The Huawei sales space on the Cellular World Congress in Barcelona, 2025.
Arjun Kharpal | CNBC
Regardless of being crushed down by years of U.S. commerce restrictions, China’s telecom big Huawei has quietly emerged as one of many nation’s fiercest opponents throughout the complete AI panorama.
Not solely does the Shenzhen-based agency seem to signify Beijing’s reply to American AI chip darling Nvidia, nevertheless it has additionally been an early adopter of monetizing synthetic intelligence fashions in industrial purposes.
“Huawei has been pressured to shift and develop its core enterprise focus over the previous decade… because of quite a lot of exterior pressures on the corporate,” stated Paul Triolo, associate and senior vice chairman for China at advisory agency DGA-Albright Stonebridge Group.
This growth has seen the corporate get entangled in the whole lot from good vehicles and working techniques to the applied sciences wanted for the AI growth, resembling superior semiconductors, information facilities, chips and enormous language fashions.
“No different know-how firm has been capable of be competent in so many alternative sectors with excessive ranges of complexity and limitations to entry,” Triolo stated.
This 12 months, Nvidia CEO Jensen Huang has grow to be more and more vocal in calling Huawei “one of the most formidable technology corporations on the planet.” He has additionally warned that Huawei will change Nvidia in China if Washington continues to limit U.S. chip companies’ exports to the Asian nation.
Nvidia surpassed $4 trillion in market capitalization final week to grow to be the world’s Most worthy firm. Its cutting-edge processors and a associated “CUDA” computing system remain the industry standard for coaching generative AI fashions and purposes.
However that moat could also be narrowing, as Huawei proves that it not solely does all of it, it does it properly. Whereas difficult American AI stalwarts like Nvidia is a tall order, the corporate’s historical past exhibits why it could actually’t be counted out.
Phone switches to nationwide champion
Huawei, which now employs greater than 208,000 folks throughout over 170 markets, got here from humble beginnings. Based by formidable entrepreneur Ren Zhengfei in 1987 out of an condo in Shenzhen, the agency began as a small phone change distributor.
Because it grew right into a telecoms participant, it gained traction by focusing on much less developed markets resembling Africa, the Center East, Russia and South America, earlier than finally increasing to locations like Europe.
By 2019, Huawei could be well-positioned to capitalize on the global 5G rollout, turning into a frontrunner out there. Round this time, it had additionally blossomed into one of many world’s largest smartphone producers and was even designing smartphone chips by its chip design subsidiary, HiSilicon.
However Huawei’s success additionally attracted rising scrutiny from governments exterior China, notably the U.S., which has ceaselessly accused Huawei’s know-how of posing a nationwide safety risk. The Chinese company has refuted such risks.
The export controls have satirically pushed Huawei into the arms of the Chinese language authorities in a manner that CEO Ren Zhengfei all the time resisted.
Paul Triolo
associate and senior vice chairman for China at DGA-Albright Stonebridge Group
Huawei’s enterprise suffered a serious setback in 2019 when it was positioned on a U.S. commerce blacklist, stopping American corporations from doing enterprise with it.
Because the influence of the sanctions kicked in, Huawei’s shopper enterprise – as soon as the corporate’s largest by revenue – halved to about $34 billion in 2021 from the 12 months earlier than.
The corporate nonetheless managed a breakthrough on AI chips, and pressed forward regardless of further U.S. restrictions in 2020 that minimize the corporate off from chipmaker Taiwan Semiconductor Manufacturing Co. A 12 months earlier, Huawei officially launched its Ascend 910 AI processing chip as a part of a technique to construct a “full-stack, all-scenario AI portfolio” and to grow to be a supplier of AI computing energy.
However the U.S. focusing on of Huawei additionally had the impact of turning the corporate right into a martyr-like determine in China, constructing upon consideration it acquired in 2018 when Meng Wanzhou, Huawei’s CFO and daughter of Ren, was arrested in Canada for alleged violations of Iran sanctions.
Because the U.S.-China tech struggle continued to develop and broad superior chip restrictions had been positioned on China, Huawei was an apparent option to grow to be a nationwide champion within the race, with extra impetus and state backing for its AI plans.
“The export controls have satirically pushed Huawei into the arms of the Chinese language authorities in a manner that CEO Ren Zhengfei all the time resisted,” Triolo stated. On this manner, the restrictions additionally turned “the steroids” for Huawei’s AI {hardware} and software program stack.
