Patrick T. Fallon

Alibaba (9988.HK) raised HK$80 billion through a share placement in August. Tencent Holdings (0700.HK) issued US$2.45 billion and 15 billion yuan (about HK$16.2 billion) in notes under its global medium-term note programme during the same month. ByteDance, meanwhile, signed a US$29.6 billion syndicated loan with 28 financial institutions led by Industrial and Commercial Bank of China and HSBC Holdings. All three companies are holding substantial cash reserves. As of the end of June, Alibaba had about 474.5 billion yuan (about HK$512.5 billion) in cash and other liquid investments, while Tencent held 511.2 billion yuan (about HK$552.1 billion) in cash.

Borrowing aggressively despite having ample cash may appear illogical, but the strategy reflects a carefully calculated financial plan.

A highly synchronised borrowing drive

First, it is worth clarifying the origins and details of the three financings. One point needs correcting at the outset: the use of Alibaba’s HK$80 billion differs from some descriptions circulating in the market.

Alibaba’s HK$80 billion was not a conventional share buyback or capital-structure optimisation exercise. It was the company’s first new share placement since its Hong Kong listing in 2019, involving 710 million new shares sold to professional and institutional investors outside the United States at HK$112.70 each. According to a Hong Kong Exchange filing, net proceeds after commissions and expenses amounted to about HK$79.7 billion, with 100% to be used to invest in full-stack AI capabilities and strengthen AI infrastructure. The proceeds are therefore directly tied to AI rather than shareholder returns. Although the placement was more than 10 times oversubscribed within an hour, the secondary-market reaction was muted. Alibaba’s US-listed ADRs fell 8.6% in a single session after the announcement, while its Hong Kong shares at one point dropped more than 10% after the market opened. Investors were concerned about dilution, while some analysts noted that the size of the placement was close to the 80 billion to 100 billion yuan Alibaba had invested in instant retail, food-delivery subsidies and related businesses over the previous 18 months, raising questions about whether the company was simply plugging one gap with another.

Tencent’s move is also noteworthy, although the original description requires correction. The 15 billion yuan was not a domestic syndicated loan. It was an offshore renminbi bond, or dim sum bond, issued alongside the US$2.45 billion in notes under Tencent’s global medium-term note programme. It was Tencent’s second dim sum bond issue and the largest such issue since the start of 2025. Of the bonds, 4 billion yuan had a 30-year maturity and a 3.1% coupon. The dollar notes had combined maturities of 10 and 20 years. The company said in an official announcement that net proceeds would be used for “general corporate purposes, including refinancing”, without explicitly stating that all the funds would go towards AI. Tencent president Martin Lau Chi-ping said after the issue that strong demand reflected investors’ recognition of the company’s progress in AI efficiency platforms. The market has broadly interpreted the fundraising as part of preparations for AI-related capital spending.

ByteDance’s financing was the largest and has made the most recent progress. The syndicated loan was formally signed in the middle of this month, with 28 banks taking part. Chinese lenders provided the largest share: ICBC committed US$3 billion, Bank of China US$2.5 billion and China Construction Bank US$1.5 billion. In total, 15 mainland Chinese banks committed US$18.9 billion, or 64% of the facility. HSBC was one of the largest foreign-bank lenders, providing US$1.5 billion. The loan carries an initial spread of 68 basis points over SOFR, lower than the 85-basis-point spread on ByteDance’s previous US$10.8 billion offshore loan in 2024. The official purpose is likewise “general corporate purposes”, but the market broadly sees the borrowing as linked to ByteDance’s consideration of raising capital expenditure this year to as much as US$70 billion to expand data centres and AI infrastructure.

The three financings were completed within a six-month period. This was no coincidence, but reflected similar capital-management choices by mainland China’s leading internet companies during the same strategic cycle.

Why borrow when they have cash?

The rationale can be summed up in three concepts: maturity matching, cash preservation and interest-rate locking.

Matching long-term debt with long-term assets

AI infrastructure is a typical capital-intensive, long-cycle investment. From project approval to power-up, deployment and stable operation, a data centre generally takes two to three years to build. If short-term internal funds are invested in one go, the balance sheet becomes less resilient and flexible. A slowdown in business or deterioration in the macroeconomic environment could leave the company with little room to absorb the shock.

