Who is the winner after the DeepSeek shock | Analyzing Alibaba, Tencent, Huawei, and Baidu

[Summary]

It has been about a year and a half since the impact of DeepSeek-R1 in early 2025. Furthermore, DeepSeek-V4 Preview was announced in April 2026, and the 75% off price of V4-Pro was scheduled to transition to the official price.

The essence of the DeepSeek shock is not that it ended the competition in AI model performance.

The AI model itself has rapidly become a commodity, forcing Chinese AI companies to compete in terms of profit margins and free cash flow (FCF, cash that companies can use freely).

Low-cost APIs and open weight are a boon for developers and businesses. The profitability line for AI agents, in-house AI, local AI, and in-house operating models will decline.

On the other hand, there will be headwinds for companies that sell models externally via API and make a profit. If prices fall, profit margins will be difficult to increase even if usage increases.

In this article, as of May 31, 2026, we will summarize the price destruction that DeepSeek has brought to the Chinese AI market, DeepSeek's own profit model hypothesis, and the winners and losers after the DeepSeek shock.

The DeepSeek shock refers to the phenomenon in which the profit structure of the Chinese AI industry has changed due to the low-priced API and open weight strategy of the DeepSeek-R1 and V4 series models.

Just the conclusion first

Company	View after the DeepSeek shock
Tencent	Most likely winner. Easy to incorporate AI into existing advertising, payment, and business services
Alibaba	High risk, high return. AI cloud infrastructure is strong, but focus is on CAPEX and FCF
Huawei	Candidate infrastructure winner. Targeting national infrastructure demand with Ascend, Huawei Cloud, and Pangu
Baidu	Waiting for structural change. Decrease in search advertising and reversal of FCF are more important than AI sales

What you can learn from this article

✓ The essence of the DeepSeek shock
✓ Why AI models are becoming harder to monetize
✓ How Alibaba, Tencent, Huawei, and Baidu compare
✓ Five metrics investors should watch

The essence of the DeepSeek shock is that the “price of AI” has changed

The impact that DeepSeek had on the market was not simply that a high-performance model from China was released.

More importantly, it broke the pricing structure for AI models.

DeepSeek official API Docs as of May 2026 lists V4-Flash and V4-Pro. V4-Flash costs $0.14 per million input tokens and $0.28 per million output tokens. V4-Pro is said to be scheduled to transition to the official price after 75% off, and is announced to be $0.435 per 1 million input tokens and $0.87 per 1 million output tokens.

This level will significantly change the profitability of AI agents and business automation.

AI agents use more tokens than regular chat. You break down tasks, search for them, call tools, fix them if they fail, and do them again.

If the unit price of a model remains high, agents tend to stop experimenting. When the unit price decreases, there will be more opportunities for work.

The DeepSeek shock can be summed up in one sentence.

AI model performance improves
↓
DeepSeek low-price APIs and open weights
↓
Standalone models become commoditized
↓
Profit focus shifts from models to agents, SaaS, cloud, and FCF

This is a seismic shift in China's AI market.

Illustration: Where will the profit pool shift after the DeepSeek shock?

Raw token value approaches zero

Of course, the value of AI tokens will not truly drop to zero.

However, from an investor's perspective, the model of ``earning a high profit margin just by selling raw tokens'' has become quite difficult.

If DeepSeek lowers its prices, Alibaba's Qwen, Baidu's ERNIE, ByteDance's Doubao, etc. will have to be conscious of price competition.

As a result, the model API is likely to enter the next loop.

Performance improves
↓
Prices fall
↓
Usage increases
↓
Competitors also cut prices
↓
Standalone model margins thin out

This is good for users.

However, it is difficult for model providers.

Even if API sales increase, profit margins will not increase if inference costs, GPU/NPU, electricity, data center, and human resources costs cannot be absorbed.

That's why the evaluation axis for Chinese AI companies has shifted from ``how smart are the models?'' to ``how can AI be turned into cash?''

Is DeepSeek really profitable?

This is what investors are most concerned about.

How does DeepSeek itself make money by releasing its API at such a low price and releasing its models with open weight?

To be honest, DeepSeek is a private company and detailed revenue and profit margins are not disclosed. Therefore, this should not be read as a fact that can be confirmed in the financial results, but as a hypothesis of the profit model from the perspective of the public disclosure strategy.

