Huawei has announced a comprehensive roadmap for its Ascend AI chips, planning four new releases by 2028, signaling a direct challenge to Nvidia's dominant position in the artificial intelligence chip market.

The Technology Sector and the broader Semiconductor Industry are experiencing significant shifts following Huawei Technologies Co.'s unveiling of an ambitious multi-year roadmap for its Ascend series of artificial intelligence (AI) chips. The Chinese technology giant aims to introduce four new Ascend AI chip iterations by 2028, explicitly positioning itself to compete directly with Nvidia Corp. (NVDA), the current leader in AI semiconductor solutions.

The Event in Detail

Huawei's detailed three-year AI chip roadmap outlines annual upgrades to its AI and server chips, with a stated goal of doubling computing power with each generation. The company's current offering, the Ascend 910C, manufactured using a 7nm process by SMIC, delivers 800 TFLOP/s in FP16 mode and 3.2 TB/s memory bandwidth, presenting a viable alternative to Nvidia's H100 GPU. Analysts project Huawei could ship over 700,000 Ascend 910 series AI chips in 2025, despite an estimated 30% yield from SMIC's 7nm process. Reports indicate the Ascend 910C achieves approximately 60% of the H100's inference performance.

Looking ahead, Huawei plans for substantial advancements:

  • The Ascend 910D, scheduled for testing by late May 2025, is designed to rival Nvidia's Blackwell B200, targeting a theoretical peak of 1.2 PFLOP/s and offering HBM3e memory with 4 TB/s bandwidth.
  • Q1 2026 will see the launch of the Ascend 950PR chip, featuring proprietary HBM (HiBL 1.0) with 128GB memory, 1.6TB/s bandwidth, 1 PFLOPS of FP8 compute, and 2 PFLOPS of FP4 performance.
  • This will be followed by the Ascend 950DT in Q4 2026, and the Ascend 960 chip in Q4 2027, boasting 288GB HBM and 2.2TB/s interconnect bandwidth.
  • The roadmap culminates with the Ascend 970 chip in Q4 2028, projected to directly compete with Nvidia's Blackwell series by that year. Huawei has also confirmed upgrades to its Kunpeng server chips in 2026 and 2028.

Analysis of Market Reaction

Nvidia shares (NVDA) experienced a -2.62% change following these developments and related geopolitical tensions. The market's reaction reflects increased investor scrutiny on Nvidia's competitive moat, particularly given its significant exposure to the Chinese market, which historically contributed 20-25% of its revenue. Recent reports indicate that major Chinese tech firms have been directed by Beijing to halt purchases of Nvidia's cutting-edge AI chips, further intensifying competitive pressures for the U.S. chipmaker. This follows a series of U.S. export restrictions that initially banned Nvidia's A100 and H100 chips, and subsequently less powerful "export-compliant" versions like the A800, H800, and H20.

Adding to Nvidia's challenges, China's State Administration for Market Regulation (SAMR) has reportedly launched an antitrust investigation into Nvidia's 2020 acquisition of Mellanox Technologies. This probe could result in substantial penalties, potentially reaching up to $1.03 billion, representing 1% to 10% of Nvidia's annual revenue generated in China. Nvidia CEO Jensen Huang has acknowledged the severity of the situation, stating that China's $50 billion AI chip market is now "effectively closed to U.S. industry" for Nvidia's most advanced data center products. Consequently, Nvidia's dominant 90-95% share of China's AI chip market has reportedly declined to approximately 50%, with domestic competitors gaining ground.

Broader Context & Implications

Huawei's aggressive strategy is intrinsically linked to China's broader push for technological self-reliance, significantly accelerated by U.S. export controls. These restrictions have inadvertently fostered a protected market environment for domestic champions like Huawei, SMIC, and SiCarrier, boosting China's chip self-sufficiency from 18% in 2023 to a projected 28% by 2025. The Chinese government's substantial investment in AI infrastructure, including a $53 billion pledge from companies like Alibaba for cloud computing and AI over three years, further underpins this domestic growth.

To counter performance disparities arising from reliance on older fabrication technologies, Huawei is employing a "group computing" strategy, where multiple chips work in tandem. This modular approach includes the development of Atlas 950/960 supernodes, capable of housing between 8,192 and 15,488 chips. While Huawei's individual chip performance may lag behind Nvidia's 4nm and 3nm offerings due to its 7nm technology, its emphasis on cost-effectiveness and clustered architectures aligns with China's strategy to leverage domestically produced hardware. Reports also suggest Huawei has relied on a "Die Bank" of chips manufactured by TSMC to meet its 2025 production targets, a move that could potentially entail export control implications.

Expert Commentary

Industry analysts suggest these developments signal the emergence of a bifurcated global semiconductor market. One ecosystem will likely remain Nvidia-led in Western markets, while a distinct Huawei-centric alternative gains traction in China. This transition is not merely about component substitution but a recalibration of value chains, with Huawei's CANN software stack aiming to replicate CUDA's developer appeal. For investors, the key lies in balancing exposure to Huawei's ecosystem partners with a cautious assessment of manufacturing and software ecosystem risks, recognizing that while Nvidia maintains its performance and ecosystem maturity edge, Huawei's cost-effective, clustered solutions are carving a distinct niche.

Looking Ahead

The coming years will be crucial for the global Semiconductor Industry. Investors will closely monitor Huawei's ability to scale its production, particularly its reliance on domestic manufacturing capabilities and potential workarounds for export control limitations. The performance benchmarks of its upcoming Ascend chips against Nvidia's future generations will be a key determinant of its market penetration. Furthermore, the evolution of Huawei's software ecosystem and its ability to attract developers will be pivotal. The ongoing geopolitical dynamics, including potential shifts in U.S. export policies and China's continued drive for technological self-sufficiency, will undoubtedly shape the competitive landscape and investment opportunities within the AI chip market.