The comeback
After one other 12 months of declining gross sales within the shopper section, the unit began to show round in 2023 with the discharge of a smartphone that analysts said contained an advanced chip made in China.
The 5G chip got here as a shock to many within the U.S., who did not anticipate Huawei to succeed in that degree of development so shortly with out TSMC. As an alternative, Huawei was reportedly working with Chinese language chipmaker SMIC, an organization that has additionally been blacklisted by the U.S.
Whereas semiconductor analysts stated the dimensions that Huawei and SMIC may produce these chips was severely restricted, Huawei nonetheless had proved it was again within the superior chip sport.
It was additionally round this time that reviews started surfacing about Huawei’s new AI processor chip, the Ascend 910B, with the corporate trying to seize upon gaps left by export controls on Nvidia’s most superior chips. Mass manufacturing of the next-generation 910C is reportedly already on the way in which.
To fill the void left by Nvidia, Huawei “has been making massive strides in replicating the efficiency of high-end GPUs utilizing mixtures of decrease chips,” stated Jeffrey Towson, managing associate at TechMoat Consulting.
In April, Huawei unveiled its “AI CloudMatrix 384”, a system that hyperlinks 384 Ascend 910C chips in a cluster inside information facilities. Analysts have stated CloudMatrix is ready to outperform Nvidia’s system, the GB200 NVL72, on some metrics.
Huawei is not simply catching up, “it is redefining how AI infrastructure works,” Forrester analysts said in a report final month about CloudMatrix.
In the meantime, Huawei has additionally developed its personal “CANN” software program system that acts as an alternative to Nvidia’s CUDA.
“Profitable the AI race is not nearly quicker chips. It additionally consists of delivering the instruments builders have to construct and deploy large-scale fashions,” Forrester’s report stated, although authors famous that Huawei’s merchandise are nonetheless not built-in sufficient with different generally used instruments for builders to change over shortly from Nvidia.
The ‘Ascend Ecosystem Technique’
Whereas Huawei’s purpose to surpass Nvidia is seen as a key improvement in China and the U.S.’s race for AI, it is vital to notice that chips signify only one constructing block of Huawei’s broader AI plans.
Huawei now has its fingers all through the bogus intelligence worth chain, from chips to computing, to AI fashions and AI purposes. These totally different AI enterprise avenues additionally leverage different areas of the corporate’s huge know-how empire.
In truth, the corporate’s “ICT Infrastructure” enterprise — which incorporates 5.5G mobile community deployment and AI techniques for industrial use — turned the corporate’s largest income driver at 362 billion yuan in 2023.
The corporate has been deploying its Ascend AI chips and AI CloudMatrix 384 at its rising portfolio of AI information facilities, that are operated by its cloud computing unit, Huawei Cloud, established in 2017 to compete with the likes of Amazon Net Providers and Oracle.
These information facilities, in flip, have supplied the coaching capabilities and computing energy utilized by Huawei’s suite of AI fashions underneath its Pangu sequence.
Not like different general-purpose AI fashions like OpenAI’s GPT-4 or Google’s Gemini Extremely 1.0, Huawei’s Pangu mannequin is designed to assist extra industry-specific applications across the medical, finance, authorities, industrial and automotive sectors. Pangu has already been utilized in additional than 20 industries during the last 12 months, the corporate said last month.
Rolling out such AI purposes usually entails having Huawei tech workers working for months on the venture web site, even when it is in a distant coal mine, Jack Chen, vice chairman of the advertising and marketing division for Huawei’s oil, gas and mining business unit, which offers digital and clever options to rework these industries, informed CNBC.
That analysis enabled the corporate in Might to deploy more 100 electric-powered trucks that may autonomously transport dirt or coal utilizing the telecom firm’s 5G community, AI and cloud computing providers.
And it isn’t restricted to China. The know-how can “be replicated on a big scale in Central Asia, Latin America, Africa, and the Asia-Pacific,” Chen stated.
Huawei has additionally open-sourced the Pangu fashions, in a transfer it stated would assist it develop abroad and additional its “Ascend ecosystem technique,” which refers to its AI merchandise constructed round its Ascend chips.
Talking to CNBC’s “Squawk Field Asia” on Thursday, Patrick Moorhead of Moor Insights & Technique stated he anticipated Huawei to push Ascend in international locations a part of China’s Belt and Street Initiative — an funding and improvement venture aimed toward rising markets.
Over a interval of 5 to 10 years, the corporate may start to construct critical market share in these international locations, in the identical manner it as soon as did with its telecommunications enterprise, he added.