The AI Mirror Behind DeepSeek's Financing: Alibaba Goes Left, Tencent Goes Right

When a company brings both Alibaba and Tencent to the negotiation table and then shows them the door one after another, it in itself is a signal.

In May this year, DeepSeek exceptionally launched its first round of external financing. The financing target of 50 billion yuan, the valuation of 350 billion yuan, the founder personally investing 20 billion yuan, and the leading investment by the National Big Fund - these figures are explosive enough. But what really caught the industry's attention was what followed: Alibaba withdrew, while Tencent is expected to enter the scene as a minority financial investor. Alibaba might have wanted ecological binding, while Tencent only wanted a financial stake without interfering in the technical direction. However, it seems that neither side reached an agreement.

One exits, and the other enters. This is not an accident based on who offers a higher price but a mirror moment of the strategic divergence between the two giants in the AI era. Reviewing the ins and outs of this financing and dissecting the underlying logic of Alibaba's iron - fisted self - research and Tencent's ecological embedding might be more interesting than just chasing the valuation figures.

Self - Research Closed - Loop vs. Ecological Embedding: Three Different AI Approaches

For Alibaba and Tencent, two Internet giants with a combined market value of nearly one trillion yuan, it's not surprising to many insiders that they hit a wall in front of a startup. They know very well that the big tech companies' calculations when investing in AI are never as simple as just financial returns.

This approach has a vivid name in the industry called "cloud cash - back investment". The rules are roughly as follows: A big tech company claims to invest one billion yuan in an AI company, but more than half of it has to "flow back" to the company's cloud business in the form of purchasing cloud computing power.

In other words, the cash that the AI company actually gets is far less. And the big tech company not only completes the bookkeeping action of external investment but also boosts its own cloud revenue. More importantly, once the money is taken, the AI company has to accept "ecological binding": The model is deployed on the big tech company's servers, and the API is preferentially connected to the big tech company's product system. The technology roadmap and commercialization rhythm have to match the big tech company's strategic rhythm.

Let's take a closer look at Alibaba and Tencent, who both appeared at the negotiation table. Their motives are quite different.

What Alibaba wants is "ecological control". This logic is not new in the industry: Investment means procurement, and equity participation means binding. The conditions offered by Alibaba imply the requirement to deeply integrate the DeepSeek model into its own product lines such as Taobao, Gaode, and Alibaba Cloud.

A potential scenario is that the DeepSeek API is preferentially deployed on Alibaba Cloud, the inference traffic is preferentially directed to Taobao's intelligent shopping guide, and the technology roadmap needs to be customized and optimized according to Alibaba's e - commerce scenarios. For Alibaba, this is not a financial investment but a way to "weld" external top - notch technology into its own business foundation using capital means. This continues Alibaba's consistent strategy, just like when Alibaba Cloud acquired Changting Technology and invested in Shumeng Works in the early days, all of which required the technology teams to be deeply integrated with the Alibaba ecosystem.

But in my opinion, the consequence of this "welding - style" integration is that the technology roadmap of the invested company will gradually lose its universality and eventually become a "customized model" only serving Alibaba's scenarios, narrowing its industry influence. However, in the AI era, Alibaba's anxiety is very specific. If it doesn't "lock" the best models into its own territory, its competitors may use them to strengthen their e - commerce and cloud businesses.

Tencent has a different calculation. Historically, Tencent has built a huge ecosystem through investment but never forces the invested companies to be exclusive to it.

In the field of AI, Tencent continues this style: mainly financial investment, without interfering in the technology roadmap and without requiring ecological exclusivity. People close to the deal revealed that Tencent's offered conditions didn't even require DeepSeek to use Tencent Cloud preferentially. It just hopes to have the priority to cooperate in the directions of WeChat intelligent agents and game NPCs in the future.

This "light - participation" attitude reflects Tencent's different positioning of AI. It doesn't expect to reconstruct its business with an external model but hopes to use external top - notch technology to "lubricate" its existing social and content scenarios. Tencent's confidence lies in WeChat, a super - entry point. No matter who invests in DeepSeek, as long as users want to use the best AI capabilities in WeChat, Tencent can achieve this through cooperation rather than controlling the company.

DeepSeek hasn't clearly chosen a side for the time being, which precisely shows that it neither wants to be "locked in" nor is satisfied with being just a "plug - in". It wants to become the infrastructure for the whole society, rather than an appendage of any giant.

