China’s AI is increasingly burning money
Qifan Business
2026.11.11 14:55
· From Jiangsu
Full text 4448 words
Text | Qifan Business, Editor丨Lu Yao
On November 9, a SpaceX plan to invest an estimated US$24 billion surfaced. According to public hearing documents, the company is preparing to build a gas-fired power plant and supercomputing center in Texas, USA.
The project has not yet been approved, but a technology company famous for its aerospace business has begun to consider solving its own power supply for computing facilities.
Microsoft, Google, Amazon and Meta are continuing to increase AI capital expenditures. In addition to chips, servers and data centers, power supply has gradually become an important factor affecting project progress and construction costs.
The investment scope of AI competition has extended from computing equipment to energy infrastructure.
China's major Internet companies are also expanding investment. Alibaba, Tencent and Baidu continue to purchase chips and expand computing facilities, and their AI business revenue is also growing. The supply of advanced chips is limited, competition in model services is fierce, and the growth of some mature businesses is slowing down. Chinese companies are under different operating pressures from U.S. technology companies.
More and more funds are flowing into AI, and financial statements are beginning to show another side: the growth of business income may not keep up with the speed of investment.
01 208 billion yuan, where was it spent?
On September 17, research institution Rhodium Group released a report on China’s AI financing. The research covers 13 companies, including Alibaba, Tencent, Baidu, Huawei, telecom operators, model companies and independent data center operators. The report estimates that the AI infrastructure capital expenditures of these companies will reach 932 billion yuan in 2026, a year-on-year increase of 103%; it may exceed 1.2 trillion yuan in 2027.
China's AI infrastructure investment scale is still only 15% to 20% of that of the United States, but financial pressure has already appeared in the financial statements of major Internet companies. According to Rhodium Group statistics, in the first half of 2026, the total capital expenditures of Alibaba, Tencent and Baidu reached 208 billion yuan, and the total free cash flow during the same period was negative 16 billion yuan.
The three major manufacturers invested more than 200 billion yuan in half a year, and the cash generated from operating activities was no longer enough to cover capital investment in the same period. Not all of these capital expenditures are used for AI, and the negative free cash flow does not mean that the company is losing money. What is really worthy of analysis is that the continued expansion of infrastructure investment has begun to change the way Internet companies use funds.
Alibaba’s financial report provides a specific example.
In the second quarter of 2026, Ali's net cash flow from operating activities was 22.945 billion yuan, a year-on-year increase of 11%; cash expenditures for the purchase of properties and equipment were approximately 67.660 billion yuan, and free cash flow was net outflow of 44.670 billion yuan, compared with a net outflow of 18.815 billion yuan in the same period last year.
Alibaba Cloud's operating performance is not bad. In the same quarter, AI-related product revenue achieved triple-digit year-over-year growth for the twelfth consecutive quarter, and the cloud segment's adjusted EBITA margin reached 12%. As of the end of June, Alibaba held 474.505 billion yuan in cash and other current investments, and its capital reserves are still sufficient.
Source: Alibaba official: June 2026 quarterly results
Alibaba currently has no shortage of money to continue investing in AI. The real pressure comes from the gap between the speed of investment and the speed of revenue realization.
The purchase of chips, servers and network equipment requires funds to be paid in advance, but the revenue generated by customers' use of cloud services is gradually realized later. Alibaba also has to bear the cost of electricity, maintenance and equipment depreciation of the data center. The growth in computing power demand can drive revenue, and whether new equipment can be fully used determines the return on investment.
Just because a data center is built does not mean that the corresponding computing power has been sold. Enterprises need to continue to obtain customer orders in order to improve equipment utilization and dilute fixed costs. Alibaba Cloud's AI business is growing rapidly, but revenue growth has not yet allowed the group's free cash flow to improve simultaneously.
Tencent’s capital occupation has taken another form. Citing information disclosed by Tencent, Rhodium Group said that in the second quarter of 2026, Tencent's AI-related advances of approximately 51 billion yuan were included in operating cash flow, and capital expenditures in the quarter were approximately 59 billion yuan.
Prepayments and capital expenditures are different accounting items and cannot be added directly. Together, they reflect a purchasing characteristic: In order to lock in advanced chips and computing resources, companies may pay large sums of money before the equipment is delivered and the assets are confirmed.
Computing power competition is not only about who has the ability to purchase chips, but also about who can come up with funds in advance, endure the delivery cycle, and convert procurement resources into actual business. The larger the purchase size, the higher the importance of equipment utilization and subsequent market demand.
Baidu's situation is more complicated. In the second quarter of 2026, Baidu's AI-driven business revenue was 12.5 billion yuan, a year-on-year increase of 25%; AI cloud infrastructure revenue was 7.3 billion yuan, a year-on-year increase of 50%. Traditional online marketing business is still declining, but basic models, cloud computing and autonomous driving require continued investment.
Alibaba can rely on e-commerce, Tencent has gaming and advertising businesses, and Baidu needs to continue to fund new businesses amid pressure on search advertising growth. AI cloud revenue is growing rapidly, but it is not enough to prove that new businesses have completely filled the changes in traditional businesses.
