---
title: "BABA (Trans): Cloud +45%, CapEx RMB 67.7 bn in the quarter"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/dolphin/post/43516618.md"
description: "External cloud revenue growth accelerated to 45%. AI-related product revenue recorded triple-digit growth for 12 consecutive quarters, reaching RMB 12.376 bn.However, Qtr CapEx of RMB 67.678 bn drove a FCF net outflow of RMB 44.67 bn. Group Adj. EBITA fell 30% YoY.Management shortened the AI payback period to 2.5 years, even 2 years. They also reaffirmed 2030 targets of $100 bn external cloud revenue and a 20% EBITA margin."
datetime: "2026-08-20T14:03:44.000Z"
locales:
  - [en](https://longbridge.com/en/dolphin/post/43516618.md)
  - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/43516618.md)
  - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/43516618.md)
author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)"
generator: "portal-rs"
---

# BABA (Trans): Cloud +45%, CapEx RMB 67.7 bn in the quarter

**Below is Dolphin Research's transcript of** $Alibaba(BABA.US) **FY27Q1 earnings call.**

**I. Key Financial Highlights**

1\. **Shareholder returns**: In the quarter ended Jun 30, 2026, BABA repurchased 13.4 mn ordinary shares (Approx. 1.7 mn ADS) for $162 mn. Management will continue disciplined capital allocation between AI+Cloud growth investment, buybacks and dividends, adjusting priorities with market conditions and strategy.

2\. **CapEx, 3-year budget and ROI framework**

a. CapEx was RMB 67.678 bn ($9.975 bn), +75% YoY (vs. RMB 38.676 bn). Growth reflected procurement-cycle timing, increased CPU capacity to support accelerating AI agent adoption, and higher chip/component prices.

b. Of the planned RMB 380 bn over three years, RMB 190 bn had been deployed by end-Jun, tracking plan. Management cautioned against annualizing this quarter's spend or assuming a linear cadence.

c. ROI: Chip-enabled servers typically break even in ~3 years on a 5-year life, leaving at least 2 years of positive FCF post payback. With higher GPM and rising in-house chip substitution, payback could shorten to 2.5 years or even 2 years.

3\. **Key quarterly metrics**

a. Consolidated: Revenue RMB 269.0 bn (+9% YoY), driven by Cloud and on-demand retail. Adj. EBITA RMB 27.3 bn (-30% YoY) on higher tech investment, partially offset by Cloud ops improvements and efficiency gains; GAAP net income RMB 10.4 bn (-75% YoY) on lower operating profit, reduced disposal gains, and FV changes of equity investments.

b. E-com Group: Revenue RMB 205.862 bn (+4% YoY), including China on-demand retail RMB 53.295 bn (+45%, led by Freshippo and Taobao Flash). Adj. EBITA RMB 39.749 bn, roughly flat YoY; customer management revenue -7% YoY, or +1% YoY excluding headwinds from new biz initiatives.

c. AI & Tech: AI Cloud & Compute Services revenue RMB 48.437 bn, with total and external revenue growth both accelerating to +45%. Adj. EBITA RMB 5.628 bn; margin 11.6% (CFO's '12%' was rounded). AI-related product revenue RMB 12.376 bn ($1.824 bn), 35% of Alibaba Cloud external revenue; AI Labs & Apps revenue RMB 3.338 bn with adj. EBITA loss RMB 13.861 bn; All Other revenue RMB 28.803 bn (flat) with adj. EBITA loss RMB 3.343 bn.

d. Cash flow & balance sheet: Operating cash flow RMB 22.945 bn (+11% YoY vs. RMB 20.7 bn). FCF net outflow RMB 44.67 bn (vs. RMB 18.8 bn outflow), driven by Cloud infra investment; net cash was Approx. $30.7 bn as of Jun 30, 2026, or Approx. $46.5 bn excluding debt maturing in 5+ years.

4\. **Segment reporting changes**: From this quarter, segments are four: Alibaba E-com Group; AI Cloud & Compute Services (incl. Cloud Intelligence Group and T-Head); AI Labs & Apps (incl. model labs, Qwen consumer and QwenWork); and All Other. E-com revenue is further split into China E-com, China on-demand retail, Intl E-com, and Global wholesale.

