Alibaba is redirecting capital away from instant retail and e-commerce toward AI infrastructure and large model capability upgrades, positioning AI as its sole growth path starting this quarter.
Alibaba has once again restructured its core business units, consolidating them into three primary segments, with artificial intelligence accounting for two of them. Notably, only these two AI-related divisions are posting double-digit growth.
In 2024, Alibaba’s portfolio comprised six major businesses: Taobao and Tmall, International Digital Commerce, Cloud Intelligence, Cainiao, Local Services, and Digital Media & Entertainment. At that time, AI-related cloud services represented just one-sixth of the mix, while four of the remaining segments revolved around e-commerce operations centered on the Taobao and Tmall apps.
By the second quarter of this year, International Digital Commerce, Cainiao, and Instant Retail were consolidated under the Alibaba E-Commerce Group. Meanwhile, AI was elevated to two independent business units, shifting its proportion from one-sixth to two-thirds of the company’s strategic focus.
Revenue from China-region e-commerce stood at RMB 110.9 billion, reflecting an 8% year-over-year decline. Within this segment, direct logistics revenue fell 10% to RMB 28.35 billion, while customer management revenue dropped 7% to RMB 82.5 billion.
Instant retail emerged as the sole bright spot, generating RMB 53.3 billion in Q2 revenue—a figure that includes the instant delivery operations of Tmall Supermarket. This represents a robust 45% year-over-year increase. While impressive, the surge is partly attributable to a low base effect; Alibaba only recently entered the food delivery arena last year, meaning prior revenues were comparatively modest.
During the earnings call, Jiang Fan emphasized optimization and loss reduction for instant retail, marking a stark departure from Alibaba’s previous mandate to “not worry about losses for three years and go all out to scale.” This strategic pivot signals a shift away from blind expansion toward sustainable, healthy growth.
Under this new approach, instant retail’s growth momentum has already decelerated. Given the higher comparative base moving forward, the division is likely to revert to low-single-digit growth or even contract in the coming quarters.
The instant retail campaign has demonstrated that Alibaba cannot decisively marginalize Meituan. Heavy capital deployment yields diminishing returns and runs counter to regulatory preferences, ultimately straining the broader commercial ecosystem rather than strengthening it.
In international e-commerce, geopolitical volatility and deglobalization headwinds lie beyond Alibaba’s control. No amount of resource allocation can reverse these macro trends. Much like the other three members of the “Four Little Dragons” expanding overseas, Alibaba can ride industry beta growth, but managerial intervention has limited upside potential.
The foundation for Alibaba’s current AI trajectory traces back to 2008, when Wang Jian proposed the independent development of the “Apsara” (Feitian) cloud operating system. The initiative faced fierce internal skepticism, with whispers labeling Wang Jian a “fraud” circulating quietly within the group for years.
Trained in psychology and having spent a decade teaching engineering psychology after graduation, Wang Jian’s background naturally invited doubt regarding his ability to lead a cloud computing venture. Between 2010 and 2012, Alibaba Cloud consistently ranked last in internal performance assessments for three consecutive years. Engineer attrition peaked at 80%, and annual cash burns yielded no visible breakthroughs.
It took Jack Ma’s personal intervention to steady the team: “I will invest RMB 1 billion annually in Alibaba Cloud for ten years. We will reassess if we fail to deliver.”
Fueled by Ma’s unwavering support and Wang Jian’s eventual fulfillment of commitment, Alibaba Cloud officially launched in 2013, carving out an independent strategic frontier beyond e-commerce.
After weathering the early capital-intensive phase, Alibaba Cloud’s growth flywheel began to accelerate. By the first quarter of this year, the division captured a 32.8% share of China’s IaaS market, up 2.7 percentage points year-over-year. It firmly holds the top position, commanding a larger share than the second- and third-place competitors combined.
Responding to the transformative impact of AI, Alibaba executed a structural realignment this quarter, bifurcating its AI operations into an Infrastructure Layer and an Application Layer. The Infrastructure Layer merges Cloud Intelligence and T-Head into the AI Cloud & Computing Power Services unit, dedicated to delivering foundational infrastructure and compute capacity.
The Application Layer consolidates the AI Model Lab, the Tongyi Qianwen Consumer Business Group, and Tongyi Qianwen Office into the AI Lab & Application Business. This unit focuses on advancing model innovation, consumer-facing applications, and enterprise productivity solutions.
Alibaba’s commitment to deep-tech infrastructure dates to 2020, when it joined peers in pledging RMB 200 billion over three years to fund core R&D in cloud operating systems, servers, semiconductors, networking, and data center construction.
Capital expenditures have accelerated sharply in recent quarters. Q2 capex reached RMB 67.678 billion, surging 75% year-over-year, with the overwhelming majority directed toward AI infrastructure build-out.
