# Alibaba’s AI Empire Has No Checkout > Source: > Published: 2026-08-03 09:10:08+00:00 # Alibaba’s AI Empire Has No Checkout ### Equity gains, cloud orders and hardware costs sit in separate accounts, leaving investors unable to measure the portfolio’s net return. Ten days in July compressed Alibaba’s AI portfolio into a single frame. On July 17, Moonshot AI, one of China’s best-funded AI laboratories, released Kimi K3, its 2.8-trillion-parameter model. Two days later, on July 19, Alibaba [launched Qwen3.8-Max Preview](https://www.alibabagroup.com/document-2016703577908576256), the latest in its Qwen model family, targeting the same coding-agent and enterprise-API market. That evening, Moonshot [paused new consumer subscriptions](https://www.reuters.com/legal/transactional/chinas-moonshot-pauses-kimi-subscriptions-amid-hot-demand-ipo-push-2026-07-20/) after demand pushed close to its compute ceiling within 48 hours of K3’s launch. Eight days later, ChangXin Memory Technologies (CXMT), China’s leading DRAM manufacturer, [opened trading in Shanghai](https://www.sse.com.cn/listing/renewal/ipo/qsmzq/index_listing_detail.shtml?uniformCode=91340100MA2MWUT60Q). CXMT closed at Rmb 49, up 466% from an issue price of Rmb 8.66. Alibaba holds [2.995bn shares](https://static.sse.com.cn/stock/disclosure/announcement/c/202605/002170_20260527_23QQ.pdf) through two subsidiaries. At the closing price, those shares were worth approximately Rmb 147bn. Against a total investment that Chinese financial media put at roughly Rmb 7.6bn, the paper gain approached Rmb 139bn. Across that sequence, Alibaba occupied four positions simultaneously: shareholder in the model company that ran short of compute, cloud provider whose infrastructure might serve the expansion, competitor releasing a rival model into the same market, and investor in the chipmaker whose debut sharply marked up Alibaba’s CXMT holding. In the US, analysts have questioned whether supplier capital finances the demand later booked as cloud revenue; regulators have [focused on lock-in and competitive control](https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-issues-staff-report-ai-partnerships-investments-study). The Alibaba-Moonshot arrangement is the clearest Chinese structural counterpart, and in some respects less transparent than the American cases. [As I reported recently](https://hellochinatech.com/p/moonshot-kimi-k3-ipo-valuation), Moonshot had moved in seven months from saying it was in no rush to list to circulating a [shareholder resolution for a Hong Kong IPO](https://finance.yahoo.com/technology/ai/articles/china-moonshot-talks-pre-ipo-142358163.html). K3 sat between those two decisions. Parts of the potential loop already exist: capital can support a model company, and that company can buy Alibaba Cloud services. What is missing is a portfolio-level account that reconciles those cloud orders with equity gains, infrastructure costs and the expense of competing through Qwen. Public reporting leaves the value in three separate ledgers. **When Every Investment Had a Destination** Alibaba once ran a model where every investment had a destination inside the organizational chart. In 2009, it invested in UC, a mobile browser company. By 2014, it owned UC outright and folded it into a mobile division. In 2016, it entered [Ele.me](https://ele.me/), a food delivery platform. Two years later, it [bought the remaining shares](https://www.sec.gov/Archives/edgar/data/1577552/000110465918021637/a18-9422_1ex99d1.htm) at an implied enterprise value of $9.5bn and pulled the company into its local services group. Traffic arrived, transactions were processed, value was settled. The platform was the register. The AI portfolio has no such register. Alibaba holds [approximately 36% of Moonshot’s preferred equity](https://www1.hkexnews.hk/listedco/listconews/sehk/2024/0523/2024052301569.pdf), a figure dated to March 2024 and almost certainly diluted by subsequent rounds. If the full round closes, Alibaba would hold 1.11% of Kling, the video-generation unit [Kuaishou, China’s second-largest short-video platform, carved out at an $18bn post-money valuation](https://hellochinatech.com/p/kling-kuaishou-ai-carve-out) while retaining at least 68% on a fully diluted basis, with more than 30 independent investors and strategic participants including Tencent sharing the capital burden. It held [12.52% of MiniMax](https://www1.hkexnews.hk/listedco/listconews/sehk/2025/1231/2025123100025.pdf), a Chinese AI model company that listed in Hong Kong in early 2026, after the global offering, excluding shares acquired through its cornerstone subscription. It holds 4.48% of CXMT. The full portfolio is larger and less precise. Estimates from Chinese financial media range from at least 24 companies, per **Securities Times**, a state-backed financial daily, to 29 per **36Kr**, with approximately Rmb 36bn in total deployed capital. No figure comes from Alibaba’s own audited disclosures. None of these companies has been absorbed into Alibaba’s organizational boundary. Moonshot runs Kimi independently. Kling reports to Kuaishou. MiniMax published its own IPO prospectus. The model companies among them are potential Alibaba Cloud customers and direct Qwen competitors. CXMT sits upstream: its share price affects Alibaba’s equity ledger, while the broader DRAM pricing cycle affects cloud costs. The value from all these positions sits in three separate ledgers. What follows traces each one, and asks why public reporting provides no way to add them up. *If the idea that Alibaba’s AI gains, cloud orders and hardware costs land in three separate financial ledgers is new to you, this is a preview of what Hello China Tech does three times a week: reading China’s AI, chip, robotics, and EV sectors from primary sources most English coverage never touches. Subscribe free to get every new analysis as it publishes.*