**II. Earnings Call Details**

**2.1 Management Remarks**

1\. **AI and Cloud commercialization**

a. External cloud revenue growth accelerated to 45%, a 22-quarter high; Alibaba Cloud EBITDA rose +133% YoY, with adj. EBITA margin at 11.6%, up ~440 bps YoY. 

b. The 45% growth was broad-based, driven by compute, storage, MaaS and AI apps; the company is proactively exiting low-margin lines to upgrade growth quality. 

c. AI product revenue was RMB 12.376 bn, implying an annualized run-rate above RMB 49.5 bn (Approx. $7.3 bn). It has delivered triple-digit growth for 12 straight quarters, now 35% of external cloud revenue, with GPM well above the segment Avg.

d. AI revenue spans compute, MaaS and application layers; demand growth in any layer translates directly into monetization. This structural advantage supports sustained rapid growth in recurring AI product revenue.

e. The AI agent boom is driving token and GPU demand, while lifting CPU, storage, DB and networking needs across traditional cloud. Alibaba Cloud is upgrading to an Agentic Cloud.

f. Including MaaS, ARR for models and application services has exceeded RMB 16.0 bn. Based on market feedback and backlog, compute demand will outstrip supply; as capacity ramps, AI and Cloud revenue growth should further accelerate with improving profitability over the next few quarters.

2\. **Full-stack AI: in-house chips and data centers**

a. Deeper synergy between in-house T-Head chips and proprietary foundation models is improving AI commercialization efficiency. T-Head now offers a full in-house line across GPU, CPU and networking chips.

b. As of early Aug, the Zhenwu series has served 650+ customers on Alibaba Cloud. 

c. Ultra-node instances powered by T-Head's next-gen Zhenwu M890 AI processors have entered scaled commercial deployment on Alibaba Cloud; supply will ramp in 2H to meet strong demand. 

d. Zhenwu M890 ultra-nodes efficiently run inference workloads of 2T+ parameter foundation models; models such as Qwen3.8-Max are now offered via MaaS on this platform. 

e. At the DC layer, delivery cycles for ultra-scale AI data centers have been compressed to 100 days, among global leaders, enabling faster global compute rollout. 

3\. **Model iteration and open-source ecosystem**

a. Release cadence tightened over the past month across language, image, audio, video and music models, with performance in the global first tier. 

b. Last week, weights for the 2.4T-parameter Qwen3.8-Max and the Qwen3.8-27B series were open-sourced. 

c. Qwen downloads have exceeded 3.0 bn cumulatively, with 300k+ derivative models built on top. A thriving open-source model ecosystem will back-feed cloud demand and form a virtuous cycle.

4\. **AI-native applications**

a. On the enterprise side, QwenWork was launched as an AI productivity suite for office scenarios, delivering agentic capabilities at scale; productivity agents are expected to be another ARR growth engine. 

b. On the consumer side, Qwen App users continue to grow, with expanding value-added services; since launch, 250 mn users have completed their first AI-powered shopping experience across e-com and other services via Qwen App. 

c. Tight coordination between Alibaba Token Hub and Alibaba Cloud has built an efficient flywheel across compute, models, tokens, apps and monetization. 

d. AI Labs & Apps losses narrowed materially QoQ, driven by lower Qwen App marketing spend; with improving training efficiency and ongoing opex optimization, losses should continue to narrow over the next few quarters. 

5\. **E-com and on-demand retail**

a. On-demand retail grew 45% with sharply narrower losses and improved unit economics QoQ, driven by higher AOV and logistics efficiency, while maintaining share. 

b. Legacy e-com aims to keep profits stable, while on-demand retail pursues profitability improvement; AliExpress delivered operating profit this quarter. 

c. E-com Group adj. EBITA was roughly flat YoY despite increased UX and tech investments, underscoring cost discipline. 

