Why this update. The original thesis framed Alibaba's EBITA decline as deliberate capital spending on cloud infrastructure, comparable to Amazon 2014. That framing remains correct. What has become clearer since the May 14 annual results is what the cloud capex is actually financing. The capex is not adding more IaaS capacity, where renting bare GPU and CPU time competes purely on price and earns thin margins. It is building the inference layer (MaaS) for an emerging token-consumption economy where AI agents, not humans, will drive the bulk of API calls. The agentic shift changes what cloud revenue is worth, and BABA's current 10.8× EV/EBIT does not price it.
The agentic shift, and why token volume is the right frame. Until 2024, AI usage meant a human typing a prompt and reading a response. A single query consumed a few thousand tokens. The next phase, already visible in 2025 production deployments, is agentic: AI systems that plan multi-step tasks, call external tools, verify their own outputs, and increasingly negotiate with other AI agents. A single agent task can consume 100 to 1,000 times more tokens than one human prompt. Multi-agent workflows where agents query, validate, and transact with each other compound this further.
The available proxy for this trajectory is Google. At I/O 2024, Sundar Pichai disclosed 9.7 trillion tokens processed per month across Google's AI surface. At I/O 2025, twelve months later, that figure was 480 trillion. A 50× increase in one year, driven mostly by Gemini integration into Search and Workspace, not by agentic workloads, which were still nascent. The agentic phase is the next leg of that curve, not the previous one. If token consumption grows even 10× per year from here, the cumulative volume by 2028 sits two orders of magnitude above today.
The cloud stack, in four layers. Alibaba Cloud, like AWS and Azure, sells different things at different margins. Token consumption flows through a specific layer in the stack, and that layer carries the highest gross margin once the model is trained.
| Layer | What is rented | Gross margin profile |
|---|---|---|
| IaaS (Infrastructure) | Raw GPU and CPU compute, storage, network bandwidth | Low, commodity pricing |
| PaaS (Platform) | PolarDB managed databases, Function Compute serverless, MaxCompute data warehouse | Medium, software is the value-add |
| MaaS (Model) | Qwen3 large language model inference, billed per million tokens consumed | 60–80% once the model is trained |
| SaaS (Software) | Finished end-user applications | Variable, depends on product |
PaaS: the customer rents a working database (PolarDB) or a serverless runtime (Function Compute) instead of a bare server. Alibaba operates the software, patches it, scales it. The customer pays a higher rate than raw IaaS because the managed layer eliminates ops work. PolarDB competes directly with Amazon RDS and Aurora; MaxCompute competes with Google BigQuery and Snowflake. This is the layer where Alibaba Cloud built its enterprise credibility before AI.
MaaS: the customer sends a prompt to Qwen3 through an API call and pays per million tokens (roughly per million words processed). No GPU rental, no model training, no infrastructure to manage. Alibaba already paid the training cost; every subsequent inference is near-pure margin. A startup building a chatbot, a bank automating document review, a logistics firm summarising shipping manifests, all consume MaaS without ever touching a server. This is the layer that has no equivalent in pre-2023 cloud reporting and where the disclosure asymmetry sits.
Why MaaS deserves its own weighting in the valuation. Qwen3 is the inference engine Chinese enterprise agents will call. Qwen has over 300 million monthly active users. Every workflow that today returns a single answer will tomorrow execute a chain of tool calls, each one consuming tokens at 60–80% gross margin without additional capital spending. AWS Bedrock and Azure OpenAI run the same playbook. If Alibaba begins reporting MaaS revenue separately, as AWS did with Bedrock from 2024, the market can finally model the high-margin layer in isolation rather than buried inside the consolidated cloud number. That single disclosure decision can drive multiple expansion before any operational change.
What this means for the price: re-rating to HK$165–200 over 12–18 months. Price has drifted to HK$127.00 (May 22 close) since the May 14 update, leaving the EV/EBIT multiple at roughly 10.8×. Tencent sits at 15× for a comparable mix of platform commerce and cloud. The arithmetic: if Alibaba re-rates to 14× with EBIT flat, the stock prints HK$165 (+30%); at 15× parity with Tencent, HK$176 (+39%); at 14× with EBIT growing 15% on cloud margin inflection, HK$190 (+50%). This is the near-term monetisation of the thesis. The token volume curve underneath is the longer leg.
Five catalysts trackable through HKEX filings before FY2027:
- Q1 FY2027 results (August 2026): cloud growth sustained above 40%, AI-related share rising from 30% toward 35%+
- MaaS revenue breakout: any separate line item disclosure for model inference revenue
- T-Head AI chip shipment milestone: next disclosure beyond the current 470,000 cumulative units, particularly the external-customer ratio
- International datacenter activations: first revenue contributions from Brazil, France, Netherlands buildout disclosed in segment notes
- Southbound Stock Connect flows: net mainland buying into 9988.HK as a re-rating accelerant
FY2026 Annual results (filed May 13, 2026): Full-year revenue +3% YoY (year ended March 31, 2026). Cloud Intelligence Group external revenue +40% YoY; AI-related products represent 30% of cloud revenue. GAAP net income RMB 102.1B. Dividend: US$1.05/ADS declared.