When it comes to investing in AI, Alibaba needs to improve capital usage efficiency, Tencent needs to balance large-scale procurement and business needs, and Baidu also needs to face changes in the cash generation capabilities of mature businesses.
The three companies are all increasing investment, but they are not on the same starting line.
02 If the price of a model increases, you may not make more money
There have been changes in the commercialization of China's large model market.
Model vendors used to compete for developers through low-price APIs, open source models and free applications. Enterprise customers can switch between different models and deploy open source models on their own, with price becoming an important factor affecting procurement.
In 2026, some model manufacturers will begin to increase API prices. The Rhodium Group report mentioned that DeepSeek also adjusted pricing in August and introduced higher peak hour prices. Although continuous price reduction will help expand the number of calls, it will compress the revenue space of service providers, especially when computing resources are still expensive.
The price increase did not immediately solve the profitability problem.
According to Rhodium Group statistics, the overall gross profit margins of Zhipu and MiniMax will drop from approximately 41% and 25% in 2025 to approximately 26% and 18% in the first half of 2026, respectively. The business structures of the two companies are different, and the decline in gross profit margin cannot be entirely attributed to model pricing, but the gap between revenue growth and profitability has already emerged.
Source: Rhodium Group report "Problems with eventual profitability" chapter
Enterprise projects often provide higher unit revenue, but require sales, delivery and customization support, and are limited in the speed of market expansion. Open platforms make it easier to reach a large number of developers, the scale of calls can grow rapidly, and customers are more price-sensitive.
Model companies are facing a difficult multiple-choice question: high-margin businesses may not be easy to expand, and API businesses with faster growth may not necessarily bring equally high profits.
MiniMax is also adjusting its customer structure. Data cited by Rhodium Group shows that the proportion of annual recurring revenue related to its corporate customers has increased from approximately 30% in August 2025 to approximately 80% in August 2026. An increase in the proportion of enterprise customers is usually conducive to establishing a more stable payment relationship, but product deployment, computing power leasing and R&D investment will also increase operating costs.
Model inference also has the characteristic of continuously consuming computing resources. Once the user asks a question, the system needs to complete a calculation. Complex tasks involve longer contexts, multiple rounds of reasoning, and tool calls, and a single task may require more computing power.
Model optimization can reduce unit computing costs, and new application forms continue to increase computing requirements. What corporate customers buy is not just tokens, but whether the model can stably complete tasks and reduce manual investment. What manufacturers need to prove is how much customers are willing to pay to complete a task, not just how many tokens the model can generate.
The accounting for consumer-level AI applications is even more difficult.
The Rhodium Group report mentioned that Alibaba's newly divided AI application business achieved revenue of approximately 3 billion yuan in the second quarter of 2026, with an adjusted EBITA loss of approximately 14 billion yuan. The scope and cost allocation of this business segment cannot simply be equated to the operating results of a single AI assistant, but the gap between revenue and losses is worthy of attention.
Free AI assistants can quickly acquire users, but they also need to continue to bear the costs of reasoning and product operation. Subscriptions, advertising, e-commerce commissions and enterprise services may all become sources of income. Whether the user scale can be converted into stable payment remains to be verified by more operating data.
Alibaba Cloud and Baidu Smart Cloud have channels to sell computing power and model services to enterprises, while consumer-level AI assistants need to establish another set of charging logic. Having a large number of active users is not enough to prove that the business can be independently profitable.
Chinese AI has begun to compete from competing for model performance, developers and user scale to competing for customer payment capabilities.
03 Who can keep AI alive?
The capital market is providing more and more funds for Chinese AI.
According to Rhodium Group statistics, in the first eight months of 2026, major model companies such as Zhipu, MiniMax, DeepSeek and Dark Side of the Moon have received a total of 179 billion yuan in funds through private equity financing, listings and private placements. In 2025, the private equity and venture capital financing scale of these companies will be approximately 9 billion yuan, and the range of financing channels in the two sets of data is not exactly the same.
Large manufacturers have also begun to adjust their capital arrangements. Rhodium Group research pointed out that some Internet companies have reduced stock repurchases, sold short-term investment assets, and increased borrowings to free up funds for AI infrastructure investment. ByteDance received a US$29.6 billion syndicated loan in September. According to Reuters, most of the funds are expected to be used for AI-related projects.
Competition for AI is extending from corporate internal R&D budgets to bank loans, stock markets and long-term asset financing. Technical capabilities determine how much capital a company needs to invest, while financing capabilities affect whether it can continue to invest.
The capital structure of Chinese AI companies is significantly different from that of American technology giants. Large U.S. technology companies use the bond market more extensively for financing, while Chinese companies rely more heavily on operating cash flow, equity financing and bank loans. Different financing channels lead to different risks taken by enterprises.
Large Internet companies have mature businesses and balance sheets that can support long-term investments through operating income, loans and other financing arrangements. Independent model companies rely more on the equity market, while data center operators need to resort to bank loans, asset securitization, real estate investment trusts and financial leases.