6\. **Strategy**

a. The AI commercialization inflection was crossed last quarter; this quarter saw faster growth and margin expansion. AI is increasingly self-funding and self-sustaining, giving the company confidence to invest more.

b. AI is now Alibaba's most certain growth engine. AI has moved from incubation to scaled commercialization, giving greater strategic and financial flexibility across full-stack capabilities and consumer opportunities.

**2.2 Q&A**

**Q: What drove the sharp CapEx increase, and how should we read the trend for the next few quarters? Any updates to the RMB 380 bn 3-year budget?**

A: We announced a RMB 380 bn 3-year plan in Feb last year; by end-Jun, RMB 190 bn had been deployed, broadly on track. This quarter was higher due to hardware deliveries following varied procurement cycles, which are inherently uneven, so timing drove the increase; we also increased CPU purchases as agent-era demand surged, and semi component prices rose.

Therefore, this quarter's spend should not be multiplied by four for annualization, nor assumed to progress linearly. Build-out continues at a steady cadence.

**Q: Why is full-stack AI a heavy-asset model requiring front-loaded investment?**

A: It is a heavy-asset business. All monetization forms—software subscriptions, API calls, MaaS, training, inference—require compute centers at each step, and monetization only happens once capacity is in place, so we must invest upfront to scale and monetize across these vectors.

This is why we entered a hardware-heavy investment cycle from 2025: to capture future growth, we first need the CapEx to build the necessary capacity. 

**Q: Why is the return on AI-related CapEx highly certain? How can you lift ROIC and shorten payback?**

A: Industry consensus is AI compute shortage likely persists until at least 2030, so from a sector view, investing in AI compute has high certainty. At current Avg. GPM, AI CapEx can break even in ~3 years, and GPM is rising; we expect to shorten payback to ~2.5 years. Post a 3-year payback, these assets generate robust, steady cash flows—for instance, A100s bought in 2020 and even V100s from 2018 still run at full load today.

We have three levers to lift GPM and ROIC. First, advance frontier models to raise AI product margins, expand higher-margin MaaS, and optimize hardware/software mix to lift overall GPM; this is already showing—Alibaba Cloud segment margin rose ~440 bps to 11.6% this quarter.

Second, scale deployment of in-house chips. T-Head covers GPU, CPU and networking—the costliest DC components are chips and storage—so higher in-house penetration replacing external commercial chips will meaningfully lift margins and profitability.

Third, improve cash-efficiency, e.g., co-building DCs with partners and adopting prepayment models for compute services. With these, payback can shorten to 2.5 years or even 2 years.

A simple framework: at current AI product GPM and a 3-year payback assumption, keeping growth ≤33% would already turn cash flow positive. But that is not our strategy now—AI remains early, and we choose to invest CapEx proactively to scale fast; as product margins improve and in-house chips substitute more, payback shortens to ~2.5 years, letting us pursue 40%+ growth while staying FCF-positive—our long-term direction.

**Q: What's the latest on on-demand retail, and post-reorg, what are strategic priorities across the E-com Group lines?**

A: In the new fiscal year, we reorganized E-com, and will update around four pillars: China E-com, on-demand retail, Intl E-com, and global B2B. 

On China E-com: the domestic market faces near-term macro pressure, while our long-term strategy is to strengthen supply and use AI to enhance shopping experience and improve operations. On supply, since last year Taobao/Tmall supported original-brand merchants and unlocked quality white-label supply from key industrial belts; we will deepen cooperation with top brands to drive stable, sustainable growth—Tmall remains a core operating ground for major brands and many original merchants.

We are going deeper into industrial belts to source directly, enabling factories to operate on-platform, and using platform AI to simplify white-label operations via managed services; the share of GMV from managed industrial-belt merchants is rising. During the 618 season, despite macro pressure, results met expectations with solid growth among core merchants.

AI has notable opportunities on both demand and supply. For consumers, we will keep rolling out AI-powered experiences such as multimodal search and virtual try-on, with goals to improve current journeys and unlock new forms of AI interaction; we already see meaningful efficiency gains in product recommendations.

