Mispricings usually live in the corners of the market. Small caps nobody covers, emerging markets foreign institutional investors avoid, post-reorganization equities where the old shareholder base sold and the new one has not arrived yet. You expect to find inefficiency in places where capital cannot or will not go.
What you do not expect to find is a mispriced $295 billion megacap trading on the New York Stock Exchange, covered by every sell-side analyst on Wall Street, held by every emerging markets fund on the planet. And yet that is precisely where Alibaba sits in March 2026. Not mispriced by 15%. Mispriced by a factor of more than two.
The marginal price-setter in a stock like BABA is no longer a fundamental analyst. It is a momentum algorithm reacting to consolidated headline metrics. Total revenue up 2%. Adjusted EBITA down 57%. Non-GAAP EPS down 67%. Free cash flow down 71%. These numbers are, taken at face value, brutal.
They are also, taken at face value, wrong. The business underneath them is in the middle of a deliberate transformation that will look obvious in retrospect and is currently invisible to anyone who does not read the segment breakdowns.
There is one question missing from every Western model of Chinese equities. Not earnings growth, or take rate, or cloud margin expansion. A more fundamental question:
I call this the B2G framework – business-to-government alignment. It is not a political loyalty test. It is a structural question about whether a company is building products that improve the metrics by which every provincial governor, every city mayor, every cadre in the Chinese system is evaluated and promoted.
If you are building what the plan calls for, the entire weight of the Chinese state – procurement budgets, permitting, talent incentives, regulatory forbearance – flows in your direction. If you are building something that competes with what the state considers its own domain, that same weight reverses.
This is the single most predictive variable for regulatory risk in Chinese equities, and it is why Alibaba was in trouble five years ago and is not today.
Western coverage of Alibaba reduces a fifteen-year arc to a single moment: Jack Ma's October 2020 speech in Shanghai, the cancelled Ant Group IPO, the punitive antitrust fine. The standard narrative is that an autocrat punished a critic, and Western observers should treat all Chinese companies as politically vulnerable as a result.
Alibaba's relationship with the Chinese state has moved through three distinct phases, which also account for why CATL and BYD never faced the same treatment.
Taobao gave fifty million small merchants and rural producers access to national consumer markets. Manufacturers in Yiwu sold directly to households in Tier 4 cities. Farmers in Shaanxi reached urban consumers without intermediaries. The "Taobao village" phenomenon – four thousand rural communities where digital commerce reversed economic decline – was 共同富裕 (gòngtóng fùyù, common prosperity) in action a decade before the phrase became official Politburo language. Jack Ma was on national television. The state celebrated him.
Alibaba began importing the Western tech monopoly playbook. 二选一 (èr xuǎn yī, "choose one of two") – forcing merchants to exclude competing platforms. Algorithmic price discrimination. Predatory acquisitions of smaller competitors. Data lock-in. These behaviors are tolerated reluctantly even in the United States. In a system where common prosperity is not a slogan but an evaluation criterion the Politburo uses to grade officials, they are negatives. By 2020, "platform economy disorder" had been named explicitly as a target in central economic work conferences. Alibaba had drifted from swimming with the current to swimming across it.
Ant Group was not just a payments company. It had built a shadow banking system – Huabei consumer credit, Jiebei personal loans – lending against thin capital with the returns of a bank and none of the regulation. This directly contravened two binding Five-Year Plan priorities: 金融强国 (jīnróng qiángguó, financially powerful country) and "preventing and resolving major financial risks". When Ma stood on the Bund and called Chinese regulators a "pawnshop mentality," he was attacking the system that was about to discover what his company had been doing. The IPO was pulled. The crackdown began.
Western analysts watched the crackdown unfold across Alibaba, Didi, online gaming, and after-school tutoring – and concluded that the Chinese state had become hostile to private tech, that "China risk" was now systemic, that all Chinese equities deserved a permanent governance discount. That generalization was, and is, wrong.