What really widens the gap may be the investment recovery cycle.
After cloud vendors purchase GPUs, they need to gradually recover costs through customer calls and computing power leasing. Model companies continue to pay R&D and inference expenses, but revenue growth may not quickly cover expenses. Data center operators need to maintain a sufficiently high rental rate within the equipment depreciation and financing period.
AI infrastructure is a business that requires a large amount of money to be paid in advance and then the future income will be used to slowly repay the capital. When equipment procurement is completed, the financial risks borne by the enterprise have already been formed; whether customer demand can continue to grow will take time to test.
Related risks are already emerging in the data center industry.
Citing data from the China Academy of Information and Communications Technology, Rhodium Group said that in the first half of 2026, 54.3% of China’s new computing power center projects involved financial leasing. Financial leasing lowers the financial threshold for enterprises to purchase equipment in one go, and also allows some operators to expand their computing power through leverage.
The report also mentioned that Sichuan Binji Technology purchased GPUs through sale and leaseback methods. After the company went bankrupt, 15 licensed leasing companies suffered direct losses.
This case involves the financing and operating issues of a specific enterprise and cannot represent the entire computing power leasing industry. However, it exposes a risk of asset-heavy expansion: equipment procurement relies on financing, and future income depends on market demand. Once there are changes in GPU prices, rental rates or customer orders, debt repayment pressure may rise rapidly.
Computing power assets are not equivalent to stable cash flow. Servers can be purchased, but long-term customers cannot be purchased directly through financing.
The advantages of large Internet companies are reflected here. Alibaba has e-commerce business, Tencent has gaming and advertising revenue, and Baidu still has search business. Even if the AI project has not recovered its investment, mature businesses can share the financial pressure to a certain extent.
Independent model companies do not have established operations of the same size. Financing can help them expand their R&D teams, purchase computing power, and win customers, but it cannot replace operating income indefinitely. How high the valuation the capital markets are willing to pay and how fast the company can sustain revenue growth will affect the next round of financing.
For data center operators with long-term contracts, stable customer demand can reduce investment risks. Companies that rely on short-term computing power leasing and have high debt ratios are more susceptible to changes in equipment prices and market demand.
China's AI industry still has room for growth. Enterprise applications, agents, domestic chips, and inference optimization may all change future computing power requirements and service costs. The expansion of industry scale does not guarantee that all participants will receive the same returns.
Making the model is only the first step to participate in the competition. How far a company can go depends on whether it can continue to pay for R&D and computing power costs, retain customers, and recoup investment before equipment is updated.
China’s AI is increasingly burning money. Well-funded companies can strive for longer payback cycles, while companies that rely on external financing need to prove business value faster. In the next round of competition, model capabilities will still be important. How long companies can endure and how much customers are willing to pay will gradually determine their respective positions.
Respect for the original work, if you are qualified, you should be appreciated.
According to the real-name requirement of the Cybersecurity Law, please bind your mobile phone number before posting a comment.
Express
2026-11-11 23:07
Wang Yi speaks on the phone with US Secretary of State Rubio
2026-11-11 23:06
Domestic commodity futures markets mostly closed higher at night, with chemical products leading the gains.
2026-11-11 22:41
Mexico's main stock index rose to 72,231 points after the opening, hitting a record high
2026-11-11 22:36
AI cloud computing service providers collectively strengthened, CoreWeave rose 14%
2026-11-11 22:34
New York gold futures fell below $4,030 per ounce, down 0.01% on the day
2026-11-11 22:33
Nasdaq China Golden Dragon Index fell more than 1%
2026-11-11 22:32
The Philadelphia Semiconductor Index opened up 4.5%, recovering all of its 12% loss
U.S. stocks opened: the three major indexes rose collectively, CoreWeave rose more than 16%
2026-11-11 22:26
Russia plans to select lunar nuclear power plant plan in 2027
JD Logistics launches Tibet train
2026-11-11 22:25
EU 2027 budget initially finalized
2026-11-11 22:24
A magnitude 3.0 earthquake occurred in Hejing County, Bayinguoleng Prefecture, Xinjiang, with a focal depth of 17 kilometers
2026-11-11 22:20
China, Turkey and Egypt launch international cooperation initiative to strengthen high-quality carbon credits
2026-11-11 22:19
Hualin Securities: CEO Dong Dong received the regulatory talk decision from the Jiangxi Securities Regulatory Bureau
Eton Electronics: The private placement investment project will start construction in December
2026-11-11 22:15
Guosheng Securities: Received decision on administrative penalties and administrative regulatory measures
2026-11-11 22:02
Xiangcai Shares: Subsidiary Xiangcai Securities received penalties and regulatory measures from the Hunan Securities Regulatory Bureau
2026-11-11 21:53
The dollar's intraday gains against the yen expanded to 0.5% and are now at 159.53
2026-11-11 21:33
OPEC maintains forecast for global oil demand growth of 2.6 million barrels per day in 2027
OPEC: Crude oil production of member countries increased by 1.97 million barrels per day in October from the previous month
Scan to download App
