ByteDance, Tencent, Alibaba and Baidu take on heavy debt to bet on AI: The survival multiple-choice question for large tech companies

When AI competition evolves from a simple contest of algorithmic creativity into a capital-intensive, asset-heavy melee over computing power infrastructure, leading tech giants that hold national-level cash cows have collectively launched intensive capital operations.

Hundreds-of-billion-dollar syndicated loans have been approved, large-scale bonds are issued one after another, equity placement and financing are carried out, and the existing investment portfolio is realized in batches. All these moves point to the reserve of AI computing power.

On the one hand, revenue scale continues to expand and the profitability of core businesses remains strong; on the other hand, profit growth is narrowing, free cash flow turns negative in a single quarter, and historical large-scale borrowing is witnessed. Behind this seemingly contradictory business landscape lies the fundamental reversal of the growth paradigm of the entire internet industry.

This article will focus on analyzing

1. Why have tech giants collectively fallen into the business paradox of being highly profitable yet still short of capital?
2. What strategic judgments and trade-offs of risks are implied behind different financing paths?
3. What risks are hidden in the off-balance-sheet leverage and computing power race of global tech giants?
4. Is the high-stakes bet of exchanging debt for computing power a valid ticket to the future or just a bubble?

Over the past decade, if you asked a believer in the classic internet era what the most attractive business model for a tech company is, he would definitely tell you it is asset-light, high-margin, and exponential growth.

From writing the first line of code to running servers smoothly in the cloud, the internet leveraged extremely large-scale wealth with an ultra-low marginal cost. In that era, "asset-heavy" was a derogatory term belonging to traditional manufacturing and real estate. The balance sheets of tech giants were clean and lightweight, and their cash flow was so abundant that all traditional industries envied them.

But when the singularity of general artificial intelligence looms, the rules of the game at the table have been brutally rewritten.

The Information latest disclosed data shows that in the first half of 2026, ByteDance's revenue was about 1200 billion US dollars, up 30% year on year; its net profit was about 200 billion US dollars, down by a single-digit percentage year on year, due to increased investment in AI. In the same period, Alibaba's revenue reached 753 billion US dollars and Tencent's 590 billion US dollars, totaling 1343 billion US dollars. ByteDance's revenue alone is already close to the sum of Tencent and Alibaba. At the same time, ByteDance borrowed 296 billion US dollars from banks, the largest single loan in the company's history.

The largest company, the most powerful cash generator, is growing at a rate of 30% on the one hand, part of its profit is eaten up by AI on the other hand, and it is still borrowing heavily.

This is not the situation of ByteDance alone. In the past few months, Tencent, Alibaba and Baidu have each completed a round of large-scale financing. AI is no longer a mere code fantasy, but an asset-heavy melee that competes for infrastructure construction.

In this battle, the profits and equities of the old era are all being drawn and cast into the steel and concrete of the computing power foundation of the new era.

What we need to clarify is the capital expenditure logic of the giants, the underlying rules of competition, and the hidden risk boundaries behind prosperity under the reconstruction of AI.

Part.1

The Shift to Asset-heavy Operation

Cash Flow Generation Cannot Keep Up With Computing Power Consumption

After the arrival of the AI wave, the asset-light model of the internet industry has been rewritten. In the past, enterprises could rely on endogenous cash flow generated from operations to cover the expansion needs of most businesses. However, the core investments in the AI track, including large model training, inference services, GPU procurement, and data center construction, all belong to rigid asset-heavy inputs.

Different from the flexible marketing expenses and elastic R&D investment in the internet era, once the computing power hardware is purchased, it will enter a fixed depreciation period of five to seven years, which will not disappear with business contraction. There is no once-and-for-all solution for model iteration: every version update and every user call will continuously consume computing power resources.

This continuous capital expenditure with the nature of an arms race fundamentally breaks the asset-light growth logic of the internet industry.

The consequent result is widespread cash flow gaps. Even if the profitability of core businesses is still considerable, the disposable cash retained by enterprises from each period of operation is still difficult to match the pace of computing power expansion. The profits of mature businesses need to take into account shareholder dividends, daily operation maintenance and iteration of existing businesses, and there is a natural ceiling for re-investment in new tracks. Therefore, obtaining funds from the external market has become a common choice for large tech companies.

The 29.6 billion US dollar syndicated loan that ByteDance has just finalized is the second largest US dollar-denominated financing in Asia this year. The initial plan was to borrow 20 billion US dollars, but it was finally expanded because the bank's subscription orders exceeded 30 billion US dollars. The spread was compressed to SOFR plus 68 basis points, 17 basis points lower than the overseas loan in 2024. This is almost the cheapest price for overseas borrowing by Chinese tech companies.