The counterfactual is sitting in plain sight. At the exact moment Alibaba was being humbled, CATL and BYD were receiving every tailwind the Chinese state could provide. Procurement preferences for new energy vehicles. Subsidies for battery R&D. Fast-track permitting for gigafactories. Talent visas for materials scientists. These companies were swimming with the current – they were building 新质生产力 (xīnzhì shēngchǎnlì, new quality productive forces), the heart of where the Politburo wanted the economy to go. Same period. Same Politburo. Opposite outcomes.
The Alibaba episode happened to be the loudest event of its era, so analysts mistook a company-specific story for a systemic one. The discount is still there in BABA's price, applied to a company that has spent five years climbing back into alignment.
The Alibaba of 2026 is a different company. Eddie Wu, who replaced Ma as CEO, has restructured the entire firm around what he calls the Alibaba Token Hub – create tokens, deliver tokens, apply tokens. On the Q3 FY2026 earnings call, he committed to $100 billion in annual external cloud and AI revenue by 2031. That number maps directly to the 15th Five-Year Plan's target of 12.5% digital economy share of GDP – a target that is now a binding KPI for every governor in the country.
Eddie Wu is not making a forecast but sending an alignment signal – telling Beijing that Alibaba is now building what the plan needs: cloud infrastructure, AI compute, the digital economy layer the entire cadre evaluation system is measured against.
Joe Tsai, the chairman, holds the geopolitical flank – Yale-educated, Canadian citizen, fluent in the language of Western capital markets, personally reassuring institutional investors on VIE structure and delisting risk. Wu builds the product. Tsai holds the bridge.
Alibaba owns a great deal beyond its two operating businesses, and the market is ignoring a large pile of owned-assets while it fixates on consolidated EBITA:
T-Head Semiconductor. Alibaba's chip design subsidiary. 470,000 AI chips shipped through February 2026, over 60% to external commercial customers. Annualized revenue of approximately RMB 10 billion (US$1.45B) – larger than Cambricon (RMB 6.5B / US$940M) and Moore Threads (RMB 1.5B / US$220M) combined, the two largest publicly traded Chinese AI chip companies, which carry a combined market capitalization near $100 billion. T-Head is not yet public, but I predict it soon will be.
Cash. RMB 560 billion (US$81B) gross in cash and liquid investments, minus approximately $40 billion in debt and convertibles (none currently in the money). Net position: roughly $40 billion.
Ant Group (33% stake). Alipay's parent. Restructured, regulated, cash-generative. Alibaba's equity income from Ant was RMB 4.3 billion (US$620M) in the first half of FY2026 alone. Conservative value of the stake: $35 billion.
AIDC. International e-commerce – AliExpress, Lazada, Trendyol. RMB 39 billion (US$5.7B) quarterly revenue, losses narrowing to RMB 2 billion (US$290M)/quarter. Worth perhaps $15 billion on improving trajectory.
Cainiao logistics. China's largest logistics network. Roughly $10 billion in standalone value.
T-Head, against those comparables, carries roughly $60 billion. Total: approximately $160 billion in assets that are not Commerce and not Cloud.
Alibaba's entire market capitalization at $124 per ADS is approximately $295 billion. Subtract the asset pile: the market is paying $135 billion for Alibaba's two cash-generating businesses – China Commerce and Cloud – combined.
Alibaba China E-commerce Group reported adjusted EBITA of RMB 34.6 billion (US$5.0B) in Q3 FY2026, down 43% year-over-year. The headline is ugly. The story underneath it is not.
The decline is almost entirely explained by deliberate investment in quick commerce – the rebranded "Taobao Instant Commerce" (formerly Ele.me). This is an expensive land grab against Meituan and JD. Management confirmed that Q2 FY2026 was peak investment, with per-order losses halving since summer. Unit economics are improving month over month: higher average order value, better fulfillment efficiency, strong customer retention.
Strip out the quick commerce burn and the base e-commerce business – Taobao, Tmall, 1688 – is generating roughly $23 billion in annualized EBITA. This is a utility-like platform with 59 million 88VIP premium members, dominant market position, and modest but real revenue growth through take-rate improvement.