There are three possible hypotheses:

DeepSeek's profit model hypothesis	Contents	Key points for investors
Low-priced API	Attracts mass usage and becomes standard for developers	Is it possible to make ends meet by usage volume even at low profit
Inference optimization service	Support for companies to operate AI inference cheaply and quickly	Can cost reduction itself be sold at a high unit price
Enterprise introduction	On-premises, dedicated environment, custom model	Can we get contracts for finance, telecommunications, and large companies

DeepSeek's real weapon is not its low price.

It is inference efficiency that realizes low cost.

For companies, the biggest question when introducing AI is not just "Is it smart?" Will it be profitable to use a large amount of data every day? Where will the data be processed? Can it be incorporated into existing systems? Can the operational costs be calculated?

If DeepSeek can capture this, the cheap sale of model APIs will be an entry point.

However, this is still a hypothesis.

Rather than viewing DeepSeek's price destruction as proof that DeepSeek is definitely profitable, it is safer for investors to view it as a force that lowers the profit margin of the entire Chinese AI market and pushes the monetization layer upwards.

Is DeepSeek itself a winner?

Whether DeepSeek itself will be a winner will not be determined simply by increased API usage.

There are four points to look at: technology standardization, corporate introduction, inference optimization, and monetization.

Points of discussion	Conditions for becoming a winner
Technology standardization	Developers continue to use DeepSeek models as standard parts
Corporate introduction	Dedicated introduction will increase in finance, communications, manufacturing, in-house AI, etc.
Inference optimization	Low-cost inference technology can be sold as a service
Monetization	Not only increase usage of low-priced APIs, but also lead to high-priced contracts

DeepSeek is a price disruptor.

However, it does not necessarily mean that the company that crashes the price will take the maximum profit.

The real challenge lies in how to connect the usage base that has been expanded at a low price to inference optimization and dedicated implementation for businesses.

Winners and losers after DeepSeek shock

After the DeepSeek shock, the Chinese AI market has been divided into winners and losers.

Importantly, companies that sell AI models won't necessarily win.

Companies that can incorporate lower-priced AI into their advertising, e-commerce, payment, cloud, business SaaS, and industrial infrastructure will become stronger.

Position	Main companies	View after the DeepSeek shock
Price competition instigator	DeepSeek	APIs alone tend to have low profits. Focus on inference optimization, enterprise implementation, and technology standardization
AI-powered	Tencent	Easy to incorporate low-cost AI into WeChat, advertising, payments, and Business Services
AI Cloud Core	Alibaba	Model price wars will wear out, but increasing AI usage will be a tailwind for Alibaba Cloud
National infrastructure type	Huawei	Apart from private price competition, targeting government and state-owned enterprise demand with Ascend, Huawei Cloud, and Pangu
Structural change candidate	Baidu	Focus is on whether the decline in search advertising can be replaced with AI Cloud, AI advertising, and Apollo Go

The most important thing in this table is the difference between Tencent and Alibaba.

Tencent is a company that “uses” AI. Mixing AI with existing high-frequency touchpoints such as WeChat, advertising, games, and payments. This is more effective in improving advertising efficiency and business services profit margins than AI alone sales.

Alibaba is also a company that “sells” AI. Provide Qwen and Alibaba Cloud to external customers and turn increased AI usage into cloud sales. However, it also suffers from CAPEX, inference costs, and price competition.

Huawei is yet another layer. Although unlisted, the company aims to capture AI demand for governments, state-owned enterprises, and critical infrastructure through Ascend, Huawei Cloud, and Pangu.

Baidu is a candidate for reversal. Although the AI ratio is increasing, the decline in search advertising and weakness in FCF remain.

Alibaba after the DeepSeek shock: There will be some headwinds, but there is also room for benefits

Alibaba is one of the companies most affected by the DeepSeek shock.

Qwen faces price pressure from DeepSeek. It is difficult to just sell the model API at a high price.

In this sense, Alibaba will also face headwinds in the short term.

On the other hand, if the demand for AI application development and inference increases with DeepSeek, the demand for the cloud that runs them will increase.

In this sense, Alibaba is also a beneficiary.

Cloud Intelligence Group's revenue in Q4 of the fiscal year ending March 2026 was 41.626 billion yuan, an increase of 38% compared to the same period last year. AI-related product sales were 8.971 billion yuan, representing 11 consecutive quarters of triple-digit growth year-on-year.

However, FCF is heavy.