Alibaba took another path. The HK$80 billion new share placement completed at the end of August was its first new share placement since it was listed in Hong Kong in 2019, and it was oversubscribed in less than an hour after launch. On the day of the placement, Alibaba's share price fell by 8.54%. Shortly after, Joseph Tsai and Wu Yongming successively increased their holdings by about HK$202 million, and Jack Ma himself also increased his holdings by more than HK$600 million, using real money to reassure the market.

Tencent's moves are more representative. In June, it issued US$24.5 billion in dollar bonds plus 15 billion yuan in dim sum bonds, totaling nearly US$47 billion, the largest bond issuance since 2020.

Capital expenditure in the second quarter soared to 52.784 billion yuan, up 176% year on year, which was mainly invested in AI computing power. As a result, free cash flow in the single quarter was negative 13.8 billion yuan, which was also the first time Tencent turned negative since its listing.

A company with operating cash flow exceeding 300 billion yuan and net profit exceeding 250 billion yuan in 2025 actually burned out free cash flow to negative seasonally, which is a very striking signal. Of course, this gap mainly comes from the concentrated prepayment for computing power procurement in the current period, which belongs to upfront capital investment rather than operating loss.

What can better illustrate the problem is that Tencent quietly sold about 17 billion yuan of listed company equity in the past three months. In July, it reduced its holdings in Kuaishou to cash out about HK$12 billion, in August it sold its shares in the South Korean game company Netmarble to recover 1.8 billion yuan, in September it reduced its holdings in Bilibili to get 400 million US dollars, and then it cashed out another HK$2.2 billion from Zhaopin. Referring to the capital expenditure intensity of the second quarter, this 17 billion yuan is only enough for Tencent to last for one month.

Baidu converted its secondary listing status on the Hong Kong Stock Exchange to primary listing, and was included in the Stock Connect on September 7, opening up the southbound capital channel for high R&D investment in AI and autonomous driving.

Among the four companies, the one most worthy of separate discussion is actually ByteDance. Although the four companies have different financing tools, ByteDance's case reveals a deeper fact: in the AI infrastructure track with extremely high capital intensity, even the company with the best cash flow in China needs the leverage of external capital to maintain competitive intensity.

According to data from The Information, ByteDance's full-year revenue in 2025 was about 200 billion US dollars, up 29% year on year; in the first half of 2026, it continued to grow at a rate of 30%, while in the same period, the revenue growth rates of Tencent, Alibaba and Pinduoduo all dropped to around 10%.

The problem arises: these companies that hold the most profitable businesses in China, especially those like ByteDance, which still maintain a 30% growth rate as a global cash generator, why have they suddenly become short of money together?

The answer is actually not complicated: the speed of spending money in the AI battle has exceeded the cash flow that the core business can generate in the current period.

According to Bloomberg reports, ByteDance is considering raising its 2026 capital expenditure to a maximum of 700 billion US dollars, more than double that of last year, and may rush to 1 trillion US dollars by 2027. Alibaba's three-year plan of 380 billion yuan in AI infrastructure investment has spent about 190 billion yuan by the end of June 2026, reaching half of the progress.

This is not a one-off investment project, but a continuous, large-scale and endless capital expenditure.

Old businesses such as Douyin, Taobao, WeChat and Baidu Search are still making profits, but even if all their quarterly profits are poured into AI, it is still not enough, so the capital market has become the only supply line.

Part.2

Common Multiple-choice Questions for Survival

Loosening of Old Moats and Seizure of the Window Period of the Era

Taking a broader view, the four companies have various financing methods but highly consistent goals, and behind the choice of each tool are implied different strategic judgments and risk preferences.

ByteDance chose syndicated loans, which have the advantages of fast speed, large volume and no equity dilution, at the cost of bearing rigid interest burden. More importantly, ByteDance is not listed, and there is no public market share price as the pricing anchor for equity financing, so syndicated loans are its most effective large-scale financing channel.

Alibaba chose equity placement, at the cost of diluting old shareholders, but in exchange for long-term funds that do not need to be repaid. Alibaba wants to build full-stack AI, from chips to cloud infrastructure, to large models and to the application layer, which requires long-term, large-scale and low-cost funds, and Alibaba is willing to exchange dilution for time.

Tencent chose the combination of bond issuance and disposal of old assets, which is between the two choices. The total equity assets of Tencent in the second quarter report were 875.1 billion yuan, of which the equity of listed companies was 487.2 billion yuan, which is an ammunition depot that can be realized at any time. But only domestic business holdings such as Kuaishou and Bilibili can be disposed of; unlisted parts such as WeBank, Epic Games, Moonshot AI and DeepSeek cannot be realized in the short term, and four newly listed AI-related companies still have lock-up periods.

