Not conclusively. Huawei’s rotating chairman said in September 2026 that he believed Ascend had a larger share of China’s AI-chip market than Nvidia, but Huawei provided no supporting figures. A separate estimate put the companies roughly level in 2025 and forecast a Huawei lead in 2026. Neither establishes that Ascend has overtaken Nvidia across every measure of performance, workload or adoption.
Has Huawei Ascend surpassed Nvidia in China?
It depends on what “surpassed” means—and the available market-share figures are estimates, not an audited count. On September 17, 2026, Huawei rotating chairman Eric Xu said reliable data on Nvidia’s share in China were difficult to obtain. He added, “I think Ascend market share should be bigger than Nvidia’s.” Reuters reported that Huawei offered no data to support that view. Reuters’ account of Xu’s remarks also quoted him saying Huawei could not produce enough AI computing equipment to meet domestic demand.
As an Amazon Associate I earn from qualifying purchases.
A separate estimate reported by the Associated Press gives a different, more specific picture. Bernstein estimated that Nvidia and Huawei each had about 40% of China’s AI-chip market in 2025, then forecast Nvidia at around 8% and Huawei at about 50% in 2026. The 2026 figures are projections, not final results; the AP report does not make them an audited market tally.
| Period | Nvidia | Huawei | What the figures represent |
|---|---|---|---|
| 2025 | About 40% | About 40% | Bernstein estimates reported by the Associated Press in 2026 |
| 2026 | Around 8% | About 50% | Bernstein forecast reported by the Associated Press in 2026, not a final market result |
The estimate and forecast concern market share, not a head-to-head performance score. They also should not be treated as confirmation of Xu’s claim: they come from a separate analyst estimate, and the underlying market is difficult to measure. The AP report notes that Nvidia still plays a role in Chinese AI development.
#1 Best Overall
- Axial-tech fans now feature a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- 2.5-slot design allows for greater build compatibility while maintaining cooling performance
- 0dB technology lets you enjoy light gaming in relative silence
- Dual BIOS switch lets you toggle between Quiet and Performance BIOS profiles
- Dual ball fan bearings last up to twice as long as sleeve bearing designs
What is Huawei trying to replace?
Huawei’s effort is broader than designing a processor. It is building an alternative stack: Ascend accelerators, software for using them, and connected computing systems that combine many chips. For AI workloads, the useful unit is often the complete system—not one chip in isolation—because processors must work together and software must schedule their work efficiently.
The goal is greater control over access to computing hardware and its supporting technology. Xu described the issue as one of technological independence: “We cannot accept a destiny where we cannot control our fate being determined by others in terms of willingness to sell chips to China or not.” That ambition does not mean Huawei has already removed every foreign dependency or can satisfy all domestic demand.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Why a market-share lead would not prove Ascend is better at every AI task
Market share and technical capability answer different questions. A chip can gain sales because of availability, policy, price or a buyer’s need for a dependable supply, even if another product performs better on a particular workload. Conversely, a technically strong accelerator may be hard to deploy at scale if it is scarce or difficult to support.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesTraining and inference are not interchangeable benchmarks
Training develops a model; inference uses a trained model to generate answers or predictions. Performance can differ by task, model, software and system configuration, so a result for one workload cannot establish a general winner. Gregory C. Allen’s April 2025 testimony to the U.S. House summarized a reported assessment that Ascend 910C delivered roughly 60% of Nvidia H100 performance for inference. That is a historical, reported comparison—not an independently verified benchmark of Huawei’s newer chips, and not a comparison of training performance. Allen’s testimony provides that earlier supply-chain and chip context.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Connected systems and software matter
Nvidia’s CUDA software ecosystem is an established advantage: developers have built tools and workflows around it. Huawei is trying to make Ascend easier to program and deploy, including work on software libraries and chip-to-chip communication. On October 1, 2026, Tom’s Hardware reported that DeepSeek and Huawei released programming tools for Ascend, including compute and communication libraries and TileLang support. These are steps toward a broader ecosystem, not evidence that the gap with CUDA has closed. The report on the DeepSeek-Huawei tools describes the collaboration.
