On October 20-22, by the Huangpu River in Shanghai, the 2026 Sustainable Global Leaders Conference returns once again. Nobel and Turing Award winners will share the future of technology and trends on-site, and there will even be a dialogue between the internet-famous professors who "rescue" people from anxiety and the era of intense competition—Macfarlane and Xiang Biao! Limited seats, first come first served. Let us join together for the sustainable appointment! [Click to register] For stock trading, look at Jinlin Analyst Reports—authoritative, professional, timely, comprehensive, helping you uncover potential thematic opportunities! Source | Bond Market Observer Author | Liu Yinping. The company that needs money the least is also borrowing now.

On October 12, Tencent Holdings (0700.HK) announced on the Hong Kong Stock Exchange that it plans to rely on its 30 billion USD Global Medium Term Note (GMTN) program to conduct an international offering of notes to certain professional investors and apply for listing on the Stock Exchange. Moody's assigned an A1 senior unsecured rating to the proposed notes. Ten institutions serve as joint global coordinators—four foreign institutions including JPMorgan, Goldman Sachs (Asia), HSBC, and Morgan Stanley, and six Chinese institutions including ICBC Asia, Agricultural Bank of China, Bank of China, Bank of Communications, China Construction Bank, and CITIC Securities. According to Bloomberg, citing people familiar with the matter, the financing scale is about 5 billion USD, structurally divided into two parts: USD bonds and offshore RMB bonds, with maturities covering 10 to 30 years. The proceeds will mainly be used for AI capital expenditures, most of which will become chip procurement orders. Tencent stated, "No binding agreement has been entered into regarding this note offering at present." In other words, the notes have not yet been priced and the issuance has not been completed.

The interesting part is that Tencent is not short of money on its books. According to the 2026 interim report, as of the end of the second quarter, the company held cash and cash equivalents of 206.93 billion yuan, time deposits and other deposits of 304.224 billion yuan, totaling 511.154 billion yuan; after deducting borrowings of 298.942 billion yuan and notes payable of 154.021 billion yuan, the net cash position was still 58.191 billion yuan—it remains a net cash company to this day. And the 5 billion USD it plans to borrow is about 33.5 billion yuan at current exchange rates, equivalent to nearly 60% of its net cash balance on the books.

A company that is not short of money—why is it entering the international bond market at this time?

Financing Fork in the Road for the Three Giants

Tencent's GMTN program was established in 2014 with an initial quota of only 5 billion USD, and has now expanded to 30 billion USD. This shell itself is like a switch: quota ready, issue when the window is suitable. For this 5 billion USD financing, the terms disclosed by people familiar with the matter are 1.5-2 billion USD bonds plus 20-25 billion yuan offshore RMB bonds. In terms of pricing guidance, the 10-year USD bond is 115 basis points higher than comparable US Treasuries, and the 20-year is 140 basis points higher; the initial guidance for the 10-year offshore RMB bond is in the 2.9% area, and the 30-year is in the 3.5% area. Regarding the use of proceeds, foreign media and Tencent have two different accounts. Bloomberg, citing people familiar with the matter, said the proceeds are mainly used for AI capital expenditures, most of which will be used for chip procurement; while the standard wording in Tencent's Hong Kong Stock Exchange announcement is general corporate purposes.

The real highlight is not the 5 billion USD itself, but the timing of its appearance. Over the past few months, China's three internet giants have almost simultaneously, within the same time window, successively used different types of financing tools: Alibaba (9988.HK) finalized an 80 billion HKD new share placement in August and completed signing in early September, with 100% of the net proceeds invested in full-stack AI; ByteDance signed a 29.6 billion USD syndicated loan in August-September, setting the second-largest USD loan in Asia this year; Tencent chose to restart offshore bond issuance in October. The three took completely different channels—one used equity, one used bank credit, one used the public bond market. The difference in tool selection is essentially a difference in capital structure, listing status, and rating system, but the destination is highly consistent: all are using their balance sheets to exchange for a ticket to the AI era.

Image source: Canned Image Library. Some media statistics show that these funds plus other top-tier financing in the same period total nearly 400 billion yuan. This number need not be viewed as an exact account, but it is enough to illustrate one thing: the money for the AI race has grown to a scale that free cash flow cannot carry. Four months ago, in June, Tencent had just completed an offshore financing of about 4.66 billion USD—2.45 billion USD plus 15 billion yuan. That was its first return to the USD bond market since April 2021, after a five-year gap. The subscription heat for that deal exceeded expectations, with total orders exceeding 6 billion USD, and the final pricing tightened significantly from the initial guidance.