Baidu's approach is more clever. After being included in the Stock Connect, southbound funds can directly buy its shares. Third-party forecasts show that southbound net inflows may reach 60 billion US dollars in the next 2 to 4 months, which is equivalent to opening a new liquidity channel in the existing market.

The four approaches represent four financial philosophies, but the underlying logic is only one sentence: in the AI computing power arms race, get the admission ticket in hand first.

From a global perspective, this trend is even more striking. Data from the London Stock Exchange Group shows that tech giants including Alphabet (Google's parent company), Amazon and Meta have issued nearly 2.2 trillion US dollars of bonds since 2026, more than double the total of 2025.

Among them, Alphabet's cash flow from operating activities in the second quarter was about 39.1 billion US dollars, and capital expenditure reached about 44.9 billion US dollars, resulting in free cash flow of about negative 5.9 billion US dollars in the quarter, which was the first time Alphabet's quarterly free cash flow turned negative since its listing. At the same time, it issued 250 billion US dollars of bonds in August, and the subscription orders reached 1.15 trillion US dollars.

In addition, data from Bloomberg shows that AMD has just completed the largest US dollar bond issuance in the company's history, raising 47.5 billion US dollars at one time. And Nvidia issued 250 billion US dollars of bonds in June, with orders exceeding 850 billion US dollars.

There is a deeper hidden worry here. Studies by Goldman Sachs and Morgan Stanley point out that major AI cloud giants have accumulated nearly 2 trillion US dollars in off-balance-sheet financial commitments through lease agreements and procurement commitments that have not yet started.

Goldman Sachs estimates that off-balance-sheet lease commitments are about 1.5 trillion US dollars, of which 1 trillion has not yet been included in the balance sheet; Morgan Stanley estimates that Alphabet, Microsoft, Amazon, Nvidia and Oracle's procurement commitments on chips, equipment and power total 982 billion US dollars.

The typical case is that Meta and Blue Owl jointly established Beignet to develop the Hyperion data center in Louisiana. Meta only holds 20% of the shares, but promises to lease it for at least 20 years. Based on this, Beignet issued a record-breaking 270 billion US dollar amortizing bond. Meta actually bears most of the repayment obligations, but this huge debt does not need to be directly recorded on its own balance sheet.

Back to Chinese large tech companies, the logic is actually the same. In a track that everyone thinks must be fully committed, any company that falls behind may never catch up. Therefore, even if cash flow is under pressure, even if shareholders are diluted, even if interest is borne, they must reserve sufficient ammunition first.

The sentence from Wang Hua of Sinovation Ventures is quoted repeatedly for a reason: "If you lose this battle, at most you will burn tens of billions in vain; but if you don't fight, it may be a matter of life and death."

Part.3

The Two Sides of the Game

Dual Tests of Commercialization and Capital Cost

But is this high-stakes bet really a sure win? Not necessarily.

Judging from traditional credit indicators, the books of these companies are indeed still very healthy. Data from Morgan Stanley shows that the average net leverage ratio of major cloud giants is only 0.5 times, far lower than the 0.8 times average of the tech industry and the 1.8 times average of US non-financial enterprises, and their cash holdings even exceed the book debt.

However, there are several hidden dangers in this apparent health that are worth analyzing.

Purchased computing power needs to be depreciated, built data centers need to be amortized, and borrowed loans need to pay interest. These are all rigid costs. No one can give a definite answer now whether AI revenue can outpace these costs within the depreciation period.

Although the interest rate of ByteDance's loan is low, it is also a floating rate; Tencent's dollar bonds and Alibaba's exchangeable bonds are not free lunches. When AI investment enters a long cycle and the financing scale keeps increasing round by round, the interest burden will change from negligible to a considerable proportion.

Tencent's free cash flow turned negative in the second quarter, which essentially means that the climbing of capital expenditure has not yet reached the peak. The future trend of this curve depends on whether AI revenue can climb up synchronously. And there is a bottom line for the pace of disposing old assets: after Tencent sells all the holdings of Kuaishou, Bilibili and Zhaopin that can be sold, the remaining assets either involve strategic synergy and cannot be easily disposed of, or involve overseas assets that will affect the overall situation if one part is touched.

The internet investment portfolio of the old era is being split up to supply blood for the AI war of the new era. This logic is cost-effective for shareholders in the short term. After all, the valuation premium of old assets has been continuously compressed in the past few years. But at the same time, it also means that the patient capital Tencent can squander in the future is limited.

To be frank, none of the AI businesses themselves have achieved a real second growth curve in commercialization.

Alibaba Cloud's external commercial revenue growth rate reached 40%, the proportion of AI-related product revenue exceeded 30% for the first time, with quarterly