In the final weeks of 2022, just as China was unwinding three years of pandemic controls, Jack Ma was at a low point.
Alibaba, his colossal internet empire, was under assault from regulators and from rivals stealing market share in ecommerce and cloud computing. Its share price was down 80 per cent from its peak. The billionaire himself had relocated to Tokyo, retreating from public view after falling out of favour with Beijing.
The release that November of OpenAI’s powerful bot ChatGPT only seemed to underscore to Ma how far Alibaba had fallen, according to a source close to him. He recognised the significance of a technology with the potential to transform the global economy. But he also knew that Alibaba was far behind.
In that moment, the seeds of a transformation were sown. In the two years since, Ma has quietly presided over a strategic turnaround that investors are betting puts Alibaba in a leading position in China’s push to roll out AI nationwide.
Alibaba is now seen to be in peak position to capitalise on an expected boom in AI demand in China, which has been galvanised by the buzz around DeepSeek, a homegrown start-up whose advanced AI model was built with fewer resources. Over the past three years, the tech giant has ramped up investment in AI, backing start-ups, spending lavishly on chips and hiring researchers by the dozen. Qwen, its family of large language models, is now considered a market leader in China, and Apple has chosen it to run AI functions on iPhones in the country.
Ma himself is also back from exile. In a highly symbolic move, he was selected to sit in the front row among other business leaders at a meeting with Chinese President Xi Jinping last month. Investors reacted with enthusiasm, with Alibaba’s share price soaring by 66 per cent since the start of the year.
Alibaba’s reversal in fortunes marks a remarkable comeback for a company that had embodied the rise of China’s tech industry in the first two decades of this century. It also underscores the country’s emergence as a global player in AI.
The FT interviewed more than two dozen current and former Alibaba employees, business partners, competitors and analysts, who described how the company executed its change in strategy — but also raised questions about whether it can establish a long-term lead in a ruthlessly competitive AI race. Alibaba declined to comment.
The company’s sprawling campus in Hangzhou, two hours’ drive from Shanghai, has had an influx of guests looking to peek inside the company’s bid for redemption. During a visit in February, visitors were given tours of a showroom with a range of AI product offerings, including Alexa-like smart speakers and TVs.
“The difference between the Alibaba campus now and six months ago is quite palpable,” says Brian Wong, an entrepreneur and former Alibaba executive. “These days, there is a sense of vibrancy and energy. People have a much clearer direction now.”
Alibaba’s fall from grace was precipitous. At the beginning of November 2020, it was one of the world’s most valuable companies and ready to achieve another capital-market milestone.
Ant Group, Alibaba’s fintech arm and a national champion, was set to raise $34.5bn from the world’s largest IPO, which was many times oversubscribed, valuing the company at over $310bn. Alibaba’s own stock closed at a valuation of almost $700bn the week before in New York.
But the music suddenly stopped. Chinese regulators shocked the markets by abruptly calling off Ant’s IPO, citing belated findings of shortcomings in its businesses. Ma was summoned to Beijing by Chinese regulators about a speech he had given the month before in Shanghai in which he called the nation’s banking system outdated and operating like pawn shops. Soon after, he would vanish from public view.
Alibaba’s shares tanked. Investors panicked that being on the wrong side of Beijing could be detrimental for business and they were not wrong. Ant was subject to three years of rolling investigations until regulators eventually slapped it with several billion dollars worth of fines and moved to reduce its monopoly power.
Things at the corporation’s core business were not going well either. The strategic push by Daniel Zhang, appointed chief executive in 2015, was dubbed as “new retail”, which involved investing in brick-and-mortar business. The plan echoed similar moves by Amazon, which bought WholeFoods for $13.7bn in 2017.
But Alibaba’s ambition was much grander. It made dozens of multibillion-dollar investments, from furniture stores and healthcare centres to malls and supermarkets. It also built a premium grocery chain, Freshippo, from scratch.
Revenue from its offline businesses expanded, before the pandemic struck and kept shoppers at home to buy mainly online. But Alibaba’s flagship ecommerce division, its primary revenue source, also started losing market share to its competitors as China’s economy buckled.
Middle-class shoppers previously loyal to its Taobao and Tmall began flocking to PDD Holdings’ cheap platform as they reined in spending. Meanwhile, ByteDance lured young shoppers away with its addictive short videos and livestreaming. Alibaba’s profit declined to $10bn in 2023, less than half its peak two years earlier.
