Huawei Aims to double chip power every year

[retaxis]

While the EUV machine is being built and logic folding is in the works, Huawei has been keeping busy with brute forcing and doubling the huawei chips compute power every year. By the time 2030 rolls around China would have their own EUV machine and own EUV chips (according to reports). Therefore this will plug the gap for the time being until China is able to roll out the EUV tech.

According to Huawei, the company aims to double its chip computing power every year through an accelerated annual release cycle. [1, 2]

Chip Roadmap

• • Ascend 960DT: Scheduled for the first quarter of 2027.
  • Ascend 960PR: Scheduled for the third quarter of 2027.
  • Ascend 970: Planned for release in 2028.
  • Ascend 980: Planned for release in 2029. [1]

Key Technologies

• • Tau Scaling Law: A custom development framework that prioritizes data speed and system efficiency over shrinking physical transistors. [1]
  • LogicFolding: An architectural method that stacks active circuitry vertically to increase transistor density without advanced lithography machines. [1]
  • SuperClusters: Systems that link thousands of processors together to match the power of major competitors like Nvidia. [1, 2]

Chinese tech giant Huawei has made a bold move in the AI "race," announcing that its next-generation AI chip to rival current Nvidia models will be released as early as the start of 2027. This has the potential to release China from the restrictions the US has imposed on its AI development.


The launch of Huawei's compute-comparable next-generation Ascend 960DT AI chip has been moved from Q3 2027 to Q1, which may not mean much to a lot of us, but it's a big deal.


At the end of August 2022, the US blocked China from accessing any new Nvidia technology. Because at the time, all Nvidia chip hardware was made at the Taiwan Semiconductor Manufacturing Company (TSMC) mega-factory. While protecting Taiwan's autonomy was cited as the main reason, in hindsight it feels like it was far more to do with a different geopolitical battle: the control of AI.

Skip forward to January 2025, and the launch of China's DeepSeek model took the world by surprise. It was powerful – but ran on old Nvidia compute technology. Less than three months later, Nvidia CEO Jensen Huang announced a massive shift in chip production to the US.


Essentially, access to advanced chips unlocks AI computing power, which is really all you need to know. However, this has been increasingly referred to as a modern-day Cold War between China and the US, even if both countries are running completely different "arms races."

At this week's Huawei Connect conference, the company announced it was moving the release of its new chip forward to early 2027, which has the potential to reshape China's AI ecosystem. Huawei's rotating and acting chair David Wang, in his keynote speech, revealed the new chip would surpass the power of previous models.

"We're evolving our Ascend chip series on a one-generation-a-year cycle," said Wang. "In 2028 and 2029, we will roll out the Ascend 970 and 980 chips, respectively. Thanks to the Tau Scaling Law, not only will their compute specifications continue to double, but you can also expect to see huge improvements across the board in terms of memory bandwidth, memory capacity, interconnect bandwidth, and more."


He also introduced Huawei's plan to power China's agentic AI (or AGI) world, which we haven't heard a lot about. You can read about it here, but essentially it underpins how different the US is approaching the age of AI by comparison. While the focus has been on compute power in the West, with frontier companies like OpenAI and Anthropic leading the way, Chinese engineers have had to design AI infrastructure differently with the chips they've had access to. So partly – not wholly – the ecosystem has prioritized open-source innovation.

However, even if 100,000 software-engineer brains may be better than the 100 or so currently at Anthropic, there are limitations that come down to hardware. This was revealed in a leaked call from DeepSeek founder Liang Fengweng with investors in July.

The Ascend 960DT chip, however, could change this.

We'll have to wait and see if it can match Nvidia's Blackwell GPU chip, which company founder Jensen Huang said this week is expected to double in sales in 2027.

Interestingly, in his keynote conclusion, Wang noted that while AI "may well be the final technological revolution in human history," its advancement to benefit societies around the globe relies on collaboration, not tech isolationism. "No single company," he said, "can build an intelligent world alone."

And even in the wake of news this week that China has opened the world's first "smart factory" where robots build robots, people are reportedly much more optimistic about the future of AI and its impact on their lives, versus the US. This may be an overlooked aspect of the so-called AI arms race that could shape it as much as the technology itself

 

[retaxis]

Huawei is heavily involved in China's national push to develop a homegrown extreme ultraviolet (EUV) lithography machine to bypass strict U.S. export controls. [1, 2]

Homegrown EUV Prototype and Testing

• • Dongguan Facility: Reports indicate that Huawei has been testing an experimental, homegrown EUV prototype at its facility in Dongguan, China. [1, 2]
  • Laser-Induced Discharge Plasma (LDP): Unlike ASML's laser-produced plasma method, this system reportedly uses an LDP approach—vaporizing tin between electrodes with a high-voltage electrical discharge to generate the critical 13.5-nanometer EUV light wavelength. This design aims to be simpler, smaller, and more energy-efficient. [1, 2]
  • Current Status: While prototypes exist and tests have taken place, industry experts and reports note that the machine fills a factory floor and still faces major hurdles regarding stable throughput, commercial yield, and mass-production readiness. Many analysts estimate that viable commercial chip production using domestic Chinese EUV may not stabilize until 2028–2030. [1, 2, 3]

Reducing Dependence and Alternative Strategies

• • Bypassing EUV for Now: Because true commercial EUV remains out of reach, Huawei has worked to reduce dependence on ASML by optimizing older deep ultraviolet (DUV) machinery with multi-patterning techniques (such as self-aligned quadruple patterning) to produce advanced nodes. [1, 2, 3]
  • Accelerated AI Chips: Huawei recently accelerated its roadmap, pushing the launch of its non-EUV Ascend 960 AI chip ahead of schedule, signaling that the company continues to scale performance without waiting for domestic EUV tools to mature. [1]