In today’s AI landscape, the most reliably profitable segment remains the “picks-and-shovels” or “selling water” business—a domain squarely occupied by the combined Alibaba Cloud and T-Head operations.
According to publicly disclosed information, Alibaba’s latest-generation AI processor, the Zhenwu M890, has achieved commercial deployment across more than 650 external clients spanning over 20 industries, including autonomous driving and internet services, via the Alibaba Cloud platform.
The preceding generation of T-Head chips has already seen production and shipment volumes exceed 500,000 units. The latest generation was deployed on Alibaba Cloud’s AI infrastructure in August, delivered as Supernode configurations. Leveraging the new T-Head “Zhenwu M890” supernodes, Alibaba has established partnerships with Kimi K3 and Tongyi Qianwen 3.8 Max. Consequently, the delivery cycle for large-scale AI data centers has been compressed to just 100 days.
In Q2, AI Cloud & Computing Power Services generated RMB 48.4 billion in revenue, a 45% year-over-year increase. Profitability remains robust, with EBITA reaching RMB 5.6 billion, up 133% year-over-year. This division currently serves as the sole profit anchor, offsetting declines in core businesses and funding continuous investments in the AI application layer.
JPMorgan estimates that Alibaba Cloud’s current ~12% profit margin may not yet fully capture the operational leverage expected once AI infrastructure reaches maturity. This suggests sustained dual growth in revenue and profitability ahead, providing a critical confidence buffer for the group.
While e-commerce is no longer the central battleground in the AI era, Alibaba’s enduring competitiveness owes a substantial debt to Wang Jian, whose vision was once dismissed by insiders as fraudulent.
Nevertheless, despite the formidable strength of its infrastructure (“water-selling”) division, Alibaba begins on level ground with competitors in the application tier.
Regarding model competitiveness, Tongyi Qianwen stands toe-to-toe with Kimi, Zhipu, and DeepSeek. However, in terms of total user base, Qianwen likely trails Doubao, and its agent adoption metrics probably fall behind platforms like Workbuddy and KimiWork.
Earlier this year, Alibaba attempted to drive traffic to Qianwen through its Five-Fu digital campaign. Ultimately, overwhelmed by Douyin’s massive user reach and distracted by e-commerce priorities, Alibaba lost the initial market cultivation race to Doubao.
Today, Qianwen-centric applications remain largely confined to Alibaba’s proprietary ecosystem. Per the earnings report, the Qianwen App has been integrated into Taobao, Tmall, and Taobao Flash Purchase. To date, 250 million users have engaged with AI-driven shopping experiences via intelligent agent features.
In the enterprise segment, Alibaba has launched Qianwen Office, deeply embedding Alibaba Cloud and DingTalk ecosystems to penetrate corporate productivity workflows through AI models and agents.
Remaining tethered to Alibaba’s closed ecosystem inherently restricts broader user acquisition. This structural limitation, which has plagued the company for years, remains acutely pronounced in the AI era.
In the AI application space, Alibaba holds no guaranteed winning hand. Technologically, it faces intense competition from Doubao, Yuanbao, and emerging pure-play model developers such as Kimi, Zhipu, and DeepSeek. Although Alibaba previously held an industry lead, recent months have been dominated by Kimi K3, Zhipu GLM 5.3, and DeepSeek V4. Its competitive edge typically lasts only about half a month.
Consequently, the commercialization challenges in this segment are evident, yet investment cannot be curtailed. Q2 revenue for the AI Lab & Application Business totaled RMB 3.3 billion, up 16% year-over-year. While growing, both absolute scale and growth rate pale in comparison to the infrastructure division.
Furthermore, of that RMB 3 billion, DingTalk contributes a baseline revenue stream. According to Caixin, DingTalk’s full-year 2025 revenue target is RMB 4 billion, averaging roughly RMB 1 billion per quarter. This underscores that AI application commercialization remains far from achieving economies of scale.
Break-even remains distant. The division reported an EBITA loss of RMB 13.86 billion, widening 330% year-over-year. Massive capital is required to sustain uninterrupted model training, yet clear and expansive commercialization pathways remain elusive, making the near-term outlook exceptionally challenging.
Evaluating Alibaba’s current posture reveals a clear dichotomy: Alibaba Cloud anchors the AI infrastructure thesis with rapid revenue growth and proven profitability, though it demands sustained forward-looking investment and capex to defend its lead. The AI application layer requires active commercialization experimentation but lacks the inherent traffic advantages necessary to replicate its historical e-commerce dominance. Meanwhile, the core e-commerce business now merely tracks macroeconomic beta growth, leaving Alibaba unable to unilaterally reshape the competitive landscape.
Moving forward, Alibaba will redirect capital away from instant retail and traditional e-commerce, channeling resources instead into infrastructure expansion and foundational model enhancement, awaiting the eventual commercial inflection point. Starting this quarter, AI has unequivocally become Alibaba’s only viable path forward.