ByteDance's continuous increase in investment has made the situation more complicated. ByteDance hardly appeared in the financing rumors of DeepSeek, not because it's not interested, but because it took another path: self - researching Doubao and heavily investing in the C - end entry point. In 2026, ByteDance pushed its AI capital expenditure up to 200 billion yuan, almost five or six times that of Tencent's AI investment. ByteDance's logic is straightforward: The ultimate battlefield of AI is user time. Whoever can capture the C - end super - entry point first will gain the initiative. It doesn't need to invest in DeepSeek because Doubao itself is a competitor to DeepSeek.

Three companies, three ways of dealing with external top - notch technology. Alibaba wants to "buy, use, and lock"; Tencent wants to "buy and use but not lock"; ByteDance wants to "do it by itself without buying". Behind these three choices are different corporate genes, resource endowments, and judgments on the AI end - game. And DeepSeek's rejection has torn open this gap, allowing us to see that in this AI game, the giants hold different cards and naturally have different logics for playing them. To understand where this difference comes from, we have to look back at the starting points of the three companies.

Genes, Foundations, and Strategic Logic: Two Paths for Alibaba and Tencent

The differences among these companies don't come out of thin air. The AI layout of Internet giants is almost written in their genes.

Alibaba's genes are e - commerce, cloud computing, and industrial services. The logic of Taobao is a closed - loop: from products to transactions, from payment to logistics, it tries to keep the entire chain in its own hands. This gene determines that Alibaba is used to being a controller rather than a connector.

The same is true for Alibaba Cloud. It adheres to self - research from the underlying technology to the upper - layer services. So when DeepSeek emerged, it was difficult for Alibaba to just be a financial investor. It needs a functional module that can be integrated into its own AI system, rather than an independent kingdom.

Tencent has a different logic. Tencent's genes are social networking, traffic, and ecological win - win. It is used to being an open platform, using capital means to connect rather than control.

Whether it's JD.com, Meituan, or Pinduoduo, Tencent's layout in the e - commerce field is mainly through investment rather than doing it on its own. One of the core lessons Ma Huateng learned from the 3Q War is that "openness is more viable than closure". In terms of AI investment, it means that Tencent won't be obsessed with locking DeepSeek into its own box.

The maturity of the technology foundation also determines the bargaining power of both sides. Alibaba's self - research AI system has run through the entire link. The chips of T-head Semiconductor have been mass - produced and delivered, and Tongyi Qianwen is in the same echelon as GPT - 4o. For Alibaba, an external large - scale model is just icing on the cake, not a timely help. Alibaba's withdrawal from DeepSeek's financing is less about "breaking off the negotiation" and more about having no strategic reason to invest. DeepSeek wants independent development and financial investment, while Alibaba wants technology integration and ecological binding - their demands are not on the same level.

Tencent's Hunyuan has achieved a counter - attack. At the beginning of 2025, when DeepSeek became a sensation, Tencent once fully embraced it, and the priority of self - researching Hunyuan was lowered. But after hiring the former OpenAI scientist Yao Shunyu with a large sum of money, Hunyuan was "restarted" and established the "co - design" methodology of promoting the model and products simultaneously. The preview version of Hunyuan Hy3 released in April 2026 has a Token call volume more than 10 times that of the previous generation, and it has taken the "double first" in call volume and market share on the OpenRouter platform. Although it still lags behind Doubao and Tongyi Qianwen, the gap has been significantly narrowed. The progress of Hunyuan makes Tencent more calm when facing DeepSeek: it can obtain top - notch capabilities from the outside and won't lose its bargaining power due to the lack of self - research.

ByteDance's AI foundation is the most unique among the three. It has no technical burden of self - researching large - scale models and no industrial inertia of B - end monetization. Its core goal is to seize the C - end AI entry point by all means. For this reason, ByteDance even increased its AI budget significantly to 200 billion yuan against the background of a more than 70% plunge in net profit. This almost cost - free investment has put pressure on both Alibaba and Tencent. However, Alibaba's budget is scattered across multiple links such as chips, cloud, models, and applications, while ByteDance's 200 billion yuan is almost entirely invested in C - end applications and computing power reserves. This focus has allowed Doubao to approach the monthly active user scale of Tongyi Qianwen in a short time.

ByteDance's "all - in" approach is forcing the entire industry to rethink the commercialization rhythm of AI. When a company is willing to exchange short - term profits for a long - term entry point, others either have to follow suit and burn money or find another way.

An insight worth pondering is that in the rapidly changing field of AI, the window period of technological leadership is getting shorter and shorter. Alibaba uses heavy - asset self - research to build long - term barriers, but at the cost of reduced flexibility and narrowed external cooperation space; Tencent uses light - asset ecology to maintain openness, but at the cost of continuously catching up in self - research capabilities; ByteDance uses extreme investment to buy time, but at the cost of significant short - term profit pressure. There is no absolute superiority or inferiority among these three strategies.