Revenue grows 2-4% annually – slower than GDP, because e-commerce is mature and not particularly B2G-aligned. China Commerce is not what the Five-Year Plan is optimizing for. But it is also set to benefit from Beijing's anti-内卷 (anti-involution) push – the explicit policy directive against ruinous subsidy wars in delivery and commerce. The State Administration for Market Regulation intervened in March 2026 to signal limits on the food delivery price war. This is the regulatory environment protecting margins, not compressing them.
EBITA margins will be permanently lower than the pre-2024 era. Quick commerce carries structural fulfillment costs that traditional e-commerce does not. The old 30%+ margin days are not returning. But the temporary investment phase – the furnace – is fading. By FY2028, commerce EBITA normalizes in the range of $25-27 billion annually.
At 8× a mature, low-growth, cash-generative platform multiple, that is approximately $210 billion.
The market is paying $135 billion for Commerce and Cloud combined. Commerce alone, conservatively, is worth $210 billion.
When Eddie Wu spoke on the Q3 FY2026 earnings call on March 19, the quarter he was describing – January through March 2026 – was 78 of its 90 days complete. MaaS token consumption up 6× was not a forecast; it was his dashboard.
"From H2 2025 to now," Wu said on the same call, "AI has entered the Agentic-driven era," and committed to "$100 billion in annual external cloud and AI revenue [by 2031]."
Three consecutive reported quarters of acceleration: 26% → 34% → 36%. Alibaba Cloud announced price hikes of 5-34% on core AI compute products the day before earnings, effective April 18. That's a Q1 FY27 tailwind but confirms genuine supply constraint. Tencent raised its Hunyuan AI prices by 460% effective March 13. This is not one company's optimism. This is an industry running out of capacity.
Look at the same market that has marked Alibaba down roughly 35% over the past five months. In the same period, Z.ai (formerly Zhipu) IPO'd on the Hong Kong Stock Exchange in January 2026 and has rallied more than 400% from its HK$116.20 offer price, sitting at a market capitalization near $33 billion on roughly $100 million of trailing revenue. MiniMax IPO'd the next day. Moore Threads, the GPU company, jumped 425% on its first day of trading. MetaX surged 693%.
These are the AI "tigers" Western and Chinese investors are climbing over each other to own, on share prices that assume massive future revenue from Chinese AI adoption.
Where do these companies actually run?
Z.ai's partnership with Alibaba Cloud is publicly documented – Alibaba Cloud is its global deployment infrastructure for international expansion, and Alibaba is also a major equity investor. MiniMax has a published case study with Alibaba Cloud detailing its cloud-native data warehouse architecture; MiniMax's models are also natively integrated into Alibaba Cloud's Model Studio.
Z.ai trades at roughly 300× sales. The market is enthusiastically valuing the application layer of the Chinese AI economy. The infrastructure layer underneath it – running the inference, processing the tokens, providing the serverless data warehouse – trades at negative $75 billion.
This is not a subtle mispricing.
T-Head's 470,000 chips are the vertical integration play that makes the entire stack defensible. Qwen models optimized for T-Head silicon, running on Alibaba Cloud infrastructure. Eddie Wu calls it the "AI golden triangle." Caixin reports that Alibaba is now, after Google, the second company globally with in-house full-stack capabilities across foundation models, cloud computing, and chip design.
The cloud thesis is the BABA thesis, and it aligns directly with the 15th Five-Year Plan's digital economy targets. Every governor in China has a career incentive to make cloud, AI, and digital infrastructure work in their jurisdiction. When Eddie Wu commits to $100 billion, he is aligning a corporate target with a national KPI. The regulatory current is flowing in his direction, not against it.
The five-year math: if cloud revenue reaches "just" $85 billion by FY2031 – well short of Eddie Wu's $100 billion target – and EBITA margins reach 12% as capex growth decelerates and pricing power compounds, that is roughly $10 billion in cloud EBITA.