Ma, at that time, was still avoiding the limelight. While he resigned from all official roles at Alibaba in 2019, the billionaire kept close contact with the company’s executives and remained a key decision maker about its strategies, according to a person close to him.
He tried to made sure speculation on his whereabouts did not cause too much damage to the company. When a US journalist contacted Ma in March 2023 to comment on whether he was unable to return to China, he took quick action to kill the story, according to a person with knowledge of the events. He jumped on his private jet to fly back from Tokyo and made a rare public appearance at a Hangzhou school he founded. If the story had been published it might have further eroded weak investor confidence in Alibaba.
But he knew he had to make a change. He decided to elevate his co-founder and closest ally Joe Tsai as chair to steady the ship.
Tsai, splitting his time between New York and San Diego, was initially unsure that he had the appetite for it after nearly a decade without an operational role at Alibaba. But Ma persuaded him to step in, according to people close to the two men. Tsai was tasked with divesting various non-strategic assets and improving the company’s standing with investors.
Alibaba’s first attempt at a strategic shift was a restructuring plan announced in March 2023 that was to split the group into six independently managed units, allowing them to go public on their own to achieve higher valuations. These include Cainiao, a top logistics company, and Alibaba Cloud, China’s leading cloud service provider. Investors initially reacted enthusiastically before sentiment shifted on the prospects for the Chinese economy.
Six months later, Alibaba was forced to reverse course, as investor sentiment on Chinese equities was too weak to support the plans.
Ma and Tsai decided it was time for a different strategy. They needed to make a more drastic shift: reinventing Alibaba as an AI company.
Pivoting to AI was a big bet, and new hands would be needed to ensure its success. But the process would not be smooth.
Eddie Wu, known internally as the “tech guy” among Alibaba’s initial team of 18 who previously held leadership roles at several Alibaba units, was chosen to succeed Daniel Zhang as chief executive.
A loyal lieutenant devoted to Alibaba, Zhang was initially told he would take charge of the cloud business after handing over the reins to Wu.
Alibaba started making some small investments in LLM research in 2019, but it was only after ChatGPT’s release that the company built out a sizeable internal research team and began ploughing money into LLM start-ups that were seeking to become the country’s answer to OpenAI. The investments were initiated by Zhang, who wanted to continue this work as cloud chief.
But Wu, the incoming chief executive, pushed back, arguing to Ma and Tsai that he should be in control as the cloud business held the keys to unlocking Alibaba’s future as an AI company. They agreed, and on the day Wu took over as chief executive and chair in September 2023, Alibaba announced that he would also lead the cloud business.
“The management and strategy were changing all the time. A lot of people left. I was confused for a few years too; it was not clear about where we were going,” says one veteran employee.
But the same employee says things began to turn around under Wu’s leadership. Wu set about centralising decision-making that had been splintered across the six business units in the now ditched break-up.
With control over Alibaba’s purse strings and the backing of Ma and Tsai, he began selling most of Alibaba’s struggling retail businesses to reallocate resources to the AI push, and pursued new investments.
In the 15 months since he took the top jobs, Wu has spent Rmb81bn ($11bn) on capital expenditure, up from Rmb34bn in the previous 15 months. The company has poured money into start-ups, including Moonshot, MiniMax and Zhipu, with much of the investment coming in computing vouchers for its cloud platform.
Alibaba also beefed up its model training team at Qwen to around 100 people, one of the largest in China, and started pushing out more and more competitive models. Its decision to make them open source “meant it could capture developer attention, enabling it to iterate its models quickly”, says Kevin Xu, founder of Interconnected Capital, which has invested in Alibaba.
The company got a strong vote of confidence from Apple, which chose Alibaba as its partner to introduce AI functions on iPhones in China later this year, chair Tsai revealed last month. While the iPhone maker used OpenAI’s LLM in the US and other western markets, for the Chinese market it must opt for one of the approved models developed by a local player.
Apple considered other similarly advanced Chinese models, but picked Qwen because the partnership requires significant support from dedicated engineers and platform resources, which are not available at smaller start-ups such as DeepSeek, according to one person with knowledge of the matter.
“We experimented running the small Qwen models directly on iPhone. They perform very well,” says Steve Hsu, a professor of computational mathematics at Michigan State University and founder of Superfocus.ai, which builds LLM applications. “I can see a scenario where Apple decides to roll it out globally beyond China.”
