Half of Alibaba's $53 billion AI plan spent in half a year – while profit fell 75 percent
At its Apsara conference in Hangzhou on September 22, 2026, Alibaba lifted the curtain on three coordinated bets at once: a next-generation Qwen model of 5 to 10 trillion parameters, the new Zhenwu V900 chip that CEO Eddie Wu calls China's most powerful AI chip, and a plan to scale Alibaba Cloud to more than 20 gigawatts of data-center capacity by 2032. The stock rose 5.1% in the wake of the announcements. But behind the presentation sits a set of accounts that tells an equally important story: quarterly capital expenditure of RMB67.68 billion ($9.98 billion), net profit down 75%, and roughly half of a planned RMB380 billion (about $53 billion) investment program already spent. The question this week — with a Trump–Xi summit at which AI competition is on the agenda just days away — is not whether Alibaba is betting big. It is whether the revenue will ever catch up with the spending.
What was announced
The three announcements fit together as a full-stack wager: a frontier model competing on capability, a proprietary chip to power it, and data-center capacity to sell both as cloud services.
The model plan is the most concretely quantified. Alibaba said it plans to train a new AI model at a scale of five to ten trillion parameters — a measure of a model's learning capacity. For comparison, the company's current flagship Qwen3.8-Max, the most powerful model in the Qwen series, has 2.4 trillion parameters, according to figures the company gave to the Associated Press. The company described the plan as a step closer to the most advanced US models, but there is no independent verification of the parameter figures — they come from Alibaba itself. It is also worth noting that the naming varies across sources: AP writes "Qwen3.8-Max," while Insider Monkey refers to "Qwen 3.8 Max" and implies a somewhat different baseline for the comparison. AP's figure of 2.4 trillion parameters is the clearest data point, but the new model has yet to receive either a name or a launch date.
The chip, the Zhenwu V900, was presented by CEO Eddie Wu as "the most powerful AI chip in China today," with three times the performance of its predecessor, the Zhenwu M890. Mass production is expected to begin in early 2027 — a company hope, not a documented timeline. And the company said the existing Zhenwu chips are already used by more than 650 external customers in industries including automotive, finance, energy and manufacturing, according to Insider Monkey.
The third element is capacity: Alibaba Cloud is to be expanded to more than 20 gigawatts of data-center capacity by 2032. The company today operates 107 availability zones across 31 regions, with planned expansions in Malaysia, Germany, the United Arab Emirates, France and Hong Kong.
The chip claim, examined
Wu's characterization and the threefold performance figure require two important caveats.
The first is the benchmark. The threefold performance improvement is measured against Alibaba's own previous generation, the M890 chip already in use — not against Nvidia's latest accelerators, as Insider Monkey's analysis (Neha Gupta) points out. That means the figure cannot be read as a statement about competitiveness against leading American hardware. No independent benchmark of the V900 exists in the available record, and comparisons with Nvidia's newest generation should therefore not be drawn from this.
The second is that "China's most powerful AI chip" is Wu's own characterization of his own product, not a verified result. It is not unusual for executives to use such formulations at their own events, but they should be reported as exactly that — a company description.
What is more concrete, by contrast, is the customer base. More than 650 external customers on the existing Zhenwu chips, if the figure holds, show that Alibaba's chip strategy is no longer purely internal — it is already a commercial business with customers across multiple industries.
The bill so far
The bet carries a price that is now visible in the income statement.
In the June quarter, Alibaba spent RMB67.68 billion — $9.98 billion — on capital expenditure, an increase of 75% from a year earlier, driven primarily by AI infrastructure. In the same quarter, net profit fell 75%, according to Insider Monkey's rendering of Reuters reporting and the company's results. It should be stressed that all of these financial figures were relayed through the financial press and company results in secondary sources, not verified directly against Alibaba's own earnings release.
More broadly, Reuters reported that Alibaba had already spent roughly half of the planned RMB380 billion (about $53 billion) AI investment for 2026–29 by the first half of 2026. That half of a three-year plan has been spent in half a year suggests either that the pace is running ahead of schedule or that the total bill will grow — both are interpretations, but the figure alone sets the standard for how aggressive the rollout is.
The company itself claims the AI investments should break even within three years. That is a company statement, not a guarantee, and it depends entirely on the revenue side continuing to compound as it did in the June quarter.
The revenue side — and what it does not cover
Here the picture is mixed. AI cloud and computing revenue rose 45% year over year to RMB48.44 billion, beating analysts' estimates. The cloud segment's adjusted EBITA rose fully 133% to RMB5.94 billion. That is real, rapid growth in precisely the part of the business the bet is meant to pay for.
But total revenue of $39.64 billion — up nine percent year over year — missed consensus, and earnings per share of $1.26 also missed estimates, according to Barchart. In other words: the growth in AI cloud is strong enough to beat segment estimates, but not strong enough to carry the whole group's results when investment costs are deducted.
The most optimistic external assessment comes from Citi. Analyst Alicia Yap estimates that AI cloud revenue could reach $168 billion in fiscal year 2033 — but conditional on the 20-gigawatt plan succeeding. That is a condition that makes the whole equation circular: the revenue forecast assumes the investment plan is realized, while the investment plan's financing partly assumes the revenue materializes.
Global footprint and regulatory risk
The 20-gigawatt target is measured against an existing base of 107 availability zones in 31 regions, with expansions planned in Malaysia, Germany, the United Arab Emirates, France and Hong Kong. The choice of countries shows Alibaba positioning itself as a global cloud provider, not merely a Chinese one — including in Europe and the Middle East, markets where American hyperscalers dominate.
But the chronology of the share price illustrates the risk embedded in China's model sector. After the 5.1% rise connected with the announcements, the stock fell four percent on September 23, after Bloomberg reported that Chinese authorities have opened an investigation into the startups DeepSeek and Moonshot AI over data-security concerns. Alibaba was not named in the investigation, and the company's own statements on it are not available. But the report hit shares in Chinese model developers generally, showing how regulatory risk in the model layer — where Alibaba's Qwen competes — can sweep onto the stock exchange regardless of the company's own messaging. The two share movements appear to be consecutive trading days, but the exact chronology between the sources is not fully established.
The timing is also worth noting: the announcements came days before the Trump–Xi summit, at which US–China AI competition is a major theme.
The open questions
Several things must fall into place before it is possible to judge whether the bet pays off.
First: independent benchmarks. The parameter counts, the chip performance and the "China's most powerful" claim all come from Alibaba. Until third-party measurements exist, the technical story is in practice the company's own narrative.
Second: the primary sources for the financials. The capex, the profit decline, the EBITA growth and the revenue figures were all relayed through the financial press. They are probably correct, but precise reporting requires verification against the company's own earnings release, and the chronology of the share price movements could have been sketched more sharply.
Third, and most decisive: whether cloud growth compounds fast enough. The arithmetic is easy to sketch. AI cloud revenue grew 45%, while the cost side grew 75% with net profit down 75%. The company's three-year break-even promise requires segment growth to accelerate or at least remain stable while capital expenditure flattens. Citi's $168 billion scenario for 2033 shows the market can see a path there — but it is explicitly conditional on 20 gigawatts being built out as planned, in global markets where competitors have deep experience and regulatory advantages.
The Apsara announcements commit Alibaba to an integrated whole that few companies in the world are attempting: its own model, its own chip, its own global infrastructure. The numbers behind them show a company spending as if it must win. What winning costs — and whether the payoff materializes — remains an open question.