By contrast, long-term notes such as a 30-year bond can better match the expected payback period of AI infrastructure. The borrowed funds can be repaid over a longer period, while the recurring cash flow generated by data centres can gradually cover the cost of the debt. This is the financial principle of maturity matching: aligning the term of liabilities as closely as possible with the return cycle of the assets.

Not putting all the eggs in one basket

The 474.5 billion yuan and 511.2 billion yuan in cash held by Alibaba and Tencent are nominally available, but in practice they have designated purposes, including day-to-day operations, potential acquisitions, share buybacks and strategic reserves for unexpected industry shocks.

Putting all that cash into AI expansion would amount to staking the companies’ overall strategic flexibility on a single long-cycle project. If progress falls short of expectations, other business lines could immediately become constrained. A more pragmatic approach is to borrow for investment while keeping internal cash for the core business. Even if the payback period for AI investment is extended, the companies would still have sufficient internal cash flow to maintain daily operations.

Stockpiling funds while costs are low

In the first half of 2026, global liquidity conditions were relatively loose. For large technology companies with high credit ratings, effective borrowing costs remained relatively low. Ample interbank liquidity and intense competition in the syndicated-loan market allowed major companies to secure large sums on favourable terms. ByteDance’s loan, whose spread was lower than that of a comparable facility two years earlier, is one example.

For Alibaba, Tencent and ByteDance, this represents a rare window to raise funds at low cost. If they wait until the AI race is in full swing and demand for capital has surged, their bargaining power could fall sharply, and they could even find themselves unable to borrow on acceptable terms despite having cash.

Computing power is becoming a money-making machine

For several years, AI was viewed as a cost centre: companies poured money into data centres, bought GPU chips and hired research teams, with little immediate return. That narrative is beginning to change, but the details must be recalibrated against actual figures rather than accepted wholesale.

Alibaba’s public figures offer early evidence of the shift. As of the end of June, annualised revenue from its AI-related products was about 49.5 billion yuan, with the segment growing 45% year on year, its strongest growth in 22 quarters. Alibaba Group chief executive Eddie Wu Yongming said clearly during an earnings call that the return on investment in AI computing capital expenditure was “highly certain” and could be recouped within three years. This is a verifiable statement with a specific timetable, more precise than the broad claim of “two to three years”.

As for the profitability of Tencent Cloud’s computing business, the claim that its computing-rental and resale operation had achieved a profit margin of more than 30% cannot be substantiated by Tencent’s public financial reports or public comments from management. Market reports instead suggest that Tencent could recently have secured a premium of more than 30% by reselling prepaid computing orders, but chose to forgo the one-off arbitrage opportunity to prioritise computing capacity for its in-house models, including Hunyuan, and internal applications such as WeChat AI and WorkBuddy. Rental is a “safety net” rather than the core business. In other words, Tencent’s current strategy is “self-use first, rental second”, not an established computing-rental and resale business generating stable margins of more than 30%. The narrative of monetising computing power is supported by solid data in Alibaba’s case, but the evidence remains insufficient for Tencent.

Once a modern data centre is operational, it can serve internal business needs, including search, advertising recommendations and large-model inference, while excess capacity can be sold to third-party companies and research institutions through public cloud platforms. This dual-track model, centred on internal use and supplemented by external sales, theoretically gives AI infrastructure two sources of revenue. The extent to which companies have put it into practice varies considerably, with Alibaba clearly further ahead in commercialisation.

Three indicators to watch

  • Changes in the share of AI revenue: If revenue from AI-related products and services at Alibaba, Tencent and ByteDance continues to grow significantly faster than overall revenue, it would show that the external commercialisation of computing assets is being validated in practice rather than remaining at the planning stage.
  • Free cash flow coverage of new debt: Investors should watch whether the three companies’ operating cash flow over the next 12 to 24 months can consistently cover principal and interest payments on the new debt. Tencent’s capital expenditure in the second quarter rose 65% quarter on quarter to 52.8 billion yuan, equivalent to about 25.8% of quarterly revenue and approaching Amazon’s level during the same period. This is a leading indicator worth monitoring.
  • ByteDance’s actual capital-expenditure rollout: If its final spending on hardware purchases and data-centre construction far exceeds the US$70 billion expected by the market, it would indicate that the industry’s AI arms race is entering a more aggressive phase.

Together, these indicators point to an increasingly clear shift in the business model: AI infrastructure is moving from pure consumption expenditure to a valuable asset, while technology giants are using leverage to secure key resources and lock in prices ahead of time.