Investments in MiniMax and Zhipu Are Starting to Pay Off for Alibaba

Its bets on AI unicorns years ago are starting to pay off as the “Token generation” takes shape.

The best-known of these “money printers” are the newly listed MiniMax and Zhipu, along with Moonshot AI, which is preparing to pursue an IPO. On one hand, Alibaba can book net investment gains from its equity stakes. On the other, amid the Token-consumption boom sparked by Lobster, these companies are becoming major consumers of Alibaba’s computing power.

A classic case of getting two bites at the same fish.

On May 13, Alibaba released its quarterly earnings for the period ended March 31, 2026, along with its annual report for fiscal 2026. For the full fiscal year, Alibaba’s interest income and net investment gains rose 322% year on year to 87.512 billion yuan, the highest level for this line item since 2022. Huxiu notes that the 87.512 billion yuan includes fair-value changes and unrealized gains from public listings, much of it non-cash income.

Huxiu has learned that the sharp increase in Alibaba’s interest income and net investment gains was mainly driven by the January 2026 listings of MiniMax and Zhipu, both Alibaba portfolio companies. Fueled by the Lobster-driven Token-consumption boom, both companies saw revenue and profit surge in 2025 and the first quarter:

MiniMax: Before MiniMax’s IPO, Alibaba held about 13.66% of the company, making it the largest outside institutional shareholder. MiniMax’s 2025 revenue rose 158.9% year on year.

Zhipu AI: Alibaba-related entities, including Ant Group, hold about 6%.

Moonshot AI, which has not yet listed, has an even stronger Alibaba imprint. According to disclosures in Alibaba’s fiscal 2024 annual report, Alibaba owns about 36% of Moonshot AI’s preferred shares. That means Alibaba’s related net investment gains could continue to rise if Moonshot AI completes an IPO.

Notably, these AI upstarts are also key customers of Alibaba Cloud.

MiniMax: Alibaba Cloud is one of MiniMax’s major computing power suppliers. In the first three quarters of 2025, MiniMax purchased about $58.4 million worth of computing power.

Zhipu AI: Alibaba Cloud is one of its major suppliers, and Zhipu commercializes its models through MaaS platforms such as Alibaba Cloud Bailian.

Moonshot AI: A flagship model customer for Alibaba Cloud and a case Alibaba Cloud frequently showcases externally.

As these AI upstarts benefited from the Lobster boom at the start of 2026, the massive Token consumption brought a clear windfall to Alibaba Cloud. Earnings showed Alibaba Cloud revenue rose 38% year on year to 41.626 billion yuan, while external commercial revenue growth accelerated to 40%, its fastest pace in nine quarters.

Alibaba should probably send Lobster’s father, Peter Steinberger, and the Alibaba leadership team that decisively invested in these AI companies after the pandemic, banners reading “Best Assist” and “Masterstroke Revival,” respectively.

During the quarter, Alibaba’s total revenue rose 3% year on year to 243.38 billion yuan, while adjusted EBITA fell 84% to 5.102 billion yuan. Against that broader backdrop, Alibaba Cloud, helped by Token consumption and the AI transition, far outpaced the companywide growth rate: quarterly revenue rose 38% year on year to 41.626 billion yuan, and adjusted EBITA rose 57% to 3.796 billion yuan.

Source: Alibaba earnings report

Clearly, “AI and Tokens” have become Alibaba’s new growth engine.

The pressure on Alibaba’s profit still comes from the same three “veteran actors” of the past several quarters:

Cloud and AI infrastructure buildout: Capital expenditure reached 26.9 billion yuan during the quarter, with spending focused on data centers, server computing power expansion, and underlying AI infrastructure.

Flash-buy business and Qianwen consumer user acquisition: Group sales and marketing expenses increased by 17.2 billion yuan year on year this quarter, with the incremental spending mainly going to instant retail operations and Qianwen AI marketing.

AI products and overall large-model R&D investment: Product development expenses, including algorithm iteration and R&D personnel costs, reached 18.957 billion yuan.