At 25× – a premium growth multiple for the infrastructure layer of China's AI economy, in line with what the market pays for the Chinese AI chip pure-plays – that is $250 billion. Alternative framings at 8× forward revenue or 20× forward EBITA land in a similar range.
| Component | Value (US$B) |
|---|---|
| Assets (T-Head, net cash, Ant 33%, AIDC, Cainiao) | $160B |
| China Commerce (8× normalized EBITA) | $210B |
| Cloud Intelligence Group | $250B |
| Fair value | ~$620B |
| Shares outstanding (ADS equivalent) | 2.35B |
| Fair value per ADS | ~$264 |
Round numbers: I am advocating a fair value of approximately $260–300 per ADS, depending on where cloud lands in the five-year window. Today's price is $124. That is 110% to 140% upside from here on the base case, with meaningful optionality above if cloud accelerates beyond the $85 billion path.
The market, at $124, is pricing Alibaba Cloud – growing 36%, with confirmed pricing power, supply constraints, ten consecutive quarters of triple-digit AI revenue growth, a full-stack chip-to-model-to-cloud capability matched only by Google globally, and direct alignment with the Chinese government's highest-priority economic KPI – at negative seventy-five billion dollars.
This is what happens when momentum algorithms set the marginal price based on consolidated EBITA (−57%) rather than reading the segment breakdown. The same quarterly filing contains both numbers. One requires reading a headline. The other requires reading words.
Not investment advice. Just arithmetic.
I generally do not publish quarterly predictions. A single quarter's revenue or EBITA is not particularly meaningful for a five-year thesis, and the temptation to over-fit short-term noise degrades analytical credibility over time.
But I am making an exception for Alibaba's Q4 FY2026 (January–March 2026), reporting in May. Two reasons.
First, the base effects and real-time signals make this quarter unusually predictable – and the gap between what the data says and what the market expects is unusually wide. Second, the "reading words" methodology is built on falsifiable predictions graded in public. This is the first company-level test.
| Total Cloud Revenue | YoY | Read |
|---|---|---|
| < RMB 45B (US$6.5B) | < 50% | Thesis weakened, reassess |
| RMB 46–47B (US$6.7–6.8B) | 53–56% | Acceptable, base case intact |
| RMB 48B (US$7.0B) | ~58% | Base case confirmed |
| RMB 50B+ (US$7.2B+) | 65%+ | Upside scenario, accelerating |
| Segment EBITA | Read |
|---|---|
| < RMB 35B (US$5.1B) | Burn not declining, investment timeline extending |
| RMB 35–37B (US$5.1–5.4B) | Modest improvement, on track |
| RMB 38–39B (US$5.5–5.7B) | Base case, furnace tapering confirmed |
| RMB 40B+ (US$5.8B+) | Faster normalization than expected |
The CEO told you what was happening, the Five-Year Plan told you where the system is pointed, and the price hikes told you demand exceeds supply.
The market does not often hand you a widely-held, heavily-covered, NYSE-listed megacap trading at roughly 40–45% of fair value. When it does, the reason is usually the same: a structural change is underway that does not yet show up in the headline numbers, and the observer base is still running the old model.
The structural change is three things at once: the pivot from Ant Financial to the Alibaba Token Hub, the B2G realignment from the wrong side of the Five-Year Plan to the right, and 36% cloud growth with ten consecutive quarters of triple-digit AI revenue expansion.
The headline is EBITA down 57%. The reality is a company generating $23 billion in normalized commerce EBITA, running the infrastructure layer for China's AI economy, sitting on $160 billion in non-core assets, and trading at an implied cloud valuation of negative $75 billion.
In Swimming With the Current, I broke my standing rule against making quarterly predictions, and the results confirm exactly why the rule existed.
Cloud revenue: RMB 41.6 billion actual versus RMB 48 billion predicted. Commerce EBITA: RMB 24 billion actual versus RMB 38 billion predicted. Both metrics breached downside thresholds. By the scale I published in April, RMB 41.6 billion reads “thesis weakened, reassess.” That scale graded the forecast, not the sum-of-the-parts. The analytical error was identical in both: underestimating the base scale of Quick Commerce. As volume surged, absolute cash burn accelerated despite improving unit economics.