The company’s leading position in cloud computing and infrastructure makes its models attractive, especially for large corporate clients looking for reliable long-term solutions.
“The LLM race is a marathon, and it’s not only about the performance of LLM itself,” says Charlie Dai, vice-president and principal analyst at Forrester. “Qwen has strong performance, while [cloud computing subsidiary] AliCloud has comprehensive coverage for AI platform capabilities with a border partner ecosystem.”
Yet while Alibaba has been releasing increasingly competitive AI models, they did not gain much attention outside of the Chinese developer community. It took until the start of this year for the outside world to recognise the work China’s AI players had been doing.
DeepSeek, also based in Hangzhou, sent shockwaves across the globe with its top-tier reasoning model despite the company having far fewer chips than larger US rivals. It shattered the common belief that China had a significant lag in LLMs, and investors started to look again at Chinese tech names trading at substantial discounts to their US counterparts.
Wu further boosted investor confidence by announcing last month that Alibaba was ready to spend Rmb380bn building its AI infrastructure in the next three years, more than the total from the past decade.
The company has budgeted Rmb39bn this year to purchase AI chips, more than 50 per cent more than last year’s amount, according to a person with knowledge of the plans, adding that the amount could be revised higher during the year if growth beats expectations.
Alibaba is also seeking to win over China’s 1.1bn netizens with its various AI applications. Earlier this year, it hired former Salesforce executive Steven Hoi to lead its consumer-based AI strategy.
Qwen and Quark, Alibaba’s answer to ChatGPT and Gemini, are tasked with attracting as many users as possible this year. They are competing head-to-head with ByteDance’s Doubao and Tencent’s Yuanbao, as well as start-ups led by DeepSeek. This month, Qwen released its latest model, QwQ-32B, which is even more cost-effective than DeepSeek’s R1 and has similarly rated performance.
Alibaba is betting big on Quark, a browser app that it is now pitching as an AI assistant with reasoning capabilities. Quark received an upgrade last week and is now powered by its Qwen reasoning model.
Chief executive Wu has spoken about the need to “AI-size” its existing businesses across the board. All units have been told their 2025 performance will be evaluated by how they can leverage AI to spur growth, according to people with knowledge of the discussions.
Core ecommerce units, including Taobao and Tmall, are being encouraged to adopt more AI technologies. The teams are working closely with Qwen’s engineers to co-develop functions that improve efficiency and user experience.
It is also developing a fleet of AI-native applications, some of which could launch this year, according to a person with direct knowledge. “Internally we believe the next killer app, one that’s even more popular than Douyin, could emerge soon based on maturing AI technologies,” says the person. Douyin, ByteDance’s Chinese version for TikTok, has hundreds of millions of daily active users.
The ultimate goal, Wu declared in last month’s earnings call, is to achieve artificial general intelligence, or AGI, the time when machines are predicted to be capable of humanlike critical thinking. While some experts are sceptical about AGI, Wu’s bold declaration is boosting Alibaba’s internal morale.
“When Eddie talked about ambitions like AGI and how that’s going to power all our businesses, I almost cried,” says another longtime employee. “It’s feeling like the old days again. We have something to believe in and strive for together.”
Alibaba’s leadership position is not assured, however, in a fast-changing competitive landscape.
Its arch rival Tencent has responded quickly to incorporate DeepSeek’s models into various elements of its business. Meanwhile, ByteDance is set to spend $12bn on AI chips this year and has one of the leading consumer-facing applications in the country.
“It’s hard to know whether Alibaba can keep the lead. The competition will get more fierce,” says Xu.
Mounting geopolitical tensions between China and the US could also threaten Alibaba’s AI ambitions. Washington has restricted Nvidia from exporting its most advanced AI chips used to train leading models to China and may further limit its sales of H20, a watered-down version of the chipmaker’s H100 chips.
Alibaba’s chip design capabilities are still years behind the most advanced in the world, while it is also struggling to find production capacity due to US bans.
Analysts say Alibaba also risks losing out on the AGI race to leaner, more dedicated teams like DeepSeek, which have a singular focus on research.
While Beijing has warmed up to the nation’s tech giants, counting on them to lead its slowing and ageing economy out of its slump, it remains an open question whether progress by China’s AI players will translate into economic growth.
“Our problems have not vanished overnight, but we have hope and confidence again,” says its veteran employee. “And that’s what got Alibaba through its earlier crises.”
Comments