Non-GAAP earnings fell materially as Alibaba front-loaded AI data-center and GPU expenditure. T-Head AI chip annual revenue reached RMB 10B with 470,000 cumulative units shipped (60%+ serving external customers). Quick commerce revenue +56% YoY to RMB 20.8B.
Cloud external revenue growth accelerated from +36% (Q3 FY2026) to +40% (Q4 FY2026), trajectory confirmed. Price at HKD 138.70 (+12.8% from HKD 122.90 at analysis). Next catalyst: April–June 2026 quarter results (expected August 2026).
Alibaba's adjusted EBITA fell 57% in Q3 FY2026 because the company is building data centers, buying GPUs, and constructing the infrastructure for China's largest cloud platform. Cloud Intelligence Group revenue grew 36% year-over-year. At HK$122.90, EV/EBIT is 10.8× against Tencent at 15× and Amazon at 35×. The company holds approximately US$40 billion in net cash (13% of market cap) with over US$12 billion in active buyback authorization. The market has priced the EBITA decline. It has not priced the cloud trajectory.
| Revenue | +9% YoY |
| Cloud Revenue | RMB 43.3B/Q +36% |
| Adj. EBITA (Reported) | −57% YoY |
| Net Cash | ~US$40B |
| Active Buyback | US$12B+ |
| 52W High / Current | HK$186.20 / 122.90 |
| Amazon | 35× |
| Microsoft | ~25× |
| 19× | |
| Tencent | 15× |
| Alibaba (9988.HK) | 10.8×−28% |
Source: HKEX filing Q3 FY2026 · Peer multiples Apr 2026
Alibaba's adjusted EBITA fell 57% in the December quarter. Revenue grew 9% after stripping out disposals. One of those numbers is producing the headlines. The other is producing the business.
At HK$122.90, the market values all of Alibaba (China's dominant cloud platform, its largest e-commerce marketplace, a logistics network spanning 200 countries) at roughly 10.8× forward EV/EBIT. Tencent trades at 15×. Amazon at 35× forward earnings. Google at 19×. The company holds approximately US$40 billion in net cash, which represents 13% of its market capitalization. That cash is not being hoarded. Over US$12 billion in buyback authorization remains active.
The EBITA decline is real. It is also deliberate. That distinction is the entire thesis.
What Alibaba Does
Alibaba Group Holding (9988.HK), one of the largest technology stocks listed on Hong Kong's HKEX, operates China's largest digital commerce ecosystem. Taobao and Tmall together handle more gross merchandise volume than any other marketplace in the country. That is the core business: profitable, cash-generative, and growing at low single digits in a saturated domestic e-commerce market.
The second business is Alibaba Cloud Intelligence Group. Q3 FY2026 revenue: RMB 43.3 billion, up 36% year-over-year. AI-related revenue has grown at triple-digit rates for ten consecutive quarters. Alibaba Cloud is the number one public cloud provider in China by market share.
Eddie Wu took over as CEO in October 2023. Since then, the company has executed a restructuring that divested non-core assets, consolidated reporting segments, and redirected capital toward cloud infrastructure. The free cash flow decline of 71% year-over-year in Q3 FY2026 is a direct consequence of that capital reallocation: data center construction, GPU procurement, network buildout. The cash is leaving the balance sheet through capital spending, not through losses. US$12 billion in buyback authorization, active at the same time capital spending is at a multi-year peak. A company that thought its stock was overpriced would not do both.
Why Alibaba Trades at a Discount
Three forces keep the price where it is.
First, the EBITA decline. A 57% drop in adjusted EBITA triggers every institutional screen that filters for earnings deterioration. Quantitative funds sell automatically. Analysts revise targets downward. The number is ugly on a spreadsheet with no context column.
Second, tariffs. Between February and April 2026, the stock fell 16.8% from its 52-week high of HK$186.20 as US-China trade tensions resurfaced. The tariff selloff has compressed the entire China tech sector, regardless of individual fundamentals.
Third, the AI investment timeline has no confirmed payoff date. Cloud capital spending is accelerating. Revenue is growing. But the inflection point where cloud margins recover is not yet visible in the reported numbers. The market prices what it can see. It discounts what it cannot.
Amazon's market cap tripled while its EBITA was near zero. The capital spending cycle is not a threat to the thesis. It is the thesis.
The Investment Cycle, the Cash Policy, the Revenue Scale
- 01 Amazon reported near-zero consolidated earnings from 2012 to 2016 while building AWS. During that period, Amazon's stock rose from US$250 to US$750, not because earnings improved, but because investors who understood the capital spending cycle recognised that AWS would eventually generate margins above 30%. Alibaba Cloud is following the same sequence. The revenue growth rate (36%) is comparable to AWS at a similar stage. The EBITA compression is the cost of building the infrastructure that produces those margins. The parallel is not a metaphor. It is a documented pattern: heavy capital spending depresses reported earnings for 3 to 5 years before the installed base generates recurring, high-margin revenue.
