Today's annual Apsara Conference has become one of the most closely watched tech events of the year. Alibaba is clearly taking it seriously, as AI now stands as the company's top strategic priority. CEO Wu Yongming personally appeared and delivered an extensive address focused on AI, effectively transforming what was once the Alibaba Cloud conference into an AI-centric event.
Wu shared substantial insights during his speech. He projected that the total volume of machine reasoning could eventually exceed human reasoning by more than 1,000 times. Alibaba is currently training new models with 5 trillion to 10 trillion parameters, and T-Head has unveiled a next-generation AI chip named Zhenwu V900, whose performance is three times that of its predecessor, the M890.
From my perspective, the truly noteworthy revelation in this speech was Wu's announcement that by 2032, the total scale of data centers operated by Alibaba Cloud worldwide will surpass 20GW. This signals that Alibaba is poised for a massive leap in AI investment over the next six years, with total funding likely to easily exceed one trillion yuan. Given that Alibaba's free cash flow has been negative recently, this implies significant cash consumption ahead, requiring continued large-scale financing. For someone like me, who isn't used to such massive numbers, it's a scenario that makes me nervous.
Let's dive into the details without further ado.
Why does Alibaba need to sustain this AI expansion for six years?
The 20GW target indeed represents a substantial leap. For those who may not grasp the scale of this number, allow me to clarify. Alibaba has not disclosed its precise global data center capacity, but Goldman Sachs previously estimated that Alibaba's live data center capacity is approximately 3-4GW, up from around 2GW in 2022. This 3-4GW figure refers to IT capacity that is built and operational, not peak computing power. If Goldman Sachs' estimate is roughly accurate, moving from 3-4GW to over 20GW by 2032 implies an additional 16-17GW over the next six years, expanding the data center footprint by more than fivefold.
This pace of expansion is notably aggressive. For context, Bloomberg has reported that Microsoft's current global data center capacity is around 12GW, with plans to increase it to approximately 38GW by 2032. Even if Alibaba achieves its 20GW target, its absolute scale would remain smaller than Microsoft's, but it would still place Alibaba among the top-tier players globally.
What would this scale of expansion cost? Jensen Huang of Nvidia has estimated that building a 1GW AI data center currently requires approximately $50-60 billion, with around $35 billion allocated to AI computing infrastructure provided by Nvidia. In June, Huang even suggested that the investment for a future 1GW AI factory could rise to $80-100 billion. Of course, Nvidia's chips are expensive, and a greater reliance on domestic chips could lower these costs. UBS previously calculated that if 1GW of new data center demand were entirely dedicated to AI, just the capital expenditure on IT equipment alone would amount to roughly 100 billion yuan. This 100 billion yuan figure primarily represents IT equipment spending, not the complete total cost of building a data center.
Even with conservative estimates, this means Alibaba's investment over the next six years could reach an astronomical 1.6-1.7 trillion yuan. However, there are variables to consider. The price and performance of AI chips will evolve, and domestic chips, particularly those from T-Head, might become cheaper. Additionally, the 20GW target refers to data centers "operated" by Alibaba Cloud, which doesn't necessarily mean Alibaba funds all construction itself, as it could seek partners. Wu emphasized today that Alibaba will invest in AI infrastructure "together with all partners." Even with significant discounts applied to UBS's model, one point remains difficult to change: expanding from 3-4GW to 20GW represents a trillion-yuan-scale infrastructure undertaking.
Can Alibaba sustain spending at this level?
The impact of large-scale AI investment on Alibaba's cash flow is already apparent. Over the past five quarters, Alibaba's free cash flow has been negative in four of them. In the most recent quarter, capital expenditure reached 67.7 billion yuan, resulting in a free cash flow outflow of 44.7 billion yuan. If Alibaba continues to add 2-3GW of computing capacity each year, the pace of free cash flow outflows may accelerate further.
Alibaba's cash position remains substantial, with about 474.5 billion yuan in cash and other liquid investments as of the end of June. Last month, the company also raised approximately HK$80 billion through a new share placement in Hong Kong, explicitly stating the proceeds would fund its full-stack AI capabilities, including AI infrastructure. Yet even with these resources, Alibaba's cash reserves may prove insufficient to cover what is conservatively a trillion-yuan computing investment over the long term. Consequently, Alibaba will likely need to continue its financing efforts.
The company's determination to make this massive bet on AI appears driven by its assessment that cloud revenue growth will outpace the growth in AI investment. Wu reinforced today that AI demand is currently extremely robust, with medium- to long-term demand even exceeding existing supply capacity. If AI-related cloud revenue maintains rapid growth and the data centers built today achieve sufficiently high utilization rates, then even a cumulative investment on the trillion-yuan scale might not represent an unwise venture.
However, the challenge is that AI infrastructure is not a one-time build. AI chips have depreciation cycles, servers require periodic upgrades, next-generation chips continue to emerge, and the computing power needed for model training and inference is also expanding rapidly. The 20GW target entails not only substantial initial construction investment but also ongoing equipment upgrades and maintenance thereafter. Regardless of the outcome, this will be an extremely costly undertaking, even for Alibaba. Whether Alibaba can succeed in this gamble remains to be seen.
Comments