It is worth noting the currency structure. In the June financing, the 10-year USD bond coupon was 5%, while the 10-year dim sum bond (offshore RMB bond) was only 2.5%, a difference of about 250 basis points. Same credit entity, same maturity, different currency, and the annual interest difference is two and a half percentage points. So this time, continuing to increase the RMB-side proportion is a clear cost decision, not a gesture.

Now look at debt maturity. Under the GMTN program, Tencent's outstanding balance is about 22.18 billion USD, with no maturity pressure within the year; the nearest 2.5 billion USD will only be repaid in 2028. The explanation that it lacks money is thus untenable. A company truly short of money would not lock in long-term funds for ten or even thirty years when it is still at ease. It is taking advantage of an acceptable interest rate environment and strong investor subscription willingness to pave the way in advance for funding needs in the coming years.

Cash Flow and Debt: The Truth Behind Turning Negative Is Prepayments

What truly explains why Tencent is issuing bonds is hidden in the cash flow statement. In the second quarter of 2026, Tencent's capital expenditure was 52.78 billion yuan, a year-on-year increase of 176%, a record for Tencent's single-quarter capital expenditure. Adding the first quarter's 31.9 billion yuan, the first half totaled 84.7 billion yuan. In the same period, free cash flow was -13.8 billion yuan, the first single-quarter negative in recent years. If you only look at these two numbers, it is easy to conclude that "Tencent has been dragged down by AI." But if you go one layer deeper into the cash flow statement, the picture changes.

The operating cash flow for the quarter included a 51.4 billion yuan prepayment for computing power procurement. This money has already been paid out, but the corresponding computing power and services have not yet been delivered. If it is excluded, operating cash flow for the quarter would be about 104.1 billion yuan, a year-on-year increase of 41%, and free cash flow would turn from -13.8 billion yuan to positive 37.6 billion yuan. Tencent pays the money upfront to lock in future computing power supply. There is a structural mismatch between the payment rhythm of AI infrastructure and the recovery rhythm of business revenue—data centers must be built in advance, chip orders must be locked in advance, and delivery cycles often run for years; while customers pay monthly and based on actual usage. This gap is what the financing is meant to fill.

Image source: Canned Image Library. Pressure is also accumulating on the balance sheet. According to the 2026 interim report, borrowings of 298.942 billion yuan plus notes payable of 154.021 billion yuan bring total interest-bearing debt to about 452.963 billion yuan, an increase of 66.158 billion yuan in a single quarter; net cash in the same period was 58.191 billion yuan, down nearly 88.7 billion yuan from the end of the first quarter, mainly due to large capital expenditure payments and annual dividends. Tencent is still in a net cash state, but the buffer cushion is thinning.

There is another account worthy of attention: capital commitments that have been contracted but not yet recognized are about 280.147 billion yuan. It may not all convert into cash outflows, but more than any current debt, it shows that this round of investment is far from over, and future financing needs will likely continue. Compared with peers, also in the second quarter, Alibaba's capital expenditure was 67.678 billion yuan, a year-on-year increase of 75%, and free cash flow net outflow expanded to 44.67 billion yuan. The direction is completely consistent with Tencent: the old business is still steadily generating cash, the scale of new investment has exceeded current cash flow, and it requires money to arrive earlier than revenue. The difference lies in the choice of financing tools. Tencent is the only one among the three internet giants to enter the public offshore bond market. Alibaba uses equity placement plus previously issued zero-coupon convertible bonds, with net proceeds locked into full-stack AI; unlisted ByteDance relies on bank syndicates, with the 29.6 billion USD loan finalized in August priced at SOFR plus 68 basis points, 17 basis points tighter than its 85 basis points loan in 2024. Different tools, essentially the same thing: all are using their balance sheets to bet on computing power.

Where the Money Burns and Where Returns Come From

Spending the money is only the beginning; whether it can be recovered is the question. Tencent's current round of AI investment has a positioning repeatedly emphasized by management—computing power first serves its own business, rather than being rented out externally. This forms a sharp contrast with Alibaba's "renting computing power to enterprises" and ByteDance's "stuffing computing power into Doubao." Tencent President Liu Chiping defined the capital expenditure of AI-native business at the second-quarter earnings call as a one-time large investment in model training, inference, and AI cloud, and cautioned the market not to assume that such a new investment will occur every year. The subtext of this statement is that investors need not extrapolate this intensity of investment as the norm.

