Huawei has found a chip-stacking workaround which could make US sanctions look less lethal than Washington hoped.
For those who came in late, when the US government moved in May 2019 to cut Huawei off from American chips, software and semiconductor manufacturing technology, many analysts treated it as a death sentence.
That night, HiSilicon head He Tingbo sent an internal letter calling the ban the “darkest of days”. She revealed Huawei had spent almost 10 years preparing for what she called an “extreme survival scenario”.
According to a copy of the letter seen by the Financial Times, Huawei had developed backup chips across its product range. The plan was based on the fear that foreign suppliers could one day be cut off.
For years, Huawei kept the work mostly out of sight. At a semiconductor conference in Shanghai in May, it announced it had found a way around a major US export control problem.
China cannot buy the world’s most advanced chipmaking kit. Huawei’s answer is a chip-stacking approach designed to boost computing performance without the latest manufacturing tools.
Bernstein analysts described the announcement as another “DeepSeek moment”. That was a nod to the Chinese AI lab whose models challenged assumptions about the gap between China and Silicon Valley under export controls.
Huawei’s comeback has become one of the biggest stories in the US-China technology scrap. Seven years after many wrote it off, the company is betting it can push China towards semiconductor self-sufficiency.
He, now 56 and nicknamed China’s “chip queen”, presented Huawei’s logic-stacking technology. It folds chip circuits into multiple layers to increase computing performance without relying on ever-smaller transistors.
China is barred from acquiring EUV lithography machines, the most advanced chipmaking tools made by Dutch group ASML. Many analysts believed that would stop Chinese chipmakers moving beyond 7-nanometre processes.
That was widely seen as the limit of what could be done with older equipment.
Bernstein analyst Lin Qingyuan said: “It breaks the core narrative that because of export controls China’s semiconductor is dead at 7nm. Huawei needed to demonstrate it works also to show the government that it isn’t wasting money and has a future.”
Huawei believes the technology is ready for smartphone chips due later this year, according to people familiar with its plans. That does not mean it is sorted.
Power consumption, overheating and manufacturing yields still look like awkward problems. Data centre AI chips are an even bigger headache.
He said in Shanghai that the technology would not be ready for AI processors until 2030. Industry experts say the problem is the engineering complexity of stacking more layers of dense logic and memory chips.
Huawei has another workaround. It wants to compensate for weaker individual chips by wiring large numbers of them into powerful computing clusters. That uses the networking expertise Huawei built in telecoms. Its CloudMatrix 384 platform connects hundreds of AI processors into one system.
Huawei claims CloudMatrix 384 beats Nvidia’s widely used NVL72 in overall computing power and memory. The trick is using brute-force scale to make weaker chips look less weedy.
Huawei is aiming for much larger clusters with its latest AI chips. People familiar with the plan said the company hopes they will be capable of AI training work. Advanced packaging and 3D chip designs are not unique to Huawei. TSMC and other chipmakers explored similar approaches years ago, while several Chinese startups are chasing comparable AI chip designs.
Some of those startups are preparing prototypes this year, which could match Nvidia’s B200 chips if the technology works. Investors said such products would not enter production until mid-2027 at the earliest.
Huawei, HiSilicon, China chips, US export controls, AI chips, Nvidia, ASML, semiconductor sanctions, chip stacking
