HomeArticle

Cash flow never lies: AI has forced Chinese tech giants into the same test.

Tech商业2026-11-20 15:29
Heavy asset investment in AI is reshaping the cash flow landscape of China's tech giants.

On November 19, the second trading day after Tencent released its financial results, its share price fell by nearly 6%, touching HK$350.4 during the session, hitting a new low since August 2024. What triggered the sell-off was not revenue — the quarterly revenue of RMB 207.3 billion still grew by 8% — but a figure further down the income statement: net cash from operating activities, which stood at RMB 2.6 billion. In the same period last year, this figure was RMB 85.3 billion.

A day earlier, Tencent CFO Lo Siu Hon explained on the conference call that this RMB 2.6 billion included a large advance payment for computing power procurement, which was used to lock in GPUs to support the upgrade of the Hunyuan Model and the inference demand of WorkBuddy. Excluding this advance payment, free cash flow would turn positive from -RMB 69.1 billion to RMB 16.9 billion. President Martin Lau tried to calm analysts with a more plain analogy: it is like building a factory, which can be used to train the next generation of models after completion, and can also produce continuously.

But the market took in another sentence. James Mitchell, Chief Strategy Officer, said that calculated based on the actual capital expenditure payment of about RMB 65 billion in the third quarter, Tencent's investment level has "roughly reached its peak" and will remain around this level in each quarter of 2027. This means that negative free cash flow is not a one-off pulse, but a new normal that will last for several years.

The same night, Alibaba released its results for the quarter ended September 30. Operating cash flow was RMB 15.69 billion, up 55% year on year, which looked much more comfortable than Tencent's. However, capital expenditure of RMB 50.477 billion simultaneously ate up this hematopoietic capacity, resulting in a net outflow of free cash flow of RMB 34.472 billion, further expanding from RMB 21.84 billion in the same period last year. Earlier, Baidu's Q3 operating cash flow was RMB 2.8 billion, capital expenditure was RMB 7.35 billion, and free cash flow was -RMB 4.57 billion. The three companies differ in scale, pace and investment structure, but their cash flow statements point to the same fact: AI is systematically rewriting the operating cash flow of Chinese tech giants from an "amplifier of profits" to a "financing source for capital expenditure".

Tencent's RMB 86 Billion and Alibaba's RMB 3800 Billion:

Two Kinds of Heavy Asset Rhythms, the Same Cash Flow Gravity

When you unpack Tencent's quarterly report, the most informative part is not the RMB 81.4 billion capital expenditure itself, but where it is recorded. The RMB 65.1 billion capital expenditure payment goes to cash flow from investing activities, while the approximately RMB 86 billion advance payment for computing power procurement is included in cash flow from operating activities. This means that Tencent is not "spending money to buy assets", but "paying in advance for computing power supply in the coming quarters". The long-term item of "advances, deposits and other assets" surged from RMB 24.5 billion at the beginning of the year to RMB 162.9 billion. This sum of money does not lock in current income, but the priority claim on the supply chain.

This is a procurement posture with anxiety. Tencent's capital expenditure in the first quarter of this year was RMB 31.9 billion, RMB 52.8 billion in the second quarter, and RMB 81.4 billion in the third quarter, with the single-quarter investment nearly 2.5 times higher within half a year. It is not building computing power at its own pace, but catching up with the admission ticket that Alibaba and ByteDance have already taken seats at the poker table. The cost is the flip of the balance sheet structure: at the end of last quarter, Tencent still had net cash of RMB 58.2 billion, but at the end of the third quarter, it turned to net debt of RMB 13.7 billion. In October, Tencent issued about US$5.7 billion in notes; in the four months before that, it had raised nearly US$4.7 billion through US dollar and RMB notes. Tencent, which once generated a steady stream of free cash flow from games and social networks and used it to support huge share repurchases, is learning to use debt instruments to finance capital expenditure.