The cost is reflected in two levels of the income statement. At the operating level, affected by investment in new products such as Hunyuan Hy, Yuanbao, WorkBuddy, and CodeBuddy, new AI products dragged down Non-IFRS operating profit by about 10.5 billion yuan in the second quarter, greater than the 8.8 billion yuan in the first quarter—Non-IFRS operating profit for the quarter was 75.636 billion yuan, a year-on-year increase of 9%, with an operating margin of 37%; excluding new AI products, it was 86.1 billion yuan, a year-on-year increase of 19%, with the margin rising to 42%. At the net profit level, profit attributable to equity holders in the second quarter was 56.022 billion yuan, a year-on-year increase of 0.7%, including a share of losses of associates of about 10 billion yuan, from the fair value remeasurement of convertible redeemable preferred shares of an unlisted investee company, which is a non-operating item; after excluding non-recurring items, Non-IFRS net profit attributable to shareholders was 68.415 billion yuan, a year-on-year increase of 9%.

Goldman Sachs, after the second-quarter report, lowered Tencent's 12-month target price from 700 HKD to 670 HKD, citing the group's more aggressive-than-expected pace of AI investment and its priority of using computing power for its own model training and applications rather than renting it out, which puts short-term pressure on earnings per share. Returns are indeed appearing, but they have not yet grown large enough to cover the scale of investment. The most direct example is advertising: second-quarter marketing services revenue was 43.565 billion yuan, a year-on-year increase of 22%, with AI's boost to recommendation efficiency and conversion rates as the main driver. This is also Tencent's relatively unique advantage—its AI does not need to find customers from scratch; WeChat's ecosystem of over a billion users is itself the largest landing scenario.

But the other side of the advantage is the ceiling. Jiang Han, senior researcher at Pangu Think Tank, pointed out: Tencent relies on the WeChat ecosystem to absorb computing power, and the bottleneck lies in the ceiling of monetizing traffic within the ecosystem. WeChat's user reach is unquestionable, but after AI capabilities are embedded in social scenarios, how to complete a commercialization loop without damaging user experience remains an unresolved proposition.

Image source: Canned Image Library. Putting the three together, the differences in tactics are clear at a glance. ByteDance is a heavy-asset landlord, using aggressive capital expenditure to precipitate computing power into its own infrastructure, betting on a super C-end entrance; Alibaba is a cloud-first monetization player, letting enterprise customers directly pay for AI computing power, with B-end cash flow running first. In the second quarter of 2026 (the first quarter of fiscal year 2027), AI-related product revenue has grown at triple-digit rates for 12 consecutive quarters, but in the same period the AI lab and application segment still recorded a large loss; Tencent is an ecosystem-absorption player, relying more on the WeChat ecosystem to absorb computing power and feed back into advertising and products. Bai Wenxi, vice chairman of the China Enterprise Capital Alliance, judges that the essential difference among the three paths lies in who owns the computing power and where the revenue comes from. In his view, Tencent's ratio of capital expenditure to operating cash flow is about 53%, making it the healthiest in cash flow and the most pressure-resistant among the three; but Tencent's bottleneck is that AI distribution within the ecosystem is constrained by privacy and product form, making it difficult to invoke relationship chains without boundaries. From a longer cycle perspective, the core proposition of this competition is not mysterious. In Bai Wenxi's view, whoever can stably produce tokens at large scale at lower cost will hold pricing power in the AI era. When a new batch of domestic computing power is concentrated and put into operation around 2027, supply release may push down inference prices, and then the battle to defend B-end MaaS gross margin will be the key to deciding victory or defeat.

For Tencent, the question is one layer more specific: is it ultimately selling computing power, or selling irreplaceable services? If the answer is the former, then the endgame of this AI competition is a price war; if the latter, then this 5 billion USD bond may generate excess returns. Pressure is still controllable for now. The USD bond window reopening after five years shows that the international market is willing to give it a good price; there is still net cash on the books, free cash flow after excluding prepayments is still positive, and AI has already achieved 22% growth in the advertising business. But controllable does not mean risk-free. The first-half capital expenditure of 84.7 billion yuan, 280.1 billion yuan in unrecognized commitments, plus the yet-to-be-verified pace of AI commercialization together form the financial backdrop for the next two to three years. Bai Wenxi believes the high-intensity investment period will continue for another two to three years, and the industry inflection point may not come until the end of 2028 to 2029. Before that, whether this bond can be proven to be a good deal depends on whether enterprise customers and users are willing to pay for AI. END•