Author: Huashang Strategy
Several of the richest internet companies in China have recently started to seek funding externally.
Alibaba is issuing shares, Tencent is issuing bonds, and ByteDance has secured a $29.6 billion syndicated loan from nearly 30 banks.
But the question is, do they really lack money? At least on paper, it does not seem so.
As of the end of June this year, Alibaba had 474.5 billion yuan in cash and other liquid investments, while Tencent's official "total cash" stands at 511.2 billion yuan. ByteDance, without publicly available financial statements, is still one of the most profitable internet companies in China.
With so much money already, why continue to seek more?
The answer is simple, because of AI.
The latest to join this financing wave is ByteDance.
On September 3rd, media reports stated that ByteDance originally only planned to borrow $20 billion. After the news was released, nearly 30 banks from China, the United States, Europe, and Singapore participated, with subscriptions exceeding $30 billion. ByteDance then expanded the loan scale to $29.6 billion, approximately 200 billion yuan, becoming the second-largest dollar loan transaction in Asia this year.
What’s particularly special is that such a large sum of money is unsecured.
A person directly involved in the transaction stated that such large-scale unsecured loans are very rare. Simply put, the banks are willing to lend the money based on ByteDance's own credit.
In 2024, ByteDance had also borrowed $10.8 billion from about 20 domestic and foreign banks and other lending institutions. Now, the loan scale has nearly tripled compared to two years ago, with better financing conditions.
What is the money for? ByteDance explained it as "general corporate purposes," but according to media reports, it will mainly support the company's AI expansion.
Alibaba is taking a different path.
In late August this year, Alibaba conducted a new share placement in Hong Kong, raising 80 billion Hong Kong dollars, and stated very directly that all this money will be invested in AI, primarily for expanding computing power, building large AI data centers, and upgrading cloud infrastructure.
Tencent went to the bond market. In June this year, Tencent issued bonds worth $2.45 billion and 15 billion yuan, part of which is a yuan bond that will be repaid in 2056, with a maturity of 30 years. Tencent did not specify that this money is exclusively for AI, but at the same time, its investment in AI infrastructure is also clearly accelerating.
The big companies seeking money are not limited to China.
As of July 7 this year, Amazon, Alphabet, Meta, and Oracle have issued about $194 billion in bonds this year, nearly 80% more than the $108 billion for the entire year of 2025.
Among them, Meta issued $25 billion in bonds all at once at the end of April. Oracle plans to raise $45 to $50 billion this year through bonds and equity to fund its ever-expanding cloud and AI infrastructure investments.
From China to the United States, more and more tech giants are starting to actively seek external financing. It’s not because they lack money, but because AI is too capital-intensive.
If you go to Guangling County in Datong, Shanxi, and look at ByteDance's computing power infrastructure, you may gain a more intuitive understanding of the "AI is costly" issue.
There, ByteDance's Volcano Cloud has already established a large-scale Taihang computing power center. The second-phase project alone has reached a total investment of 4.5 billion yuan, planning to have over 15,000 server cabinets, with a 220 kV power transformation project built alongside it.
▲Source: Guangling County Government website
The AI we usually see are chatbots and videos generated in seconds, but when it truly lands, it is a series of server rooms, rows of servers, and a massive set of supporting infrastructure.
Although chips often receive the most attention, acquiring chips is just the first step. The real challenge is how to turn those chips into a stable computing power operation.
Since this year, the sales of NVIDIA's H200 to China have undergone repeated approval and delivery processes. Even if permission is granted, it doesn’t mean the chips can be immediately obtained. Meanwhile, data centers also take time for construction, equipment deployment, and ultimately powering up.
This also means that investments in AI infrastructure are not only vast in amount but must also be planned and continually invested in advance.
As the demand for computing power continues to grow, the capital expenditure of large companies is also rapidly expanding.
At the beginning of 2025, Alibaba announced it would invest at least 380 billion yuan in building AI and cloud computing infrastructure over the next three years, having already spent about 67.7 billion yuan in the second quarter of this year.
ByteDance is even more aggressive. According to media reports, ByteDance previously discussed increasing its 2026 capital expenditure to a maximum of $70 billion, equivalent to about 470 billion yuan, with a focus still on data centers and other AI infrastructure. This number may ultimately be adjusted, but it sufficiently indicates the scale of this competition.
This sense of urgency can also be seen in Zhang Yiming.
When he announced his resignation as CEO of ByteDance in 2021, Zhang Yiming expressed in an internal letter a hope to "take a ten-year period" to spend more time learning, systematic thinking, and researching new things.
By July this year, he rarely spoke out at an internal meeting of the Seed team, clearly stating not to rely on distilling competitor models for short-term rankings, and expressed a willingness to sacrifice some short-term profits for long-term goals.
