--- title: "Alibaba is betting US$52 billion on AI infrastructure — more than its last decade combined" date: 2026-10-02 category: Capital & Markets site: NeuroAI canonical: https://neuroai.site/a/na-capital-alibaba-380b-capex language: en --- # Alibaba is betting US$52 billion on AI infrastructure — more than its last decade combined > In February 2025 Alibaba committed at least ¥380 billion (≈ US$52.4B) to cloud and AI hardware over three years, and later said it would spend even more. The cheque is bigger than the company's entire AI and cloud spend of the previous ten years. For a firm better known for e-commerce, the message was unmistakable: the next battle is compute. On February 24, 2025, Alibaba said it would invest **at least ¥380 billion** (≈ **US$52.4 billion**, at the Reuters rate of 7.247 RMB/USD) in cloud computing and AI hardware infrastructure over the following three years. The figure, reported by Reuters and Xinhua, exceeds Alibaba's total AI-and-cloud spending across the prior decade. ## Why the number matters This is not a vague pledge. It is a capped floor — "at least" — tied to a specific window. A few readings follow directly: - It signals that Chinese hyperscalers now treat GPU clusters and data centers as the strategic asset of the 2020s, akin to how ports and factories were in an earlier era. - It sets a spending baseline that rivals (Tencent, ByteDance, Huawei Cloud) are implicitly measured against. - It arrived days after a strong December-quarter print — revenue of about **RMB 280.15 billion** (≈ **US$38.7 billion**) — that lifted Alibaba's Hong Kong shares roughly 14% in a single session. ## Early signs it is already spending By May 2025, Alibaba told Reuters it expected to **exceed** the ¥380 billion plan, citing early commercial returns from AI demand. The cloud business was the proof point: - Cloud Intelligence Group revenue rose **38%** to about **¥41.63 billion** (≈ **US$5.8 billion**) in the quarter to March 2025. - AI-related products accounted for about **30%** of external cloud-customer revenue, and management said it expected that share to pass **50%** within roughly a year. - CEO Eddie Wu framed margins as "secondary" to gaining AI-driven market share — a candid admission that growth is being bought, not yet harvested. ## The open-model engine behind the spend Alibaba's bet is not only on hardware. Its Qwen (通义千问) family is among the most-downloaded open-weight model series from China. Xinhua noted that since August 2023, Alibaba Cloud has open-sourced four generations — Qwen, Qwen1.5, Qwen2 and Qwen2.5 — spawning **more than 90,000 derivative models**. Open weights pull developers toward Alibaba's cloud to fine-tune and serve them, turning model generosity into infrastructure demand. ## The risk on the other side Building capacity ahead of proven, durable demand is a classic hyperscaler gamble. If enterprise AI adoption in China lags the build-out, Alibaba absorbs depreciation and idle GPUs. Wu's "margin is secondary" line tells you the company accepts near-term pressure on profit to secure position — a reasonable strategy only if the AI revenue ramp is real and sticky. ## How to compare this number honestly Capital-commitment headlines invite bad comparisons, so a few calibration points help. First, this is a floor, not a budget: "at least" means the real number will be set by demand, and Alibaba has already said as much. Second, the spend is concentrated in a single business line whose revenue (about RMB 41.63 billion in the March 2025 quarter) is roughly one-seventh of the group's total — meaning Alibaba is deliberately letting its most capital-hungry unit run ahead of its cash generation. Third, the commitment spans three years, so the annual run-rate is closer to RMB 127 billion (≈ US$17.5 billion) than the headline number suggests — large, but comparable to what US hyperscalers spend in a single quarter of peak build-out. Reading the floor, the window, and the run-rate together gives a fairer picture than any one figure alone. ## What it means for the wider stack The commitment also lands on China's domestic compute supply chain. A spend of this size cannot be filled by imported accelerators alone given export controls, so it functions as a demand guarantee for domestic chips, servers and optical interconnects — the same pull-through that Chinese AI-chip makers have been counting on. If you track the semiconductor side, Alibaba's capex line is one of the clearest leading indicators of whether domestic accelerator capacity gets funded at scale or stalls at pilot volumes. ## Honest limitations - The ¥380 billion figure and the "exceeds past decade" claim come from Reuters and Xinhua; we did not see the original company filing, only the reported statement. - The May 2025 "will exceed" comment and the 30%/50% cloud-AI revenue shares are from a Reuters earnings report; exact definitions of "AI-related revenue" are not public. - RMB-to-USD conversions use the Reuters rate of 7.247; different days yield slightly different dollar amounts. - This describes a capital allocation decision, not an investment recommendation. ## What readers can do now - Track Alibaba Cloud's quarterly "AI-related revenue" share — management itself flagged 30%→50% as the watch metric. - Compare this ¥380B, three-year commitment against peers' disclosed capex to gauge the sector's aggregate build-out and any over-supply risk. - If you use open models, test Qwen derivatives on Alibaba's cloud free tier to see whether the open-weight-and-cloud flywheel actually lowers your serving cost. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-capital-alibaba-380b-capex Free to quote with attribution and a link to the original.