Alibaba's pace is different, but the gravity is the same. Its capital expenditure is not "pulsed advance payment", but infrastructure precipitation that has lasted for several years. The three-year RMB 3800 billion AI and cloud infrastructure plan announced in February 2025 has invested about RMB 1900 billion by the end of June 2026, more than half of the progress. Q2 capital expenditure was RMB 67.678 billion, free cash flow was -RMB 44.67 billion; Q3 capital expenditure was RMB 50.477 billion, free cash flow was -RMB 34.472 billion. The management has made it clear that the three-year RMB 3800 billion plan will "far exceed" the original amount. In August, Alibaba completed a HK$80 billion rights issue, 60% of the net proceeds will be used to expand global computing infrastructure, and 40% will be used for AI data centers and supporting facilities. CFO Xu Hong's statement is restrained and sober: focus on improving return on invested capital, and "commit to realizing positive free cash flow as soon as possible". The word "as soon as possible" itself admits that turning positive is not an immediate matter.

Baidu is the lightweight sample among the three. The quarterly capital expenditure of RMB 7.35 billion doubled year on year but fell 35.5% month on month, and the free cash flow gap was -RMB 4.57 billion, which is far smaller than Tencent and Alibaba in absolute scale. The management explained that it has already established sufficient computing power reserves, and a considerable part of the incremental demand can be met by existing production capacity. Baidu is the only one of the three companies that shows "restraint" in the pace of capital expenditure, but this is not strategic ease, but more the supply substitution expectation brought by volume constraints and self-developed chips (Kunlun Core M100 is expected to be mass-produced by the end of the year). In the arms race of AI computing power, not following means being eliminated, and following means continuous blood loss on the cash flow statement.

From "Cash Cow plus Repurchase" to "Borrowing plus Cash Drainage"

Over the past decade, the financial narratives of large Chinese internet platforms have been highly consistent: light assets, high cash flow conversion rate, and large share repurchases. Advertising, games, and e-commerce commissions have marginal costs close to zero, operating cash flow has long been higher than net profit, and free cash flow is the fuel for shareholder returns. AI infrastructure has cut off this chain from the middle.

Bloomberg estimates that the capital expenditure of leading domestic internet companies in 2026 will approach US$66 billion, a 77% increase from the historical high in 2025, and the funds will mainly flow to large model training, GPU procurement and intelligent computing centers. Data from Morgan Stanley provides a colder reference: the capital expenditure per server in China is about 3 times that of similar configurations in the United States, and the depreciation cost is significantly higher. The ROIC of Chinese cloud vendors is about 13%, while that of the United States is 31%. This means that for the same computing power scale, Chinese tech giants need to pay heavier capital costs, and the recovery efficiency is lower.

Therefore, financing behavior itself has become the hidden line of the earnings season. Tencent issued bonds, Alibaba did a rights issue, ByteDance signed a US$29.6 billion syndicated loan, and the four large financings totaled about US$56 billion. None of these companies are short of cash on their books — Alibaba still had RMB 487.9 billion in cash and other liquid investments as of the end of September — but they chose not to "wait for the business to recover payment slowly", but to fully prepare for the investment in the next few years at one time while the capital window is still open. A former financial head of a listed company gave a very straightforward breakdown to Securities Times: ByteDance is not listed, and syndicated loans are an option to retain control; Alibaba's debt scale is already high, and continuing to issue bonds will push up the premium, so rights issue is a realistic choice; Tencent still has unused bond quotas, so it prioritizes debt instruments to avoid dilution.

The choice of financing instruments is a tactic, but the common strategic judgment points to: locking the supply of AI computing power is more important than the cost of capital. Once the contract is signed and the advance payment is made, the money cannot be recovered if the demand judgment is wrong. ByteDance, as the underwriter of the capacity of multiple data centers in Southeast Asia, used binding contracts to buy the supply in advance, at the cost that it is irrevocable. Tencent's RMB 86 billion advance payment is essentially doing the same thing, except that it occurs in the domestic computing power supply chain.

