Huawei is a Chinese technology conglomerate best known for telecom equipment and smartphones — and it is now racing to build a self-sufficient AI stack, from chips to cloud, largely because US export controls cut off its access to Nvidia’s hardware.
What Huawei is
Huawei was founded in 1987 in Shenzhen by Ren Zhengfei, a former army engineer, and grew from reselling imported phone equipment into the world’s largest supplier of telecom network gear, passing Ericsson in 2012. The company now employs roughly 213,000 people across more than 170 countries, organized into five divisions: Carrier (telecom networks), Enterprise (business IT and data centers), Consumer (phones, tablets, wearables), Huawei Cloud, and Digital Power (solar and energy storage). Its smartphones briefly outsold Samsung’s in mid-2020, before sanctions cut off the advanced chips it needed to keep building them at that scale. Huawei has also built its own mobile operating system, HarmonyOS, as an Android alternative, and now pours a large share of its earnings back into research: about 192.3 billion yuan (roughly $27 billion) in 2025, or 21.8% of total revenue, according to the company’s own figures — funding that now goes heavily toward chips and AI.
Why Huawei is building its own AI chips
Starting in 2019, the US added Huawei to its trade blacklist over national-security concerns, and by 2020 barred any chipmaker relying on US technology — including Huawei’s longtime supplier, Taiwan’s TSMC — from selling to the company. That cut Huawei off from the world’s most advanced chips, including the Nvidia GPUs that power most AI training and inference elsewhere; restrictions like these are part of a broader set of AI-related export controls Washington has used to slow China’s access to cutting-edge chips.
Huawei’s response was to build its own. Its chip-design arm, HiSilicon, developed the Ascend line of AI chips, fabricated domestically by China’s SMIC instead of TSMC, and paired them with CANN — a software layer that lets AI frameworks such as PyTorch run on Ascend hardware, much as CUDA lets them run on Nvidia GPUs. A newer version, CANN Next, adds programming features modeled closely on CUDA, aimed at making it easier for developers to switch away from Nvidia. In 2026 Huawei unveiled the Atlas 950 SuperPod, a rack-scale cluster of Ascend chips pitched as a domestic alternative to Nvidia and AMD’s AI data-center systems, and says it plans to ship roughly 750,000 of its newest Ascend 950PR chips this year, with large Chinese buyers such as ByteDance and Alibaba placing sizable orders.
Huawei Cloud and the push beyond China
Huawei Cloud, the company’s cloud-computing division launched in 2017, runs Ascend-based AI infrastructure across dozens of regions and offers ModelArts, its platform for building and deploying machine-learning models. It’s increasingly where Huawei pitches its AI story to businesses outside China too, adding agentic AI infrastructure — systems that can plan and carry out multi-step tasks rather than just answer questions — and a coding assistant called CodeArts to its regional rollouts. Businesses anywhere can sign up for Huawei Cloud directly through its international portal.
Why it matters
Huawei’s build-it-yourself approach is one of the clearest tests of whether a company cut off from the world’s leading chipmaking tools can still compete in AI. If Ascend and CANN keep closing the gap with Nvidia, export controls designed to slow China’s AI progress will have pushed Huawei toward a homegrown alternative it can now sell well beyond China’s borders, through Huawei Cloud’s expanding footprint abroad. It’s also a signal to other governments weighing their own AI dependence on a single chip supplier: sanctions can cut off access, but they don’t necessarily stop a well-funded competitor from building around them.
In the news
Huawei Cloud’s expansion into Southeast Asia includes a new agentic AI infrastructure stack and a CodeArts coding-agent beta in Thailand, part of its broader enterprise AI push in the region.