The company's total FCF for the quarter was an outflow of 17.3 billion yuan. The company attributes this mainly to quick commerce, Qwen app user acquisition, and cloud infrastructure spending.

If you look at Alibaba, there is only one question after the DeepSeek shock.

Can the increase in AI usage be converted into Cloud EBITA margin and FCF?

Until this is confirmed, the market will likely view AI cloud as growing, but cash is still heavy.

However, Alibaba has one of the largest AI cloud infrastructures in China, and is one of the companies that could benefit from the expansion of AI usage. It is too early to write down the option value of cloud infrastructure just because of a short-term deterioration in FCF.

Tencent after DeepSeek shock: Absorb low-cost AI into existing economy

Tencent is less likely to get involved in a pure model price war after the DeepSeek shock.

The reason is that AI is more likely to be used within the WeChat economy rather than being sold as an external sales model.

In the first quarter of 2026, Tencent's sales were 196.458 billion yuan, an increase of 9% year on year, Marketing Services sales were 31.935 billion yuan, an increase of 20%, and FinTech and Business Services sales were 61.541 billion yuan, an increase of 10%. Non-IFRS operating profit was 75.6 billion yuan, non-IFRS operating profit margin was 38.5%, and FCF was 56.7 billion yuan, an increase of 20%.

This is a fairly strong cash generation ability among Chinese AI-related companies.

For Tencent, low-cost AI will help improve the efficiency of advertising, mini-shops, payments, and business services.

If ad recommendations improve, advertisers' ROI will increase. If mini-shops and corporate AI agents become more widespread, there will be more room to monetize existing traffic.

However, Tencent is not only optimistic.

AI investment will increase. Development costs for models, inference, data centers, AI talent, and enterprise features will become heavier.

What investors should look at is whether Marketing Services, FinTech and Business Services, and FCF will continue to grow after investing in AI.

In other words, the reason why Tencent appears to be a winner is not because of the sales of the AI model alone, but because it is increasing its profit margin and FCF by adding AI to its existing advertising, payment, and business services.

There are also weaknesses. If the introduction of AI into WeChat worsens the user experience, advertising efficiency will be difficult to increase. If China's consumption and advertising market conditions slow, improving targeting through AI alone will not be able to compensate. Furthermore, if AI investment becomes heavier than expected, even Tencent's FCF growth will slow.

Huawei after DeepSeek shock: Change profit distribution in a place other than price war

Huawei is in a slightly different place than the private internet price war of DeepSeek shock.

Huawei's essence is not so much an AI company as a national infrastructure company in the AI era.

By combining Ascend chips, Huawei Cloud, Pangu, CANN/MindSpore, and communication infrastructure, we will meet the AI demand for governments, state-owned enterprises, and critical infrastructure.

Looking at the relationship diagram, Huawei's strength lies in the following flow.

Ascend
↓
CANN / MindSpore
↓
Huawei Cloud
↓
Pangu / industrial agents
↓
Government, SOEs, and critical infrastructure

Ascend is the computing resource, CANN/MindSpore is the development environment, Huawei Cloud is the billing platform, and Pangu is the industrial application layer. By connecting these four factors, Huawei will be able to capture national infrastructure demand in the AI era, apart from the model price war.

The more DeepSeek lowers the price of its models, the more broadly it will become easier for businesses and governments to use AI. At that time, on the critical infrastructure side, it becomes important to consider ``which cloud to run it on'', ``which domestic chip to use'', and ``which business system to incorporate it into''.

Huawei comes in there.

Since it is unlisted, it is difficult to see the profit margin and FCF of AI alone. However, in 2025, company-wide sales were 880.941 billion yuan and operating cash flow was 127.384 billion yuan.

For investors, Huawei is not something to buy directly, but an entity that changes the cloud space for governments and state-owned enterprises from Alibaba, Tencent, and Baidu.

ByteDance: Private but can't be ignored with advertising AI and Doubao

ByteDance is also a must-see in the Chinese AI market after the DeepSeek shock.

The company is unlisted, and general investors cannot directly buy or sell its shares. Still, it has an advertising/recommendation platform based on Doubao, Volcano Engine, and TikTok/Douyin, which influences the competitive environment.

ByteDance is both a company that sells AI and a company that uses AI.

AI can be incorporated into advertising, video recommendations, content generation, creator support, and enterprise cloud, making it easy to benefit from low-cost AI. On the other hand, if Doubao and Volcano Engine intensify price competition, it will also put pressure on Alibaba Cloud and Baidu AI Cloud's profit margins.

