The largest A-share IPO of the year—ChangXin Technology’s listing on the STAR Market for RMB 58 billion—has made Alibaba the biggest winner: with an investment of RMB 7.6 billion, Alibaba increased its stake to 5% just before the IPO, expecting an unrealized gain of 17 times and a book value exceeding RMB 130 billion. This is not merely a financial investment, but also a microcosm of Alibaba’s strategic shift from “selling new retail to buying AI infrastructure”—having invested RMB 36 billion over three years, its current unrealized gains have surpassed RMB 210 billion.

Author and source: Tencent Technology

The largest IPO on China's A-share market this year has a particularly unique winner.

On July 16, CXMT (688825.SH) launched its new share subscription on the STAR Market, with an issue price of RMB 8.66 per share, aiming to raise approximately RMB 58 billion—the largest IPO in STAR Market history.

The industry investor with the highest stake is Alibaba, which, through two entities, holds nearly 5% of ChangXin Memory Technologies, with a total investment of approximately RMB 7.6 billion. This amount exceeds the direct shareholding of ChangXin’s founder and chairman, Zhu Yiming, who collectively holds approximately 2.65% to 2.68% of ChangXin Memory Technologies through indirect holdings.

Alibaba’s investment timing is also noteworthy: it first invested in CXMT in December 2021, and after multiple rounds of dilution, its stake was reduced to approximately 1%—a classic example of a small, strategic “positioning” investment. For the next three and a half years, Alibaba remained inactive. Then, just before CXMT’s IPO in June 2025, Alibaba suddenly made a significant additional investment, raising its stake to around 5%, becoming the largest industrial investor.

Based on post-listing valuation estimates, Alibaba's stake in CXMT is valued at approximately RMB 130 billion, with an expected total return multiple of up to 17x.

Why would a company that started in e-commerce bet its largest pre-IPO industrial stake on a storage manufacturer that had suffered massive losses for years?

Just a few years ago, CXMT was deeply embroiled in industry price wars. Today, giants like Apple are lining up to secure its production capacity. Several industry insiders shared a common view with Tencent Technology: Alibaba’s heavy investment in CXMT complements the self-developed chips of Pingtouge, forming a synergistic hardware foundation for computing and storage—preparing the groundwork for Alibaba Cloud’s future dominance in AI applications by removing hardware supply bottlenecks.

Behind this investment, Alibaba Investments has undergone a near-complete transformation over the past three years—focusing on AI, actively investing in chips, large models, embodied intelligence, and AI applications; backing early-stage, small, and cutting-edge companies; and avoiding tight control over portfolio companies in favor of mutual benefit. According to incomplete statistics, Alibaba has invested approximately RMB 36 billion in AI-related ventures over the past three years, with current unrealized gains exceeding RMB 210 billion.

Ten years ago, Alibaba’s investment keywords were still: strength, control, integration.

01 Sell the past, buy the future

Let’s first look at what Alibaba sold.

On December 17, 2024, Alibaba announced that it would sell its entire stake in Intime Retail to a consortium composed of Yaguo Group and Intime’s management team for approximately RMB 7.4 billion, expecting a loss of about RMB 9.3 billion. Half a month later, on January 1, 2025, Alibaba announced again that it would sell its entire stake in Sun Art Retail to CDH Capital for a maximum of approximately HKD 13.138 billion—Alibaba had initially invested in 2017 and increased its stake in 2020, having poured a total of approximately HKD 50.4 billion into the parent company of RT-Mart; this sale is expected to result in a shareholder-allocated loss of approximately RMB 13.177 billion.

In half a month, two transactions resulted in approximately RMB 22.4 billion in notional losses. Prior to this, Alibaba had transferred its stakes in seven listed companies—including YTO Express,红星美凯龙, and Focus Media—to Hangzhou Haoyue, and sequentially reduced its holdings in XPeng and Bilibili.

Yintai and Sun Art were the two most important pieces in Alibaba’s “new retail” era. At the Cloud Computing Conference in October 2016, Jack Ma declared, “In the next ten to twenty years, there will be no such thing as e-commerce—only new retail.” Following this, Alibaba privatized Yintai, took control of Sun Art, and acquired Ele.me—each transaction involving significant capital investment and accompanied by deep organizational restructuring. This was the standard pattern of Alibaba’s investments during that era: first take a stake, then increase it, then gain control, and finally achieve full ownership—UC, AutoNavi, and Youku Tudou all followed this path; Ele.me went from an initial investment of $1.25 billion to full acquisition for $9.5 billion in just two years.