In terms of model development, Zhang Yiming still emphasizes long-termism, but in terms of computing power and infrastructure investment, ByteDance is clearly accelerating.
The reason is not complex; if your models lag today, they can still iterate after six months. But if others are already running large amounts of GPUs in their data center while you are still in construction and waiting for power, it becomes very difficult to catch up in that timeframe.
Moreover, in the AI era, infrastructure is no longer just the support behind the business; it itself is becoming part of the business.
How capable models can be, how many users they can serve, and whether costs can be reduced are all directly constrained by computing power. To some extent, where the infrastructure is built sets the upper limit of the business.
Therefore, while AI competition may last for many years, the race for infrastructure positions is concentrated in the next few years. What the big companies are truly competing for is not just chips, but also time.
But this brings up another question: since they have hundreds of billions in cash on their balance sheets, why don’t they just spend their own money?
Because a company's money never goes to just one thing. Share buybacks, acquisitions, and developing new businesses all require funds, and space must be left for future risks and opportunities that may arise.
Infrastructure like servers and data centers will generate value for many years. Using long-term funds to support long-term investments is, in itself, a very natural choice.
Tencent is a very intuitive example. While it hasn't specifically tied the bond funds to AI, long-term funds clearly leave more room for future large-scale investments.
Of course, the big companies dare to spend this way because another important prerequisite has been met. This accounting is already beginning to make sense.
At Tencent's Q2 financial report meeting this year, an analyst asked a very direct question: If the capital expenditure in the second quarter was about 53 billion yuan, which annualizes to over 200 billion yuan, how will future depreciation and amortization affect profits? How long until the new income from AI can cover these costs?
Tencent's Chief Strategy Officer James Mitchell's answer reveals another layer of logic behind the company's capital expenditures.
According to him, with the current demand for computing power and rental prices, if Tencent rents out its new computing power to third parties like some emerging cloud companies, it could almost immediately cover equipment depreciation and quickly achieve good returns.
Tencent's president Liu Chiping then added another calculation: Some computing power that was pre-paid and ordered months ago can even yield more than 30% profit if turned around now.
This essentially provides a "safety cushion" for Tencent's AI capital expenditures: if its own AI business grows fast, it will use the computing power itself; if external demand is stronger, it can also be monetized through cloud services.
Computing power is transitioning from a mere cost to a productive asset that can generate income.
Alibaba has also started to calculate a similar equation. In August this year, Alibaba Group CEO Wu Yongming stated at the earnings conference that based on the current average gross profit level of AI products, related capital expenditures can be recouped in about three years, and with the improvement of gross margins and operational efficiency, it could be shortened to 2.5 years, or even around two years.
He also mentioned that the A100 purchased by Alibaba in 2020 and even the V100 purchased in 2018 are still operating close to full capacity, and their actual useful life far exceeds the theoretical depreciation cycle.
Meanwhile, income has also begun to catch up. Revenue from AI-related products at Alibaba has achieved triple-digit growth for the twelfth consecutive quarter, with annual revenue exceeding 49.5 billion yuan. In the second quarter of this year, Kuaishou's AI revenue also surpassed 850 million yuan.
These numbers do not yet prove that the hundreds of billions of AI investments have been fully recouped. But at least, the big companies are no longer relying solely on imagination to calculate the future of AI. How much to invest, how long to recoup, and how much revenue can be generated is becoming increasingly clear.
Thus, a new cycle is beginning to take shape: the more computing power there is, the more models can be trained and customers served; after revenue growth, companies are also more confident to continue investing and more easily secured the next round of funding.
Money transforms into computing power, computing power drives business, and business supports the next round of investment.
However, the dynamics of the capital market are also changing.
In the first half of this year, whether in China or the United States, capital was very enthusiastic about AI. But entering the second half, the market is becoming more cautious, with increasing attention on whether high valuations can be realized and how long large investments will take to recoup.
This is also why Alibaba's 80 billion Hong Kong dollar share placement is worth noting. Alibaba does not lack money, yet it still chose to take money now. ByteDance's expansion of syndicated loan scale and Tencent's issuance of long-term bonds also reflect similar considerations.
For big companies, as long as money is still accessible, preparing ammunition in advance for the coming years is clearly more prudent than waiting to refinance when the actual need arises.
Because the need for money will not only arise for these few companies. Platform companies, chip companies, cloud vendors, and large model companies will all continue to increase investments. When everyone simultaneously needs several hundred billion or even trillion yuan in funds, capital itself may become a scarce resource.
These large companies are increasingly clear that this battle for AI is different from the internet era. Technology determines whether you can get to the table, but how deep your pockets are can decide how long you can stay at the table.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