Rather than being an investment target, it is natural to view it as an entity that changes the competitive environment.

Baidu after DeepSeek shock: Reversal requires more FCF than AI sales

Baidu is in the most difficult position after the DeepSeek shock.

AI Cloud Infrastructure is growing. In the first quarter of 2026, it was 8.8 billion yuan, an increase of 79% compared to the same period last year. AI-powered Business also increased by 49% year on year to 13.6 billion yuan.

Meanwhile, online marketing sales were 12.6 billion yuan, down 22% year-on-year.

In other words, AI is growing, but high-margin search advertising is weak.

Price competition from DeepSeek will also put pressure on Baidu AI Cloud's profitability. Even if the growth rate is high, if prices fall, profit margins will be difficult to increase.

For Baidu to truly be reevaluated, AI Cloud, AI-native Marketing Services, and Apollo Go need to contribute to FCF.

Operating cash flow in 1Q 2026 was positive, but FCF after deducting capital investment was -3.246 billion yuan on Baidu Inc. basis.

What the market wants to see is not an increase in the AI ratio.

The question is whether AI conversion will generate cash.

Opposite scenario: model company still has a chance of winning

The discussion so far is based on the hypothesis that the DeepSeek shock will reduce the profit margin of the model alone, and that value will shift to agents, SaaS, cloud, advertising, and industrial infrastructure.

However, there are cases where this view is incorrect.

AI demand will increase faster than price decline

Even if the unit price of a model falls, if usage increases beyond that, there is a possibility that model companies will still have sales and profits.

In particular, if the usage of AI agents, code generation, video generation, and business automation increases rapidly, it will become a large market even with low profit margins.

More tasks that only high-performance models can solve

Not all AI applications will become commoditized.

As the number of tasks that are difficult to perform only with high-performance models increases, such as in finance, medicine, scientific research, complex code generation, and long-term inference, top models may be able to maintain premium prices.

Inference optimization itself becomes highly profitable

If a company like DeepSeek, which is leading in inference efficiency, sells inference optimization, dedicated environments, on-premises implementation, and technology licenses for enterprises, rather than just APIs, model companies will still have a path to high profit margins.

For this reason, it is easy to conclude that model companies are structurally unprofitable.

The important question is whether it will be enough to sell the model cheaply, or whether it can be moved to higher layers through inference optimization and corporate implementation.

Outlook after 2027: Possibility of advancing to AI OS competition

After the DeepSeek shock, the Chinese AI market may become even more layered after 2027.

Model competition
↓
Inference cost competition
↓
Agent competition
↓
AI OS competition

The AI OS referred to here is not the computer OS itself, but rather the platform on which AI agents run the work of companies and individuals across the board.

Just as Windows was the OS in the PC era, in the AI era, the AI OS that bundles agents may become the center of the profit pool.

Rather than whose model to use, what is more important is which Agent platform will run business, advertising, payments, EC, internal data, and the cloud.

In this phase, Tencent's WeChat, Alibaba's DingTalk and Cloud, Huawei's industrial infrastructure, Baidu's search AI and Apollo Go, and ByteDance's advertising and video infrastructure are all candidates for AI OS in different ways.

Investment strategy after the DeepSeek shock: what to watch in the second half of 2026

After the DeepSeek shock, the numbers to look at have changed in China's AI investment.

AI sales and model performance alone are not enough.

What to see	Why it's important
API price	Affects the profit margin of a single model
Inference cost	Determining profitability of AI Agent and business AI
Cloud EBITA margin	Is AI cloud profitable
CAPEX	Is data center investment putting pressure on FCF
FCF	Is AI finally returning to cash

Looking at the movement of the profit pool by layer, it looks like this:

Layer	View after DeepSeek shock
Model layer	Gross profit margin tends to decline due to price competition
Cloud layer	Receiving the increasing amount of AI usage. Focus on occupancy rate and EBITA
Agent/SaaS layer	Once you get into the business, it's easy to make recurring payments
Advertising/payment layer	Companies that add AI to existing touch points can easily protect profit margins
Industrial infrastructure layer	Government, state-owned enterprises, and critical infrastructure tend to have high unit prices

DeepSeek didn't destroy demand for AI.

Rather, it will work in the direction of increasing the amount of AI used.