At the time, Alibaba’s investments typically required three things: traffic, users, and technology. As business cycles and profitability have evolved, this logic has now been completely replaced.

The turning point occurred in September 2023, when Wu Yongming became CEO of Alibaba Group and immediately sent a company-wide letter establishing "User First, AI-Driven" as the two core strategic priorities. At the analyst call in November of that year, he stated, "The driving force behind future industry development will be technological innovation represented by AI."

This CEO, with an engineering background, differs greatly in character from his predecessors. Alibaba’s investment strategy during the internet era was a classic investment banking approach—skilled in valuation and adept at structuring terms.

Wu Yongming is a technical partner among Alibaba’s “Eighteen Lohan,” having written the earliest code for Taobao and led Alibaba Mama and Yitao. He is also skilled in investing. Before returning to Alibaba’s management team, he personally founded Yuanjing Capital, which has systematically invested in a number of AI companies. For instance, during Alibaba’s lead investment in Star Era’s Pre-A round, Yuanjing Capital co-led the round. Many Alibaba employees believe Wu Yongming’s return marks the beginning of Alibaba’s shift in investment strategy.

02 Two Logics: Go and Mining

An investor remarked that if Alibaba’s investments during the mobile internet era were like a Go board—where the top-level design was clear, the board was the foundation, and the goal was to collectively capture the most black stones—then Alibaba’s investments in the AI era are more like a battery factory investing in lithium mines.

“The development of AI will face a chain of challenges—first, a shortage of computing power, then a shortage of storage, and later possibly a shortage of electricity,” analyzed an investor familiar with Alibaba. “To ensure the security of the entire industrial chain, it’s necessary to invest upstream to secure resources, just as battery manufacturers invest in lithium mines.”

There is one difference between the two logics: Go seeks control, while mining seeks binding.

Take Alibaba’s investments in China’s “Six Little Tigers” of large models as an example. Starting in the second half of 2023, Alibaba aggressively invested in leading Chinese large model startups: In September 2023, Alibaba invested in Zhipu AI, with Alibaba Cloud contributing 1.2 billion yuan in its Series B round, plus an additional 440 million yuan from Ant Group, directly propelling Zhipu into unicorn status; in October, it participated in Baichuan Intelligence; in November, Alibaba Cloud led the investment in ZeroOne Technologies; in February 2024, it led the investment in Moonshot AI; and in March, it led a round of at least $600 million in MiniMax.

Six Little Tigers, and Alibaba invested in five of them. At the time, the investment community joked that Alibaba was “collecting stamps.”

Looking specifically at the structure of these investments:

First, there is no attempt to exert control. Even though Alibaba holds approximately 36% of Moonshot AI, it holds preferred shares and does not interfere in corporate governance, nor has it appointed management or restructured the organization as it did with Ele.me in the past.

Second, there is no exclusivity. Behind these five companies stand Tencent, Meituan, Xiaomi, Sequoia, and Hillhouse—Alibaba has not demanded an exclusive "choose one" arrangement.

Third, mining power equity participation.

A source told Tencent Tech that among Alibaba's investments, a portion consists of "computing power discounts," meaning that part of the investment is paid in the form of computing power, requiring the invested companies to use Alibaba Cloud's computing resources.

This model is not uncommon in the industry: overseas, Microsoft invested in OpenAI and Amazon invested in Anthropic, with part of each investment delivered in the form of computing power. Beyond financial returns, both parties gained strategic advantages—the invested companies secured access to the most scarce training resources, while cloud providers locked in their largest computing customers and secured a leading position in the era of large models.

Another individual close to Alibaba’s investment circle told Tencent Technology that, in his view, equity in computing power is “just a different investment structure. In the internet era, major companies investing in apps included a certain ‘traffic discount’; in the AI era, it’s a ‘computing power discount.’”

Of the three points, “no exclusivity” best reveals how much times have changed.

Exclusivity was once the default precondition for Alibaba’s strategic investments. In 2014, Alibaba demanded that Meituan remove WeChat Pay from its payment options and retain only Alipay; Wang Xing refused. Subsequently, between 2015 and 2016, Alibaba sold its entire ~7% stake in Meituan for approximately $900 million—a stake that, if held until Meituan’s peak market value, would have been worth over RMB 150 billion.