- 02 Net cash of approximately US$40 billion is not a static number. It is a policy statement. A company with US$40 billion in cash and US$12 billion in remaining buyback authorization is telling the market two things simultaneously: the balance sheet can absorb a multi-year investment cycle without dilution, and management believes the stock is underpriced at current levels. At 13% of market capitalization, the cash alone provides a floor that does not exist for most technology companies making comparable infrastructure bets.
- 03 Cloud Intelligence Group revenue of RMB 43.3 billion in a single quarter means that, at that quarterly pace, annual revenue exceeds RMB 170 billion. For context, AWS generated US$62 billion in 2022 (roughly RMB 440 billion) at margins near 30%. Alibaba Cloud is at approximately 40% of that revenue scale in a market where AI adoption is accelerating, government policy explicitly supports domestic cloud infrastructure, and the competitive gap between Alibaba and its nearest Chinese rival (Huawei Cloud) is widening, not narrowing. The question is not whether Alibaba Cloud will be profitable. The question is when.
What Alibaba Is Actually Worth
The valuation framework here is comparative, not speculative.
| Component | Basis | Value |
|---|---|---|
| E-commerce (Taobao + Tmall + International) | ~10× EBIT, peers 12–15× | Core of current market cap |
| Cloud Intelligence Group | 5× revenue (AWS trades at 8×) | ~RMB 850B / ~HK$930B |
| Net cash | Mark-to-market | ~US$40B / ~HK$312B |
| Buyback (remaining authorization) | Executed over 12–18 months | >US$12B accretion |
| EV/forward EBIT | Current multiple | ~10.8× |
| Tencent EV/EBIT | Peer comparison | ~15× |
| Current price | April 9, 2026 | HK$122.90 |
At 10.8× forward EV/EBIT, the market assigns no premium for the cloud business growing at 36%. If cloud were valued separately at even 5× revenue (half of what AWS commands), the implied value of the e-commerce business would be close to zero. That arithmetic does not require optimism. It requires a calculator.
Risks We Are Not Downplaying
Risk 1: Investment timeline uncertainty
EBITA recovery depends on cloud margins inflecting upward as the installed base of AI infrastructure generates recurring revenue. If AI monetization in China is slower than expected (regulatory friction, enterprise adoption delays, pricing pressure from Huawei or Baidu), the recovery could extend to FY2028 or beyond. Two more years of compressed margins means two more years of institutional selling pressure. The Amazon parallel holds only if the investment converts to margin. Amazon proved it could. Alibaba has not yet.
Risk 2: E-commerce competition
PDD Holdings (Pinduoduo) continues to gain share in price-sensitive segments. Douyin (TikTok's Chinese parent) is building a commerce ecosystem that captures transactions inside short-video content. Taobao and Tmall remain dominant by GMV, but market share in e-commerce is never permanent. If core commerce profitability erodes while cloud is still in investment mode, the company faces margin pressure from both sides of the business simultaneously.
The Decision
Written at HK$122.90 on April 14, 2026.
This is an investment cycle thesis, not a growth thesis. The stock does not need Alibaba to accelerate. It needs the market to recognise that the EBITA decline is capital spending, not deterioration, and that capital spending has a documented pattern of converting to margin in cloud businesses.
| Scenario | Observable Signal | Price Implication |
|---|---|---|
| Bull | Cloud margins inflect H2 FY2027 + e-commerce stable | HK$170–200 |
| Neutral | Cloud grows >25% but margins flat through FY2027 | HK$130–150, thesis intact |
| Bear | EBITA negative 2 consecutive quarters, or net cash < US$20B | HK$85–95, thesis invalidated |
Three signals invalidate the thesis: EBITA turns negative for two consecutive quarters, signaling that the investment cycle has become a structural loss; net cash falls below US$20 billion, meaning the balance sheet can no longer self-fund the transition; or the Chinese government reverses its current supportive posture toward Alibaba, reopening the regulatory risk that compressed the stock in 2021–2022.
The 52-week range tells the story: HK$95.70 to HK$186.20. At HK$122.90, the stock sits in the lower third of that range while the cloud business posts its strongest growth in three years. Cloud Intelligence Group grew 36% in Q3 FY2026 while the stock sat in the lower third of its 52-week range. The fears are in the price. The growth rate is not.
Sources
- Alibaba Group Q3 FY2026 quarterly results: HKEX filing, February 2026
- Cloud Intelligence Group revenue: Alibaba investor relations, Q3 FY2026
- Tech peer EV/EBIT multiples: Bloomberg consensus, April 2026
- Net cash and buyback authorization: Alibaba Q3 FY2026 balance sheet, HKEX filing
- AWS revenue reference: Amazon annual report 2022
- Per-share and EV calculations: own calculations on the basis of HKEX market data as of April 14, 2026