AI cloud grew sharply quarter-over-quarter, but legacy cloud was still ~80% of the business entering the quarter, and Chinese New Year seasonality in that legacy base held the blended number below my threshold.
On quick commerce, the miss was bigger and more honest. Unit economics are confirmed improving, but volume grew 57% year-over-year – orders at 2.7× prior year, non-food at 3×. The furnace is running hotter because they are shoving dramatically more volume through it. I was 2–3 quarters too early on the burn taper.
The headline numbers triggered pre-market algorithmic selling. The call itself delivered the most explicit forward guidance of Eddie Wu's tenure. The stock went from -3% pre-market to +7% intraday as humans read the transcript, a 10% swing priced entirely off the gap between the headline and the transcript.
For the first time, Alibaba disclosed AI revenue as a standalone number: RMB 8,971 million ($1.3 billion), representing 30% of external cloud revenue, growing at a triple-digit rate for the eleventh consecutive quarter.
Work the arithmetic backwards. AI revenue was RMB 9.0 billion this quarter, growing at a "triple-digit" rate – by the eleventh consecutive quarter at this scale, likely in the 100–150% range rather than the high end. That brackets the prior-year base and, with it, the rest of the cloud business:
| Component | Q4 FY25 (range) | Q4 FY26 | Growth |
|---|---|---|---|
| AI cloud revenue | RMB 3.6–4.5B | RMB 9.0B | ~100–150% |
| Non-AI cloud revenue | RMB 25.6–26.5B | ~RMB 32.6B | 23–27% |
| Total cloud | RMB 30.1B | RMB 41.6B | +38% |
Non-AI cloud grew somewhere between 23% and 27% – healthy for a competitive Chinese market. AI was simply not yet large enough to pull the blended rate to the 58% prediction.
Every leading indicator for cloud economics outperformed the base-case model.
MaaS margins are higher than IaaS, stated explicitly by Eddie Wu on this call. Cloud EBITA grew 57% on 38% revenue growth, so the margin expansion is already in the reported numbers, and Wu put further improvement "visible in the next 1–2 quarters" – not FY28.
How much higher? In 2024–2025, when AI was a negligible part of AliCloud, public cloud margin ran ~8.0–8.5%. Blended margin is now 9.1%, and with AI at 20% of total cloud revenue that implies AI margins of roughly 11.5–13.5%.
Three factors drive Cloud margins higher over the next twelve to eighteen months, and two are already running. First, AI mix shift: as AI moves from 20% to 50%+ of cloud on Wu's guidance, blended margin rises to roughly 10.5% on composition alone. Second: scale economics and pricing power. Revenue growing 40%+ spreads fixed costs, and the 5–34% price hikes effective April 18 flow straight to margin. Customers are queuing despite the increases. These two factors are sufficient to reach the 12% April target.
The final driver is T-Head. As domestically fabbed Zhenwu and Yitian chips replace rented foreign compute, the AI component margin has substantial room to expand above the current 11.5–13.5%. Eddie Wu confirmed that T-Head penetration within Alibaba Cloud is "still low" – the constraint is SMIC fab capacity, not product-market fit. If a price war comes, a large wedge of unrealized T-Head cost savings sits under the margin, so Alibaba can absorb pricing pressure far better than the current P&L suggests.
T-Head is going outward first: Alibaba is selling AI servers to partners and co-building data centers rather than absorbing the chips into AliCloud. The inference here is that this is a commercial track record being built ahead of a probable IPO.
Model Studio – the platform business where developers build on Qwen models via API – has crossed RMB 8 billion in annualized recurring revenue. Eddie Wu said crossing RMB 10 billion this quarter is "a certainty." The target for December 2026: RMB 30 billion. That is 3.75× in six months on the highest-margin component of the cloud business.
Token consumption on BaiLian, the domestic-market name for Model Studio, is up 10× since November–December, an acceleration beyond the 6× disclosed last quarter.
