Alibaba Bets Big on AI Infrastructure as Cloud Revenue Jumps

The Infrastructure Price of AI Ambition

Alibaba's June quarter financial results laid bare a tension familiar to anyone tracking Asia's cloud wars: building AI infrastructure at scale requires burning cash faster than revenue can catch up. The company reported RMB 269 billion in revenue for the three months ending June 30, representing 8.6% year-on-year expansion. That figure stands in sharp contrast to the 2.9% growth posted in the prior quarter, signaling a return to momentum after months of sluggish performance.

The cloud computing division drove much of that acceleration. AI-related offerings within the unit recorded triple-digit percentage growth, continuing a streak that began in late 2025 when Alibaba pivoted its cloud strategy toward inference workloads and model-training services. At DailyTechWire, we've tracked similar patterns across Tencent Cloud and Huawei Cloud over the past year, all three jockeying for enterprise customers migrating legacy workloads to generative AI stacks.

Yet the tailwinds came with a sobering counterbalance. Alibaba's capital expenditure more than doubled quarter-on-quarter, reflecting outlays for GPU clusters, data center expansions, and networking gear capable of handling the bandwidth demands of large language model training. Adjusted earnings declined, and free cash flow remained negative, underscoring the front-loaded cost structure of competing in China's AI infrastructure race.

Where the Money Is Going

The bulk of Alibaba's capex surge is flowing into compute capacity. Industry data suggests Chinese cloud providers collectively ordered tens of thousands of Nvidia A800 and H800 chips in the first half of 2026, stockpiling inventory ahead of potential tightening in export regulations. Alibaba has also invested in domestic alternatives, including chips from Biren Technology and Moore Threads, though these remain less power-efficient for training workloads above 100 billion parameters.

Data center footprint is another major line item. Alibaba operates more than 80 availability zones across Asia-Pacific, and the company has broken ground on at least three new facilities in Inner Mongolia and Guizhou since January, targeting regions with lower electricity costs and favorable provincial incentives for AI infrastructure. Cooling systems account for a growing share of construction budgets; liquid cooling deployments have tripled year-on-year as rack densities climb past 30 kilowatts per cabinet.

Networking upgrades round out the spending picture. Moving training data between nodes at the scale required for models with trillions of parameters demands low-latency fabrics. Alibaba has deployed InfiniBand and proprietary optical switches in its largest clusters, reducing communication overhead but adding material costs that won't translate to revenue for several quarters.

Revenue Growth Driven by Model Hosting and API Calls

The 8.6% overall revenue increase masks sharper gains within specific cloud product lines. Model-as-a-service offerings, which let enterprises fine-tune and deploy large language models without managing infrastructure, saw adoption jump among e-commerce platforms, fintech apps, and customer service operations. Alibaba Cloud's Qwen family of models, available via API, recorded a 140% rise in billable inference calls compared to the March quarter.

Enterprise customers are also spending more on vector databases, embedding search, and retrieval-augmented generation pipelines. These ancillary services carry higher gross margins than raw compute, helping offset some of the pressure from GPU costs. Storage revenue tied to training datasets and model checkpoints grew double-digits, as organizations archive multiple versions of fine-tuned models for compliance and rollback purposes.

International expansion contributed modestly. Alibaba Cloud added points of presence in Jakarta and Bangkok during the quarter, targeting Southeast Asian developers building local-language AI applications. Revenue from markets outside Greater China remains a low-teens percentage of the cloud division's total, but growth rates in those geographies outpaced the domestic business.

The Profitability Puzzle

Adjusted earnings fell despite the revenue pickup, a function of gross margin compression in the cloud unit. Training and inference workloads generate lower margins than traditional enterprise software-as-a-service, and Alibaba has discounted pricing to win deals away from Tencent and Baidu. Promotional credits for startups and research institutions further diluted profitability.

Free cash flow stayed negative for the third consecutive quarter, weighed down by the capex ramp and working capital tied up in chip inventory. Alibaba's cash conversion cycle lengthened as suppliers demanded faster payment terms for scarce GPU shipments, while the company extended longer payment windows to large enterprise customers as part of contract negotiations.

The funding rounds we've followed across the region suggest this dynamic is not unique to Alibaba. Tencent's cloud unit reported similar margin trends in its May earnings, and Huawei's enterprise business group flagged elevated infrastructure spending in its semi-annual report. The common thread: securing GPU supply and data center capacity today is a prerequisite for revenue growth in 2027 and beyond, even if it punishes near-term cash flow.

What Comes Next

Alibaba's trajectory over the next four quarters will hinge on two variables. First, whether AI product revenue can scale quickly enough to absorb the fixed costs of the infrastructure buildout. Model hosting and API-based inference carry better unit economics than training services, so a shift in customer mix toward inference-heavy workloads would improve margins. Second, whether domestic chip alternatives mature to the point where Alibaba can reduce reliance on imported GPUs, lowering both capital intensity and supply chain risk.

Regulatory tailwinds may also play a role. Beijing's push to centralize AI compute resources in state-backed "computing power networks" could funnel public-sector demand toward the largest cloud providers, including Alibaba. Conversely, any move to cap data center energy consumption in key provinces would force the company to slow expansion or relocate facilities to less economically attractive regions.

For now, Alibaba's June quarter underscores a broader reality in Asia's AI infrastructure landscape: growth and profitability are moving in opposite directions, and the companies willing to endure the deepest cash burn today are betting they'll emerge with the largest customer bases and the most defensible moats when the cycle turns.