In general, genes determine the path, the foundation determines the bargaining power, and the strategic logic determines the timing of action. The advance and retreat at DeepSeek's financing negotiation table are all projections of these three choices at the capital level.

Closed - Loop Deep Dive and Open Integration: Two Parallel AI Futures

Looking away from the financing negotiation, the AI futures of Alibaba and Tencent are actually quite clear: two paths that won't intersect in the short term and may not necessarily result in one swallowing the other in the long run.

Alibaba will continue to move forward along the closed - loop of full - stack self - research. CEO Wu Yongming has set the tone that AI will become Alibaba's main growth engine. The AI - related revenue of Alibaba Cloud has grown by three - digit numbers for ten consecutive quarters, and the goal for the next five years is to achieve an annual commercialization revenue of over 100 billion US dollars from cloud and AI. In terms of chips, T - head Semiconductor is increasing its investment in domestic GPUs to gradually reduce its dependence on NVIDIA. Tongyi Qianwen has formed a complete set of capabilities from inference to agents and then to multi - modality. The Qianwen App has 300 million monthly active users, and Wukong is accelerating its penetration in the enterprise - level market. The three monetization methods of computing power leasing, Token charging, and application subscription form a deep enough moat.

Tencent's approach is completely different. On the C - end, it targets WeChat, the largest ecosystem, and moves towards a "decentralized intelligent entry point" - users don't need to open a dedicated AI application but can naturally use AI while chatting or watching videos. On the B - end, WorkBuddy and CodeBuddy have been connected to Enterprise WeChat and Tencent Cloud to help customers improve efficiency. Judging from the performance of Hunyuan Hy3, Tencent cares more about whether the model can be truly put into use rather than ranking on the list. Tencent will also continue to invest in high - quality AI companies through financial investment to integrate external models into its own scenarios. If the investment in DeepSeek materializes, it will be a continuation of this strategy. Tencent's money - making path is more direct: improving advertising efficiency, reducing game R & D costs, and accelerating monetization on the cloud and WeChat ends. In 2026, Tencent Cloud has completed three price adjustments, indicating that it is starting to have pricing power in the AI office field.

Agents are becoming the next common focus of both sides. Alibaba focuses on breaking through in agent capabilities. The Qianwen App is connected to ecosystems such as Taobao, Alipay, and Fliggy, and has taken the lead in realizing AI - assisted handling of complex life tasks. Tencent focuses on the implementation of agent capabilities in B - end productivity and platform - based governance. In the multi - modality direction, the two companies are almost simultaneously making efforts: on April 16, 2026, they both released their respective world model products on the same day, pushing the AI competition from language parameters to spatial intelligence. Although their strategic paths diverge, their judgments on the technology direction are surprisingly consistent.

In this long - term battle, Alibaba holds two aces: computing power infrastructure and B - end ecosystem. Tencent has the largest C - end traffic pool in the country. ByteDance is closely pressing on the C - end with 200 billion yuan in heavy investment and nearly 400 million monthly active users of Doubao. Each of them has its own weaknesses: Alibaba has high full - stack barriers but is difficult to overwhelm in the C - end in the short term; Tencent's ecosystem is flexible, but its self - research of Hunyuan still needs to catch up; ByteDance's investment is fierce, but the net profit pressure will sooner or later be transmitted to the decision - making level.

A possible mid - term pattern is that Alibaba holds on to the B - end and cloud infrastructure, Tencent holds on to the AI embedding in social scenarios, and ByteDance seizes the independent C - end entry point. The three form a staggered competition rather than a life - and - death struggle. For the entire industry, this multi - polar pattern is more conducive to technological diversification and innovation vitality than a single monopoly.

What really determines the outcome may not be whose model is stronger but who can first turn AI from a "tool" into a "lifestyle". Alibaba bets that enterprises and developers can't do without its computing power; Tencent bets that users can't do without the intelligent assistant in WeChat; ByteDance bets that Doubao will become the next national - level entry point. None of these three paths is easy, and no one can easily deny the others.

Back at DeepSeek's financing negotiation table, Alibaba left, and Tencent got closer. This is not a disagreement over DeepSeek's pricing but different answers from two business philosophies. Alibaba builds a "castle": high walls and deep moats, self - sufficient. Tencent builds a "port": ships can freely enter and exit, and the water surface is prosperous. In the rapidly changing field of AI, there is no absolute superiority between the "castle" and the "port". The only certainty is that DeepSeek's large - scale financing in May is destined to be a page marking the divergence of Chinese Internet giants in AI.

This article is from the WeChat official account "Whale Business" (ID: bizwhale), author: Hu Duzhi, published by 36Kr