing resources outweighs AI bubble concerns.

Blackstone will also accelerate its private equity investments in Japan. In June, it launched its largest Asian fund yet, at $13.1 billion.

Podcast: Tech Latest

Welcome to the Tech Latest podcast. Hosted by our tech coverage veterans, Katey Creel and Shotaro Tani, every Tuesday we deliver the hottest trends and news from the sector.

In this episode, Shotaro speaks with chief tech correspondent Annie Cheng Ting-Fang about how AI-driven demand is pushing companies like BYD, Google, AMD and Tesla to turn to Samsung for chipmaking as TSMC's capacity comes under increasing strain.

For more great stories like this delivered to your inbox every week, sign up to our #techAsia newsletter. Current subscribers, click here to update your newsletters preferences.