For merchants, AI is now widely used; we are applying AI across operations to amplify capabilities, especially in data analytics, ad marketing and CS—areas with clear ROI—and will partner with QwenWork to launch AI agents tailored for e-com scenarios. 

On-demand retail: after over a year of investment, Taobao Flash has scaled materially with improved user perception, richer supply, better logistics and higher orders. Last quarter, users and orders grew further while UE improved markedly and losses narrowed.

We will accelerate integration of Freshippo, Tmall Supermarket and others, expand non-food categories, and focus on front-warehouse buildout; Freshippo expanded front warehouses over the past year, driving YoY GMV growth. 

We will broaden category coverage and innovate in key touchpoints to improve CX. We expect non-food GMV in on-demand retail to exceed food within the next fiscal year, lifting multiple physical categories across E-com.

The business is expected to reach overall profitability in FY29. Longer term, it could contribute ~30% of platform GMV, forming the second growth curve for E-com.

Third, Intl E-com: near-term growth is pressured by tariffs and geopolitics, but cross-border ops have improved profitability while maintaining volume growth amid a complex environment. We see long-term potential across both scale and profitability; our local platforms in Turkey and the Middle East are growing fast, with operational efficiency improving in SEA.

Fourth, global B2B: 1688 and Alibaba.com have grown for two decades, and AI may profoundly reshape B2B platforms, potentially reinventing the model, with models playing an increasingly central role. We launched 'Accio Work', an AI agent for cross-border merchants, attracting 50k+ paying merchants shortly after launch.

AI is changing how B2B, especially cross-border, operates; backed by two decades of experience, we can create new models and opportunities in AI-era B2B and trade. Overall, we have reset E-com strategy across key areas and will leverage supply-chain coordination and AI advantages to unlock greater growth, diversify revenue and profit streams, and make segment development steadier.

**Q: What is the revenue growth cadence for Cloud over the next few quarters, and what sustains acceleration?**

A: External revenue in AI & Cloud has accelerated for nine consecutive quarters, reaching 45% this quarter. Demand is very strong, and our products have clear advantages vs. peers, so we expect further acceleration in the coming quarters.

AI revenue was RMB 12.376 bn this quarter, annualizing to Approx. $7.3 bn; for next quarter, we see the annualized figure approaching $10 bn, with strong growth. We also expect EBITDA margin to improve QoQ in the next few quarters.

An important driver is rising MaaS demand. MaaS demand grew sharply this quarter, and with ongoing inference-efficiency gains, MaaS ARR has topped RMB 16.0 bn—this is the latest figure as of Aug.

Strategically, Alibaba's AI investment differs fundamentally from pure-play AI: we invest intensively across chips, AI cloud infra and models—the three critical fields—and maintain leadership in each. As AI remains early and value may migrate among chips, cloud, models and apps, full-stack investment ensures best-in-class service and economics and sustained competitiveness through each tech phase.

**Q: Longer term, what are the core Cloud growth drivers, and how do they differ from near-term drivers?**

A: Near term (the next 1–2 years), commercial inference demand has grown exponentially since late 2025, marking a structural shift—compute has become the core asset driving AI revenue, and all AI monetization modes center on compute. Industry consensus also sees compute undersupply persisting; moreover, higher-margin MaaS inference changes the equation: compute has shifted from a cost center to a core productive asset tied directly to revenue.

With GPUs used broadly and high-priced compute still scarce, pricing is converging toward the highest-margin monetization modes, influencing pricing for nearly all GPU-related products. Alibaba's leading multimodal models give us clear advantages in realizing compute value and anchoring pricing—both for new customers and renewals—supporting margin improvement in the next year-plus.

Long-term drivers are scale and network effects. The true 'super app' is cloud-based AI compute itself—training, inference, AI software and agents all run on full-stack AI cloud, needing GPUs, CPUs, storage, DBs, virtualization and toolchains.

AI Cloud is like a super city: workloads are residents, and ever-iterating full-stack services are infrastructure that attract new residents and increase stickiness—creating powerful network and scale effects. We operate the most data centers among Asia cloud providers, enjoying scale economies; large-scale deployment of in-house T-Head AI chips helps avoid external GPU price premiums, protecting margins; and frontier in-house models give us strong pricing power on compute.