However, the profit distribution will change.

Model providers will have a smaller share, and value will shift to cloud, agents, SaaS, advertising, payments, and industrial infrastructure.

Companies that can capitalize on this change will be strong.

Companies that cannot take advantage of this will not see their profit margins increase even if AI sales increase.

Conclusion Table: Potential Winners After DeepSeek Shock

The current evaluation can be summarized from an investor's perspective as follows.

Company	Current evaluation	Numbers to check
Tencent	Top Winners	Marketing Services, Business Services, FCF
Alibaba	High risk, high return	Cloud EBITA, CAPEX, FCF recovery
Huawei	Infrastructure winner candidates	AI infrastructure adoption for governments and state-owned enterprises
Baidu	Waiting for structural change	Search advertising, AI Cloud profit margin, FCF
DeepSeek	Technology standardization candidates	API usage, corporate adoption, and inference optimization revenue
ByteDance	Unlisted but a threat	Doubao, Volcano Engine, Advertising AI

This table is not a fixed winner or loser.

Rather, it's a checklist to see which companies should prove which numbers in the post-DeepSeek shock market.

Conclusion: After the DeepSeek shock, there will be competition for cash conversion power

The DeepSeek shock did not end the Chinese AI market.

Rather, it has pushed AI to the stage where it can be used cheaply and widely in real-life operations.

However, it is not an easy world for model companies.

The days of selling raw tokens at high prices are coming to an end, and profits will shift to agents, SaaS, cloud, advertising, payments, and industrial infrastructure.

What investors need to see in the second half of 2026 isn't flashy AI announcements.

With Alibaba, Cloud EBITA margin and FCF recovery.

With Tencent, you can maintain profit margins for advertising and business services using AI.

Huawei will expand the adoption of AI infrastructure for governments and state-owned enterprises.

Can Baidu replace the decline in search advertising with FCF from the AI business?

The winners after the DeepSeek shock will not be the companies with the smartest models.

It is the company that can most efficiently convert AI into cash.

This article is intended to summarize the thinking behind investment decisions, and is not intended to recommend buying or selling specific stocks. Chinese stocks, Hong Kong stocks, and US-listed ADRs are subject to risks associated with price fluctuation risk, exchange rate risk, liquidity risk, regulatory risk, geopolitical risk, and differences in accounting and disclosure systems. DeepSeek is a private company and limited financial information is publicly available.

[Who is DeepSeek? Price destruction caused by Linux in the AI world and China's AI strategy] (/strategy/2026/05/26/deepseek-ai-linux-strategy-2026.html)
[Are Chinese AI companies really profitable? Compare profit margin, FCF, and profit models [2026 edition]](/strategy/2026/06/01/china-ai-companies-profit-margin-fcf-comparison-2026.html)
[Who is the winner in China's AI agent market? Compare the monetization strategies of Tencent, Alibaba, and Huawei] (/strategy/2026/06/01/china-ai-agent-market-tencent-alibaba-huawei-2026.html)
Why Huawei is the biggest threat to Chinese AI | Analysis of Ascend, Pangu, and Huawei Cloud

Source

DeepSeek API Docs “Models & Pricing” https://api-docs.deepseek.com/quick_start/pricing
DeepSeek API Docs “DeepSeek V4 Preview Release” https://api-docs.deepseek.com/news/news260424
DeepSeek API Docs “DeepSeek-R1 Release” https://api-docs.deepseek.com/news/news250120
Alibaba Group “Alibaba Group Announces March Quarter 2026 and Fiscal Year 2026 Results” https://www.businesswire.com/news/home/20260512841182/en/Alibaba-Group-Announces-March-Quarter-2026-and-Fiscal-Year-2026-Results
Tencent Holdings Limited “Tencent Announces 2026 First Quarter Results” https://en.prnasia.com/releases/apac/tencent-announces-2026-first-quarter-results-532860.shtml
Baidu, Inc. “Baidu Announces First Quarter 2026 Results” https://ir.baidu.com/news-releases/news-release-details/baidu-announces-first-quarter-2026-results
Huawei “2025 Annual Report” https://www.huawei.com/en/annual-report/2025

This article is for educational and informational purposes only, based on public information. It is not a recommendation or solicitation to buy or sell any specific security or financial product. Although care is taken with accuracy, the content and future investment outcomes are not guaranteed. Final investment decisions should be made at your own judgment and responsibility.