China turns to US-made AI chips to boost hypersonic weapon performance

In mid-April 2024, Chinese press announced that Chinese researchers working on a joint project between Dalian University of Technology (DUT) and Beijing Power Machinery Research Institute (BPMRI) had used a US-designed artificial intelligence (AI) semiconductor chip to boost the performance of the scramjet engine on a hypersonic weapon.

The team published their results in mid-March in a paper featured in Propulsion Technology, a Chinese academic journal. The team, headed by Professor Sun Ximing, said in its paper it had placed a Nvidia Jetson TX2i GPU computer unit into a hypersonic vehicle able to travel through the air at speeds of greater than Mach 7.

It found the chip improved the speed at which computational tasks in the scramjet engine could be performed considerably, saying it enhanced “real-time optimisation of the fuel supply system, fault diagnosis and fault-tolerant control in scramjet engines”.

Related Articles

China’s AI developments in electronic surveillance extends to battlefield

US hypersonic programmes to receive almost $7 billion in 2025

UK brings 90 companies into $1.25 billion hypersonic missile programme

The researchers cautioned that further work was needed to apply the AI chip to hypersonic weapons in areas such as shockwave correction and data modelling. Nevertheless, the Chinese team’s test suggested that even relatively unsophisticated AI chips have the potential to eventually be integrated into Chinese or other states’ growing hypersonic missile arsenals.

According to Chinese media, Nvidia’s TX2i has been designed for industrial applications and was released six years earlier, making it extremely unlikely China would seek to use the chip model in future hypersonic missile projects. The design was far below the performance levels of Nvidia’s most advanced H100 chip design and below the threshold which triggers US export controls on AI chips to China.

China’s homegrown semiconductor chip industry has been capable of manufacturing designs that can match or exceed the performance of the TX2i. Chinese researchers may have selected a foreign semiconductor design simply to see if Chinese hypersonic weapon designs could operate using foreign chips.

China, however, has been prioritising the expansion of its own semiconductor manufacturing sector to assuage its own technological independence and national security concerns, which include safeguarding supply chains for critical military parts and equipment – particularly of mature chip designs of similar levels of capability to the TX2i.

The successful integration of dual-use legacy chip designs into hypersonic missiles would allow China’s armed forces to accelerate production of these systems due to the ease of access to such chips, as well as increasing the effectiveness of their missiles compared to current-day designs, despite achieving no technological breakthrough.

Chinese hypersonic missile systems like the DF-27 pose a potent threat to US naval power in the Pacific, even if work to counter the threat they pose to important naval units like US aircraft carriers has already been underway.

Dr Tom Stefanick, a non-resident senior fellow at the Brookings Institution’s Strobe Talbott Center on Security, Strategy and Technology, said open sources showed that China’s hypersonic weapons systems presently posed a theatre-range threat capability. This included the Pacific’s first and second island chains (regions such as Taiwan or parts of the Philippines), and areas covering Guam, the South China Sea and Japan.

Stefanick commented that this put areas with a US military footprint such as Guam, Okinawa, and US ships and bases in Japan, the Philippines and elsewhere in east Asia at risk of Chinese hypersonic missile attacks. He expressed some scepticism, however, that China required access to AI chips to improve the potency of its hypersonic missile systems and added that, in any case, there were many alternatives should China need to improve its designs without such chips.

“There is no evidence that AI algorithms are broadly superior to alternative algorithms for hypersonic guidance and control, for example,” Stefanick explained. “AI algorithms are very good for image recognition when images are very clear, but fog and clouds are common over the Pacific, so optical sensors are useful only part of the time.

“In such cases, AI algorithms do not perform significantly better than alternative methods,” he added.

Dr Stefanick did highlight the potential use for AI in generating deceptive signal patterns that would allow a missile attack to fool pattern recognition by enemy satellites using systems such as synthetic aperture radar (SAR) recognition. Such AI-enabled sensor blocking has wider threat implications beyond a regional US–China war in the Pacific when combined with the potential speed and range of some countries’ hypersonic missile arsenals.

Russia, which also possesses hypersonic missiles, has said some of its weapon designs could carry nuclear warheads, thereby increasing the threat of nuclear war, should one nuclear-armed country launch a hypersonic missile attack during a standoff with another nuclear-armed power – especially if the defender was unable to locate a missile accurately or determine if it was armed with a nuclear warhead.

Such a risk has encouraged some US officials to push for AI-enabled integration and automation of sensor data for the country’s surveillance systems defending the US homeland in recent years. Their aim has been to use AI systems to increase US response time to any airborne or missile attack using hypersonic technology, pointing to China and Russia’s pursuit of AI and hypersonic weapons as a counter to present-day US air and naval superiority. The US and other allies, including the UK, have also been pursuing their own hypersonic missile technology programmes.