Given sector trends and product advantages, we have high confidence in reaching $100 bn external Cloud revenue by 2030 and good visibility to 20% margin levels. 

**Q: MaaS ARR exceeded RMB 16.0 bn as of Aug. Will the year-end RMB 30.0 bn target be revised?**

A: MaaS is growing very fast, with ARR above RMB 16.0 bn as of Aug. Given current momentum and new models in the pipeline, we remain confident in reaching RMB 30.0 bn ARR by year-end.

**Q: In MaaS, what are revenue shares of in-house vs. third-party models? How do rising competition and more open-source models impact margins?**

A: On our MaaS platform, in-house models still contribute the majority of revenue, with third-party models also sizable. Customers often want to use multiple models in their AI apps because models differ in strengths; having more open-source models for inference is positive for us.

On margins, hosted in-house and third-party models on platforms like Bailian achieve highly comparable margin levels. We build in-house models both to elevate intelligence and as part of our AGI pursuit; strictly within MaaS, margin profiles are very close, and a rich open ecosystem is highly beneficial to a cloud provider like Alibaba Cloud.

**Q: In a full-stack ecosystem, where does value accumulate and monetize?**

A: It is a long-term call with high uncertainty, and we invest across the stack so value—wherever it concentrates or migrates over time—stays within our ecosystem. My short-term view: most value will sit in chips and AI cloud infrastructure when technology is early and supply is constrained—globally, value concentrates in infrastructure and core hardware, i.e., chips and storage—so at Alibaba, compute, cloud infra and AI are integrated into one core business.

**Q: What is the ultimate business model for foundation models? Is today's API monetization the endgame?**

A: Opinions differ industry-wide and internally; my personal view is that API-based monetization is transitional, not the endgame. We invest massive compute across the platform not to maximize short-term API revenue.

At or near AGI, the endgame is delivering real products and outcomes customers want—building R&D that outputs products and operational results. Heavy investment and the arms race among model companies aim at that endgame, which I believe will monetize far above today's API-call services.

**Q: If value concentrates in hardware and compute over time, and given Gov.-led allocation in those areas, how do you view competition going forward?**

A: A few points on T-Head chips, a topic we have not discussed much with investors. We have manufactured and shipped 500k+ units of the prior generation; the latest generation was deployed on Alibaba Cloud as ultra-nodes in Aug, and we are among the very few able to deploy such in-house domestic chips at scale.

A unique aspect of T-Head chips is their GPU-centric architecture, well-suited for both training and inference. Hundreds of companies already use these chips via Alibaba Cloud for inference and training, spanning embodied intelligence, autonomous driving and model companies; our second-gen will begin R&D in 2H, with very high compute and interconnect bandwidth, capable of directly replacing existing chips.

We hold a distinctive position in chips, especially for large-scale model training, and I do not think any Gov.-led allocation mechanism can produce chips with such real competitiveness. As a core part of Alibaba Cloud, T-Head has highly certain prospects, and we remain confident in our core edge here—engineers across many fields use these chips widely.

In short, T-Head chips are the best domestic chips supporting cross-industry training and inference—we are truly No.1; in future capacity and deployments, we can confidently say we will be at least top two. And on reaching customers, Alibaba Cloud is the largest player in China's Cloud and AI market, giving us strong distribution advantages; from this angle, we are highly confident in T-Head's long-term commercial value.

\<End of transcript>

**Risk disclosure and disclaimer:**[**Dolphin Research Disclaimer & General Disclosures**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [09988.HK](https://longbridge.com/en/quote/09988.HK.md)
- [89988.HK](https://longbridge.com/en/quote/89988.HK.md)
- [BABA.US](https://longbridge.com/en/quote/BABA.US.md)
- [HBBD.SG](https://longbridge.com/en/quote/HBBD.SG.md)

## Comments (1)

- **Fattycat · 2026-08-20T14:41:12.000Z**: Thanks Dolphin ☺️


---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**