Huawei Targets $12 Billion in AI Chip Revenue as Ascend 950 Orders Surge

# Huawei Targets $12 Billion in AI Chip Revenue as Ascend 950 Orders Surge

Huawei is betting that 2026 will be its breakout year in AI silicon. The Chinese technology giant is projecting $12 billion in AI chip revenue this year, a 60 percent leap from the roughly $7.5 billion it booked in 2025, as demand for its Ascend 950PR processor surges among China's largest cloud and internet companies.

The aggressive target, first reported by multiple industry outlets this week, underscores how rapidly China's domestic AI chip ecosystem is maturing under the pressure of escalating U.S. export controls. With Nvidia effectively locked out of the high-end Chinese AI accelerator market, Huawei has stepped into the vacuum with a processor that industry analysts say is finally competitive enough to anchor large-scale AI deployments.

The Ascend 950PR: China's Answer to Export Controls

Mass production of the Ascend 950PR began in March 2026 with the launch of Huawei's Atlas 350 accelerator card. The chip delivers up to 1 PFLOPS at FP8 precision and 2 PFLOPS at FP4, with interconnect bandwidth reaching 2 TB/s. It integrates Huawei's in-house HiBL memory offering 112 GB capacity at 1.4 TB/s, a deliberate move to reduce reliance on external memory supply chains.

Manufactured by SMIC using its N+3 process -- broadly comparable to a 5nm-class node -- the 950PR represents a meaningful step forward for China's semiconductor self-sufficiency ambitions. Huawei has publicly claimed the Atlas 350 card delivers 2.8 times the performance of Nvidia's H20, the most advanced chip the American company was still permitted to sell in China before the latest restrictions took hold.

Huawei plans to produce approximately 750,000 Ascend 950PR units this year, alongside 600,000 of the older Ascend 910C chips, which would double 2025 output of that model. Total Ascend production capacity is expected to reach 1.6 million units for the year.

DeepSeek V4 Lights the Fuse

The catalyst for the current order surge is the launch of DeepSeek V4, the latest large language model from one of China's most prominent AI labs. DeepSeek has optimized V4 to run natively on Huawei's Ascend hardware, a development that effectively validated the chip for enterprise-grade AI workloads and triggered a wave of procurement activity.

"If DeepSeek succeeds in running both inference and training on Ascend chips within the next one to two years -- and stabilizes the full software stack including compilers, operators, communication libraries, distributed training, and inference frameworks -- then its core model development pipeline could effectively become independent of CUDA," TrendForce noted in an April analysis of the partnership.

Within days of the DeepSeek V4 announcement, China's three largest cloud buyers moved to secure supply. ByteDance has committed approximately $5.6 billion in Ascend 950PR orders for 2026, the largest single AI chip procurement deal between a Chinese tech company and a domestic chipmaker. Alibaba and Tencent have also placed significant orders, though specific figures for those companies have not been publicly confirmed.

A Two-Chip Strategy

Huawei is not putting all its eggs in one basket. The 950PR is primarily targeted at inference workloads -- the process of running trained AI models to generate predictions and outputs. An upgraded variant, the Ascend 950DT, is expected by the end of this year and is designed specifically for training and deep learning scenarios, where computational demands are even more intense.

Beyond that, Huawei has next-generation Ascend 960 and 970 chips in its pipeline, each targeting roughly two-times performance gains over its predecessor, according to TrendForce.

"The number tells you how fast China's domestic AI stack is forming," Startup Fortune observed of the $12 billion revenue target. "This is not a government subsidy story anymore. It is a market-pull story driven by real enterprise demand."

The Nvidia Shadow

The backdrop to Huawei's surge is Nvidia's forced retreat from the Chinese market. Successive rounds of U.S. export controls have progressively restricted which chips Nvidia can sell to Chinese customers, culminating in rules that have effectively barred the company's most capable AI accelerators from the market entirely.

Market estimates now suggest Huawei could capture 60 percent of China's domestic AI chip market by the end of 2026. That would represent a dramatic reshaping of a market that Nvidia once dominated.

Still, challenges remain. Despite strong demand, Huawei's production capacity cannot yet match the full scale of interest from Chinese cloud providers. Supply constraints, partly driven by U.S. restrictions on advanced chipmaking equipment reaching China, mean that securing Ascend 950PR allocation has become a competitive process among buyers.

What It Means

The $12 billion target is more than a revenue milestone. It signals that China's AI infrastructure is reaching a tipping point where domestic hardware and software can sustain large-scale model development without relying on American technology. If Huawei delivers on its production targets and DeepSeek's CUDA-free software stack matures as planned, the global AI chip market could look fundamentally different by the time 2027 arrives -- bifurcated into two largely independent ecosystems, one built on Nvidia's CUDA and one on Huawei's Ascend.

For Nvidia, the question is no longer whether it will lose the Chinese market, but how quickly the Huawei-led ecosystem can become self-sustaining. For Huawei, the question is whether it can scale manufacturing fast enough to capture the moment. At $12 billion, the answer is already taking shape.