Today, taking Moonshot AI as an example, in February 2024, Alibaba became its largest institutional shareholder; six months later, in its Series B round, Tencent invested; subsequently, in the Series C and C+ rounds, Alibaba and Tencent jointly participated in three consecutive funding rounds. Numerous media reports have noted that Tencent’s earlier investment in Dianping once triggered Alibaba’s withdrawal from Meituan, yet today Alibaba faces no resistance as Tencent enters. In May 2026, the same structure was replicated at Zhipu AI—Alibaba, Tencent, Meituan, and Xiaomi all appeared simultaneously on its shareholder list. In the Series C round of Unitree Robotics, Alibaba, Tencent, China Mobile, and Geely entered together. On MiniMax’s prospectus, Alibaba holds an indirect stake of 13.66%, with Tencent and miHoYo listed alongside it.

An investor commented on this shift: “In the internet era, WeChat and Alipay were the era itself, so investment meant bringing you onto my board. But AI is different—in the AI era, you can only follow the trajectory of technological evolution, empowering and binding yourself through capital and computing power; the posture must shift from ‘shaping the era’ to ‘adapting to the era.’”

Another factor is that the market environment for investments is changing. An investor told Tencent Technology, "Now, early-stage investments are consensus-driven; good projects are scarce, and everyone is competing for them—even large companies no longer have the power to force these startups to choose between them."

At the same time, he believes that projects with investment value today are often hard tech initiatives, and since large companies have broad business boundaries, nearly any hard tech project can typically be explained from a synergistic perspective.

This logic also has its limits and costs. Among the five companies Alibaba invested in, several have faced selling pressure after lock-up periods expired; MiniMax suffered consecutive declines when it encountered its first large-scale lock-up expirations on the Hong Kong Stock Exchange. Meanwhile, DeepSeek, which took no funding from Alibaba, single-handedly disrupted the global AI landscape in early 2025. The market had widely rumored that Alibaba had subscribed to a 10% stake in DeepSeek at a $10 billion valuation—a claim officially denied by Alibaba.

Collecting-style investing ensures you don’t miss the top projects, but it doesn’t guarantee you’ll pick the right variables.

03 Invest in what's missing: infrastructure, models, applications

In March 2026, Alibaba established the Alibaba Token Hub (ATH) business group, led personally by Wu Yongming, with its core strategy summarized in three actions: creating tokens, distributing tokens, and applying tokens.

In the AI era, tokens are regarded as foundational resources akin to electricity, and Alibaba’s investment landscape spans the entire value chain of their production, distribution, and consumption, which can be summarized in three layers: compute infrastructure, model ecosystem, and applications & embodiment.

The computing power infrastructure layer is the main battlefield. At the beginning of 2025, Alibaba announced it would invest at least RMB 380 billion over the next three years to build AI and cloud computing infrastructure—the largest investment plan in this sector by any private Chinese company to date. In fiscal year 2026, Alibaba’s capital expenditures reached RMB 126.063 billion.

This figure places it among the top tier domestically. Industry sources indicate that ByteDance’s preliminary 2026 AI capital expenditure plan is approximately RMB 160 billion, roughly on par with Alibaba. However, on a global scale, the gap remains substantial—in 2026, Amazon’s capital expenditure guidance is around $200 billion, Google’s is $175–185 billion, Meta’s is $115–135 billion, and Microsoft is expected to exceed $105 billion for the year.

In addition to its in-house development, Alibaba has also expanded its upstream chip investments. Alibaba’s subsidiary, Alisoft China, participated as a cornerstone investor in the Hong Kong IPO of MemTech, investing $550 million and ranking among the top cornerstone investors. MemTech’s product portfolio includes memory interface chips, PCIe/CXL retimers, and CXL memory expansion controllers. Alibaba has also invested in Ascendion Technologies, Lightmatter, and Holosonic Semiconductor, covering areas such as RF baseband, silicon photonics computing, and AI inference chips.

In the storage domain, Alibaba is advancing both in-house development and industry investments. In 2023, Pingtouge released the Tianyue 510 SSD controller chip, which has been deployed in Alibaba Cloud data centers. Across the supply chain, CXMT provides DRAM, YMTC provides NAND, Montage Technologies provides interconnect chips, and Alibaba Cloud handles system integration and business applications.