The Depreciation Bill Has Not Been Sent Yet

The current pressure on the income statement is only the first wave of the impact of capital expenditure. Tencent's Q3 depreciation and amortization was RMB 13.7 billion, up 56% year on year, but this mainly reflects the result of the gradual transfer to fixed assets of earlier rounds of capital expenditure. The current capital expenditure of RMB 81.4 billion, plus the approximately RMB 280.1 billion contracted but not accrued capital commitments disclosed in the interim report, will enter the depreciation and amortization sequence in the coming quarters. The suppression of capital expenditure on profits does not happen "at the moment of payment", but after the assets are delivered, launched and transferred to fixed assets, it continuously and lagging erodes the operating profit margin in the form of depreciation.

This is the real weight of Mitchell's sentence that "the current level will be maintained in 2027 and improvement will come in 2028". The premise of improvement is not that capital expenditure declines, but that the revenue growth rate generated by computing power assets exceeds the depreciation growth rate. The revenue growth rate of Alibaba Cloud's external commercialization has increased to 67%, and AI-related product revenue has achieved three-digit growth for the 13th consecutive quarter. The revenue of AI cloud and computing power services in Q3 was RMB 60.973 billion, up 53% year on year. Revenue is accelerating, but free cash flow is still expanding outflow, indicating that the scale of revenue feedback has not yet caught up with the volume of capital precipitation. Goldman Sachs judges that Alibaba will still face a funding gap in fiscal 2027 to 2028, and it is expected to achieve capital balance driven by the release of EBITDA of the cloud business until fiscal 2029. For Tencent's turning point, the caliber given by the management itself is 2028.

What the market is currently pricing is exactly the visibility of this cash flow gap period. Analyst Catherine Lim's comment after the earnings release of Tencent and Alibaba has been repeatedly cited: the deep-seated concern of investors is whether the company can prove that AI is bringing measurable revenue growth through cloud services, advertising or transaction conversion, and the market will remain cautious before that. Morgan Stanley cut Tencent's target price by nearly 13% to HK$480, and CICC lowered the Non-IFRS net profit forecasts for 2026 and 2027 by 2% and 7%, both on the grounds that "higher AI investment will suppress short-term earnings". The sell side does not deny the AI direction, but refuses to pay in advance for a cash flow gap with no visible recovery curve.

Cash Flow Speaks Before Profit

The most memorable thing in this earnings season is not how many points the profit of a certain company has fallen. Profits can be smoothed by investment income, modified by changes in fair value, and redefined by "core operating profit excluding new AI products" — Tencent did exactly that. Excluding new AI products such as Yuanbao and CodeBuddy, the core operating profit was RMB 83.3 billion, up 8% year on year, and the profit margin remained stable at 40%.

But operating cash flow has no room for such rhetoric. RMB 2.6 billion is RMB 2.6 billion, and -RMB 69.1 billion is -RMB 69.1 billion. Every dollar in and out corresponds to a real change in the bank account. When Tencent's operating cash flow collapsed from RMB 85.3 billion to RMB 2.6 billion, when Alibaba's free cash flow gap expanded from RMB 21.8 billion to RMB 34.5 billion, and when Baidu's free cash flow remained negative even with a month-on-month decline in capital expenditure, the three companies said the same sentence with three different statements: the first bill of AI re-assetization is not sent to the income statement, but to the cash flow statement.

Martin Lau compared AI investment to building a factory, and this analogy is accurate. However, the depreciation life, capacity utilization rate and product price of the factory are far more difficult to predict in the field of AI computing power than in traditional manufacturing. The technical iteration cycle of GPUs is calculated in years, and the assets are highly specific. Once the demand growth rate slows down, impairment and depreciation will come at the same time. The ROIC of Chinese cloud vendors is already nearly 18 percentage points lower than that of the United States, and the cash payback period is about 3 years, while the rhythm of technological iteration may not wait for these 3 years.

Tencent, Alibaba and Baidu are all betting on the same thing: before the depreciation bills mature intensively, computing power can be converted into revenue, and revenue can cover the cost of capital. The bet has been placed, the financing has been received, and the advance payment has been remitted. The cash flow statements in the next few quarters will record the progress of this gamble one by one. The income statement can lag, depreciation can be apportioned, and the narrative can be adjusted — but cash flow does not lie.

This article is from the WeChat official account "Tech Business", written by Tech Business, and published with authorization by 36Kr.