These expenses can broadly be divided into two strategic areas: AI and e-commerce. In AI, Alibaba is continuing to invest heavily in infrastructure, refine its Token-centered AI industrial-chain layout, and step up its push on the consumer side. In e-commerce, beyond acquiring users through Flash Buy, Alibaba is also strengthening the AI transformation of its broader e-commerce business.

But challenges and uncertainty remain.

Across the broader AI market, independent AI model and agent companies in China and overseas are generally trying to avoid dependence on a single chip or a single cloud. As investees, MiniMax, Zhipu AI, and Moonshot AI cannot be viewed as a monolithic “Alibaba camp.” If Alibaba wants to extract multiple forms of value from the same fish, it will have to compete with strong rivals such as Volcano Engine.

In the consumer AI market, Qianwen’s user base has declined somewhat after its Lunar New Year marketing push. It is now in a kind of “second place in the arena” position. To break into the top tier, it urgently needs sustained nationwide attention and stronger product capabilities.

On the e-commerce front, Alibaba and Meituan are about to enter the decisive phase of their instant-retail battle. The summer of 2026 is already seen by the core leadership teams of both sides as a critical window. Given the performance pressure on Flash Buy’s decision-makers after several quarters of heavy investment, it is reasonable to expect June through August to become a dividing line: either Alibaba wins more support to fully suppress Meituan, or it maintains the current situation and gradually narrows the front.

The most fundamental challenge may still be the AI revolution of Alibaba’s core business: What exactly should Alibaba’s AI e-commerce look like?

After Alibaba’s earnings call, at 10:30 p.m. Beijing time on May 13, one hour after U.S. markets opened, Alibaba’s U.S.-listed shares were up 6.8% at $143.95 per share.

Token Dividends Arrive, and Alibaba Is Changing

Hangzhou, Alibaba’s Xixi campus, May 10, “Alibaba Day.”

More than a dozen “cats and dogs available for adoption” posters were posted in the lobby of Building C5 in Zone C. This is part of the annual Alibaba Day stray-animal adoption event, a regular fixture each year.

Unlike in previous years, the posters this time leaned fully into the AI mood. The cat and dog posters were packed with AI jokes:

“Equipped with powerful multimodal hearing and smell capabilities” (slogan for a dog named Ermao).

“A Claude Code-level smart dog, with 100% command-comprehension accuracy” (slogan for a dog named Liulian).

“A disguised independent large-model computing power node passing by” (slogan for a cat named Naiqiu).

At Alibaba Day, adoption posters filled with “AI memes” appeared in Alibaba’s Zone C.

Not far from the C5 lobby, in a presentation hall, mascot performers representing several Alibaba AI products, including Wukong, Yun Xiaobao, and Xiaojiuwo, danced close to the audience to TWICE’s “What Is Love?” On the floor, meanwhile, a presenter was enthusiastically explaining how to use Qianwen to order flowers for family members.

This is a snapshot of Alibaba’s current internal push to make “everyone AI-enabled.”

Huxiu has exclusively learned that after Alibaba established the Alibaba Token Hub, or ATH, business group on March 16 and set up a “Group Technology Committee” on April 8, the company has been aggressively pushing AI upgrades across its businesses and reorganizing its business map and employee management model around Token flows:

All employees receive a monthly allocation of free Tokens, which can be used for Qwen, Wukong, Qoder, Qianwen, and other tools depending on the department.

For most technical teams, a dual metric of Token consumption plus output quality, assessed through AI quality checks and human review, has been incorporated into the comprehensive evaluation system.

AI tool usage has become a key assessment focus for business units, which can apply to develop targeted agent tools based on specific scenarios.

Another person familiar with the matter told Huxiu that when ATH was first established, some departments tried to use Token consumption as a key target, but quickly changed the model after real-world practice.

“Across the group, pure Token consumption is not currently treated as a metric. The company cares more about effective Token consumption plus actual AI-based ROI,” the person said.