Last quarter, Wu introduced a $100 billion five-year Cloud revenue target during Q&A, and the market largely ignored the implied 35% CAGR. Management has now doubled down on it: the CEO has consolidated control over AI and publicly staked his tenure on this specific milestone.
| Component | April | After Transcript |
|---|---|---|
| Assets (net cash, Ant 33%, AIDC, Cainiao, T-Head) | $160B | $160B |
| China Commerce (8× normalized EBITA) | $210B | $210B |
| Cloud Intelligence Group ($85B rev, 12% margin, 25×) | $250B | $250B |
| $100B target scope | Assumed cloud-only | Confirmed cloud-only |
| 12% margin assumption | Aspirational | Supported by 3 engines (2 active) |
| Fair value per ADS | ~$264 | ~$264 ✓ |
| Current price (May 13) | $124 | $145 |
| Upside to fair value | ~113% | ~82% |
The 12% margin that was a five-year aspiration in April is now two to three quarters out, and possibly materially higher over the five-year horizon. The CAPEX plan that seemed aggressive will be exceeded because they are supply-constrained across every procurement channel.
Customer Management Revenue (CMR) grew +1% on the headline because a new merchant development program moved platform subsidies out of sales and marketing expense and into contra revenue. Like-for-like, CMR grew 8%. Group revenue like-for-like was +11%. 88VIP membership hit 62 million and is still growing double digits. The underlying commerce business is fine.
On Quick Commerce specifically: Jiang Fan confirmed unit economics breakeven by end of FY2027 – roughly 2–3 quarters later than our optimistic scenario, but years ahead of the original FY2029 guidance. He said April onwards showed "significant" unit economics improvement while maintaining volume. The burn has an expiration date – and when it arrives, the consolidated P&L will snap back in free cash flow at precisely the moment cloud margins are hitting their highest levels.
Predicting quarterly revenue conversion from token volume is not where this framework has edge. Three falsifiable milestones for the next two earnings reports:
Milestone 1: Model Studio ARR ≥ RMB 10 billion by the August report (Q1 FY27). If yes, the MaaS thesis is confirmed and the highest-margin component is scaling as guided.
Milestone 2: Cloud external growth ≥ 45% in Q1 FY27. The seasonal rebound plus price hikes plus Model Studio surge should deliver this. If it doesn't, the acceleration thesis needs reassessment.
Milestone 3: Cloud EBITA margin ≥ 10% in Q1 or Q2 FY27. Eddie Wu promised visible margin improvement in 1–2 quarters. We will hold him to it.
All three confirmed: the April base case is conservative. Any one fails: reassess, against a $94 entry and a ~$264 fair value.
The consolidated headline was terrible: EBITA -84%, non-GAAP net income essentially zero, free cash flow negative. The algorithms sold pre-market. Then the humans read the transcript and bought it back 10%.
The demand underneath is not the American kind. China skipped the SaaS revolution: millions of enterprises never had a cloud vendor, and the government is pushing an AI-native industrial base. Alibaba is not fighting to displace entrenched incumbents. It is filling a vacuum – with a physically sold-out product, accelerating growth, pricing power, and a margin structure that is temporarily suppressed by the most expensive possible way of serving current demand.
The April model survives intact: the $100 billion target confirmed as pure cloud, the $85 billion estimate standing, and the 12% margin moved from aspiration to a path with two of three engines already running. And the three milestones above will tell us, within two earnings cycles, whether the thesis is tracking or not.
The market still prices Alibaba as an ex-growth utility while margin expansion and growth acceleration are both in the reported numbers.
Get the next post by email. No spam.
The author holds shares of Alibaba (BABA) at a blended cost basis of approximately $94.
Nothing in this letter is investment advice. Every prediction here is graded on the Prediction Register.
Disclosure: Positions are disclosed live at disclosure.plainsightresearch.com.
A note on first principles. Why the Smil-Rosling synthesis is the analytical engine of Plain Sight Research – and why it keeps producing China.
Sinopec dated the gasoline peak to 2023; the actuals ran faster. The demand did not vanish – it changed meters. Five migrations carry electrification from 30% toward 60%.