ChangXin is developing HBM and may potentially complement Pingtouge’s AI chips in the future. According to IDC data, in 2025, Pingtouge’s AI chip shipments will rank second in China, behind only Huawei. Individuals close to Pingtouge told Tencent Tech that Pingtouge’s IPO process is accelerating.

During the Q2 2025 earnings call, management disclosed that Alibaba has prepared contingency plans in response to global AI chip supply and policy changes, establishing a diversified supply chain through partnerships with multiple collaborators to "ensure the timely advancement of the RMB 380 billion investment plan."

The middle layer is the model ecosystem. In this layer, Alibaba sells water, mines gold itself, and holds equity in nearly every gold miner.

Alibaba's self-developed Qwen, along with five large model companies in which Alibaba holds stakes, are all potential competitors of Qwen and also all customers of Alibaba Cloud's computing power.

This layer is also the most fiercely competitive. According to Omdia, Alibaba Cloud’s share of China’s AI cloud market rose from 35.8% in the first half of 2025 to 38.1%, surpassing the combined share of the second through fourth places; revenue from AI-related products has maintained triple-digit growth for eleven consecutive quarters. However, based on IDC’s measurement of public cloud inference volume for large models, Volcano Engine held a 49.5% market share in 2025—nearly one out of every two tokens on China’s public cloud ran on Volcano Engine—while Alibaba Cloud ranked second with 27%.

Cai Jingxin and Wu Yongming summarized the rationale for investing in AI in their first letter to shareholders: as a cloud computing provider, Alibaba aims to meet the surge in computing power demand brought about by the explosion of large models.

The top layer consists of applications and embodied intelligence.

Starting in 2024, Alibaba made concentrated moves in the embodied intelligence sector: In May, it strategically invested in ZhiJi Power, acquiring an 18.78% stake to become the largest shareholder outside the founding team; in October 2024, it co-led the Pre-A round financing of Xingdong Jiyuan, with the equity transfer officially completed in January 2025, reinforcing its strategic commitment; subsequently, Alibaba Group and Ant-affiliated capital progressively invested in Unitree Robotics, Xinghai Tu, and dexterous hand manufacturer Lingxin Qiaoshou; in September 2025, Alibaba Cloud independently led a nearly RMB 1 billion A+ round investment in Zibian Robotics—marking the first direct investment by Alibaba Cloud’s business entity in an embodied intelligence company.

Wu Yongming once said at the 2024 Cloud栖 Conference: "The greatest imagination of generative AI is not about creating one or two new super apps on a smartphone screen, but about taking over the digital world and transforming the physical world."

Large models are the brain, robots are the body, and Alibaba’s five invested large model companies, along with its self-developed Qwen, all need to find physical embodiments for intelligence.

Stacked in three layers, Alibaba plays three roles simultaneously along the AI value chain: it is a shareholder of companies like Changxin and Lanji, a computing power provider to firms like Moonshot, and the operator of Qwen and Kuaike—acting as shareholder, supplier, and competitor all at once. This is an unprecedented level of complexity in the internet era, and it embodies the concrete form of the “align with the times” strategy: no longer striving to win at every stage, but ensuring that Alibaba is present whenever any stage explodes.

And if we follow the chain of "invest in what's missing," we can see more clearly:

The first bottleneck is computing power. Structural shortages have persisted for nearly three years since 2023.

Stage two: Storage. This is the current bottleneck experiencing explosive growth. AI servers require 8 to 10 times more DRAM and over three times the NAND capacity of traditional servers. Goldman Sachs, in its research report, assessed that the supply-demand imbalance in storage is “the most pronounced among all segments,” with the tight market conditions expected to persist at least until the second half of 2027.

The third ring: electricity. This is the bottleneck just beginning to emerge. The International Energy Agency warns that global data center electricity consumption will surpass 1,000 terawatt-hours for the first time in 2026, with AI-related power usage accounting for over 60%. Electricity costs now make up around 60% of data center operating expenses, with some AI computing centers exceeding 70%.

In public information, Alibaba's investment portfolio currently lacks a presence in the power sector.

A person close to Alibaba told Tencent Technology that Alibaba is also engaging with projects in communications, energy, and other sectors, and new updates may be released in the future.