On the business-map side, ATH, which covers Tongyi Lab, Qianwen, Wukong, MaaS, and the AI Innovation Business Unit, has focused over the past two-plus months on improving the following areas:

Using the ATH framework and the Technology Committee system to plan core technology directions in a unified way, such as which base-model technology modules and which scenarios should take priority.

Reducing “duplicate R&D” across different business segments and improving the reuse of technology and product capabilities.

Creating a more fixed and clearer cross-department communication mechanism.

Huxiu understands that for the overall AI market, Alibaba has broken down demand along the main line of Token consumption and mapped different products to specific demand scenarios. At a macro level, its next core directions include three major areas:

In basic cloud, chips, and related areas: Beyond meeting the needs of its own AI businesses, Alibaba will strengthen penetration in external markets and widen its lead.

In the AI-to-consumer market: Alibaba will treat Token consumption, user scale, and ecosystem coordination as key priorities. It is pursuing not only consumer user growth and activity, but also the ability of products such as Qianwen to connect and coordinate with other business segments inside the Alibaba ecosystem.

For emerging technology directions and scenarios, Alibaba will move quickly with a small, nimble model, including new forms of AI hardware, quantum computing, cloud PCs, and more.

Notably, Alibaba will maintain its firepower in external AI investments. On the May 13 earnings call, Alibaba Chief Financial Officer Toby Xu said: “Over the past year, we have been very firm in making artificial intelligence investments, and we intend to remain firm over the next two years and continue making such investments.”

Huxiu has exclusively learned from several venture investors that after the model boom, Alibaba’s current AI investment thinking mainly follows these lines:

Targets that can organically integrate with Alibaba’s current Token strategy ecosystem, such as emerging agent projects and AI application projects.

A shift in investment perspective from “soft” to “hard,” with Alibaba remaining open to and interested in sectors such as AI hardware, embodied intelligence, and drones.

Experimenting with multiple investment models, including independent startup models for internally incubated projects, and potentially exploring a more flexible internal incubation ecosystem.

A person familiar with the matter told Huxiu that earlier rumors that “Alibaba attempted to invest in DeepSeek” were inaccurate. The real situation was: “In 2025, Alibaba-related people did have contact with the DeepSeek team, but there has been no new contact in 2026. Around that 2025 contact, companies including ByteDance also engaged with DeepSeek. For Alibaba’s current AI ecosystem, DeepSeek is not an indispensable target.”

Another industry source said the outside world may have misreported the matter. “Alibaba’s T-Head chip business did have contact with DeepSeek, and this is related to DeepSeek’s own domestic chip substitution strategy. But this is a business cooperation and product procurement relationship, not an investment.”

From the author’s perspective, Alibaba’s overall AI layout can be divided into internal and external dimensions. From the external investment perspective, it is seeking more “money coming back”: through strategic investments and other means, it is backing more emerging companies that can organically integrate with Alibaba Cloud, Qwen, and T-Head. These companies can not only bring Alibaba investment returns, but also become growth customers through Token consumption.

From the perspective of internal products and businesses, Alibaba needs its own AI tools and products to become the arsenal for its internal AI evolution, something ByteDance and Meituan are also doing. At the same time, Alibaba urgently needs these products to compete in external markets and secure market positions along the Token chain.

Alibaba has both opportunities and challenges across these two dimensions. But in the author’s view, one of the most easily overlooked issues is the relationship between its strategic resolve and public-market sentiment toward its stock price.

Right now, Alibaba’s determination on AI is in part because capital markets’ high expectations for Alibaba are mainly tied to its AI layout and vision. But the risk may also stem from that same point: it also needs a new story.

At first, the outside world was drawn to the narrative of “AI infrastructure as the utilities of a new era.” Then investors wanted to see more, giving rise to narratives around consumer AI and AI’s reshaping of e-commerce. But the latter two are much harder than the first. In essence, the infrastructure narrative is a story of resource advantage plus first-mover advantage. Alibaba built meaningful barriers through an early move and deep resources. The latter two, however, depend on consumer product strength plus operational capability in one case, and efficiency plus determination in the other, specifically the determination to reshape existing businesses and interest structures.

