Alibaba's AI Capex Is the Point, Not the Problem

Background

Alibaba (BABA) priced a primary placement of 710 million new Hong Kong shares at HK$112.70, raising HK$80B (roughly $10.2B), the largest equity raise in the company's Hong Kong history, with all proceeds directed to AI infrastructure, in-house chips, and the Qwen model family. The raise follows a June quarter (fiscal 2026) in which revenue grew 9% year over year to RMB 268.95B ($39.6B) while capex jumped 75% to RMB 67.7B and net profit fell sharply. The market's first reaction was an 8-10% selloff; the deal book told a different story.

Key Findings

The Oversubscription Is the Signal, the Selloff Is the Noise

The prevailing read on the raise was that a discounted equity sale signals weakness: the placement came at an 8.4% discount to the prior Hong Kong close (about 3.6% versus the US-listed line), and the stock dropped 8-10% on announcement. But the deal was nearly 3x oversubscribed, with sovereign wealth funds and long-only institutions crowding in. That contrast, a 10% knee-jerk drop against a 3x book, captures the entire debate in one data point. Short-horizon holders priced dilution; long-horizon capital paid for the capex plan. Dilution of roughly 3.6-3.7% of the enlarged share count is real but modest against $10.2B of fully AI-directed capital. When institutions are begging for exposure, a modest discount is cheap funding, not distress.

Equity Funding Is a Feature of Alibaba's Position, Not a Bug

US hyperscalers fund AI capex from operating cash flow and debt because they have to preserve equity multiples that already price in the AI story. Alibaba reached for equity instead, and the framing matters: this is not plugging a balance sheet hole. It is raising cheap capital at a moment of high institutional demand so the buildout does not starve the core commerce business of cash. With capex already up 75% year over year to RMB 67.7B in a single quarter, funding the acceleration externally protects the operating engine while the AI segment scales. The playbook matches Microsoft and Amazon in ambition; the financing route reflects Alibaba's cheaper currency and deeper discount to intrinsic value.

AI Demand Is Proven, Not Speculative

The skeptical case treats Alibaba's AI spend as chasing a trend without evidence of monetization. The numbers say otherwise. AI cloud and compute revenue grew 45% year over year to RMB 48.4B in the June quarter. Model-as-a-service has passed RMB 16B in annualized recurring revenue. AI product revenue has grown at triple-digit rates for ten consecutive quarters, a streak that predates the current capex surge. Alibaba is not building capacity and hoping demand arrives. Demand is already outrunning supply, which is precisely the condition under which aggressive capex is the correct capital allocation decision.

The Stack Is Deeper Than Any US Hyperscaler's, Including Google's

The assumption that Alibaba is a follower running the US hyperscaler playbook at a lag misses the structural difference. Alibaba designs its own accelerators and CPUs, operates the cloud those chips run in, builds the frontier Qwen model family on that infrastructure, and owns the consumer surfaces (Taobao, DingTalk, Quark) where inference deploys on day one. Even Google, the most vertically integrated US player, does not own commerce-scale consumer distribution for its models in its largest addressable markets. Two external forces compound the advantage. First, Qwen is open-weight and has become the default model family across China and much of the global open-source ecosystem, a distribution moat that closed-model US players structurally cannot replicate. Second, US export controls, intended as a constraint, have turned Alibaba's in-house chip effort from a side bet into a captive advantage: it is the only credible hyperscaler serving the world's second-largest AI market, where US clouds cannot operate.

The Cost of Conviction Is Real and Must Be Named

The honest tension in the thesis sits in the AI Labs and Applications segment: RMB 3.34B in revenue against an adjusted EBITA loss that widened to RMB 13.86B, driven by inference costs tied to the Qwen consumer app. Net income attributable to shareholders fell roughly 75% year over year to around RMB 10.5B. This is not a thesis-breaker, consumer AI is a land-grab phase across every major player, but it is the variable that determines whether the current raise is the last one needed or the first of several. Equity funding is dilutive by definition; repeated dilution would change the math.

Implications

Alibaba

The raise buys Alibaba time and capacity to press its full-stack advantage while the core business stays fully funded. The strategic burden now shifts from proving demand, which the 45% cloud growth and ten-quarter AI streak have done, to proving unit economics on consumer inference. If Qwen-driven losses converge toward breakeven as inference costs fall and monetization ramps, Alibaba converts a discounted equity raise into one of the cheapest AI buildouts any hyperscaler has executed.

US Hyperscalers (Microsoft, Amazon, Google)

None of the three can contest the Chinese market, which means Alibaba's home turf is uncontested by the only companies with comparable stacks. Google faces the sharpest comparison: Alibaba matches its chip-cloud-model integration and exceeds it on consumer distribution and open-weight reach. For Microsoft and Amazon, the implication is competitive rather than direct: Qwen's position as the default open-weight family erodes the assumption that frontier capability requires a closed US model, particularly in markets where cost and sovereignty matter.

Investors

The valuation asymmetry is the actionable point. Investors get the same AI-buildout growth story, 45% cloud growth, triple-digit AI product growth, aggressive capex into proven demand, at a fraction of the multiple paid for Microsoft or Amazon. The 3x oversubscribed book shows institutional capital has already made this call. The monitoring variable is specific: whether AI Labs and Applications losses narrow before Alibaba needs to return to the equity market.

Conclusion

Alibaba's $10.2B raise and 75% capex surge are not symptoms of stress; they are the visible cost of a full-stack AI strategy that no US hyperscaler can replicate in Alibaba's markets. The findings collectively dismantle the bearish read: demand is proven by 45% cloud growth and ten quarters of triple-digit AI product growth, the equity funding is cheap capital raised into a 3x oversubscribed book, and the vertical integration from chips to consumer apps is deeper than even Google's. The genuine risk is not the buildout but the pace at which Qwen-driven inference losses converge, and that is a monitoring question, not a thesis question. Structurally, Alibaba is the only hyperscaler with a captive claim on the world's second-largest AI market, and the market is pricing that position at a discount the deal book suggests will not last.