How far along is Huawei’s Ascend ecosystem?
Huawei is presenting Ascend as a growing software and developer platform, but its adoption figures are company-reported rather than independently audited. In a September 2026 account, Huawei said Ascend supported more than 90 leading third-party open-source projects, that more than 5,200 developers were active on Ascend software each month, and that over 40 AI models had been trained directly on its computing platform. Reuters separately attributed the developer and model counts to Huawei. Huawei’s account of its Ascend software and ecosystem says CANN, its software platform, is moving toward sustained, community-driven open-source development.
Rank #4
- Powered by Radeon RX 9070 XT
- WINDFORCE Cooling System
- Hawk Fan
- Server-grade Thermal Conductive Gel
- RGB Lighting
Those numbers suggest activity and investment, but they do not by themselves show how many deployments are in production, how much computing work they handle, or how easily a team can move an existing Nvidia-based workflow. Huawei’s David Wang said the company was evolving Ascend on a one-generation-a-year cycle; that is a company roadmap statement, not a guarantee that future products will ship on schedule.
What do the 950-series plans and prices tell buyers?
Reuters reported in September 2026 that Huawei’s Ascend 950DT was expected in the fourth quarter of that year. Huawei said testing had produced good results and expected model developers to begin training on systems using it the following year. Those are plans and company statements, not confirmation of broad deployment or an independent performance result.
Best Value
- Axial-tech fans now feature a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- Phase-change GPU thermal pad helps ensure optimal heat transfer, lowering GPU temperatures for enhanced performance and reliability
- 2.5-slot design allows for greater build compatibility while maintaining cooling performance
- Dual-ball fan bearings last up to twice as long as standard conventional sleeve bearings designs
- 0dB technology lets you enjoy light gaming in relative silence
Reuters described the 950DT card as integrating an AI processor with memory and other components, primarily for developing models and generating responses. It described the 950PR as intended to handle user requests before they reach a model. Two people familiar with pricing discussions told Reuters the 950DT was indicated at more than 250,000 yuan (US$37,255); this was not a published Huawei price list or an official retail quote. The same reporting said high-bandwidth memory (HBM) scarcity and cost were affecting Chinese AI-chip makers, a constraint that can limit how quickly hardware supply grows. Reuters’ report on HBM constraints and reported pricing details those indications.
Xu said Huawei lacked capacity to satisfy demand in China and was limiting overseas expansion. That makes supply access part of the competition: an organization may value a locally available system even when it would otherwise compare alternatives on performance, software or price. It also means a market-share shift should not automatically be read as proof that Chinese buyers have stopped using Nvidia.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What should an AI team compare before choosing?
A decision should be made against the team’s actual workload and deployment conditions, not a single market-share estimate or a headline performance number. Compare:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Workload: whether the priority is training, inference, or both, and whether the specific model and software are supported.
- Whole-system throughput: how the accelerators, memory and interconnect perform together at the scale the deployment needs.
- Software migration: the engineering effort to port, optimize and maintain existing code, libraries and tools.
- Supply and timing: whether the needed systems can be obtained in the relevant market and in sufficient volume, given reported capacity and HBM constraints.
- Total cost: the actual system price and operating costs available to the buyer; the reported 950DT indication is not a general retail price.
- Evidence quality: whether a performance comparison is independently tested, current, and matched for workload and configuration.
The reporting available in September 2026 does not establish an independent, apples-to-apples benchmark of Ascend 950DT against a current Nvidia accelerator. The older 910C inference comparison cannot fill that gap. For a buyer, the most defensible conclusion is that Ascend is becoming a more significant domestic option, while the right choice still depends on workload, software readiness, system availability and verified performance.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