Qianwen and Flash Buy Will Decide the Direction on Two Battlefields

One day after Alibaba Day, on May 11, Alibaba announced that Qianwen and Taobao had been fully connected.

At this point, outsiders can observe Alibaba’s thinking on AI e-commerce with greater clarity.

In the author’s view, the overall layout can be compared to a large tree:

Fertile soil, or the foundational technology layer: the Qwen large model and Taotian AIGX, a foundational technology module built on Qwen and Taotian’s self-developed multimodal products.

The trunk, or the vertical line around traffic: the AI search-and-promotion system, new AI-based product databases, and more.

Branches, or B-side merchant AI tools: AI Dian Xiaomi, Alimama’s AI advertising tool suite, AIGC tools including the Xingchen series, and more.

Fruit, or consumer-facing products: the Qianwen AI shopping assistant, including the Qianwen app and embedded Taobao modules, AI try-on, AI universal search, and more.

A schematic as understood by the author, not an official diagram.

Information obtained by Huxiu shows that in 2025, Taotian invested substantial energy and resources into key areas such as the “soil” and “trunk,” while prioritizing R&D for key scenarios in the “branches” and “fruit.” These foundational moves are the reason Qianwen and Taobao can now be connected, and why a large number of new AI tools for both B-side and C-side users can emerge.

It is not hard to see that this system will create several points of intersection.

At the entry point, Qianwen and Taobao intersect. On the traffic side, the search-and-promotion system intersects with AI tools for both B-side and C-side users. On the tool side, there is overlap between the output of B-side tools and the exploration of C-side tools.

But the real checkpoints also appear at these intersections:

As the consumer-facing entry point and tool for the e-commerce segment, Qianwen is not merely a simple “embedding.” It actually affects the existing business model. To take a simple example, Taobao’s splash-screen ads and the entrance to the Qianwen app require exploration of an entirely new traffic mechanism.

Its overall traffic mechanism may require redesigned “rules of ingenuity” as AI evolution becomes more thorough.

Consumer users want AI to shorten decision-making and reduce advertising. B-side merchants can use AI tools to produce materials more efficiently. Can AI play “Neo” in both worlds at once?

Among these intersections, the most important battlefield may be the integration of Qianwen and Taobao, because this is the first true “same-stage” moment for two of Alibaba’s strategic priorities over the past several years. It is a test of how much chemistry Alibaba’s AI can generate with its core e-commerce business.

Still, Taotian is not facing only the “new world” war of AI. The story in the traditional world is not over: the instant-retail fight represented by Flash Buy is still burning.

The war between Flash Buy and Meituan has become this: success means a business revolution that defeats a giant, Meituan, and delivers growth in users, orders, and GMV; failure means sunk costs running into the tens of billions of yuan.

The key question is how to define success and failure.

The author lists below three key comments Alibaba’s senior management made about Flash Buy on the evening of May 13 earnings call:

Since April, while maintaining order scale, Alibaba has driven a significant improvement in UE through better logistics efficiency and optimized order structure. Huxiu notes that UE, or unit economics, refers to average revenue minus average cost per order.

We are confident in achieving overall profitability in instant retail.

Flash Buy has had a clear positive effect on physical-goods e-commerce, especially in new customer acquisition, user activity, transaction growth, and logistics infrastructure.

The author believes the war for scale, meaning order volume and ranking by share, may be about to end or may already have ended. It is entering a “digestion cycle,” in which Alibaba will refine operations around its existing scale, reduce per-order costs, and use the cultivation of users’ “cross-shopping” habits to bring real incremental growth to traditional e-commerce.

The second statement is the crux. Flash Buy needs to become profitable. This may also be one of the “intersections” described above: Flash Buy needs to explore a model that does not rely on “2025-style subsidies,” yet can maintain scale and drive cross-shopping.

In fact, this is also a kind of expectation for “money coming back.”