DeepSeek is sprinting towards an IPO, and the valuations of the "Four AI Rising Stars" are all set to be recalculated

Bloomberg reports that DeepSeek has initiated preparations for an IPO, with the earliest potential listing application submission expected in 2026, targeting the domestic capital market and aiming to go public in 2027. As of now, the company has not formally submitted listing documents or announced a clear listing plan. 

Back in May this year, DeepSeek completed its first external financing since its establishment, raising approximately $7 billion, with a post-money valuation of around $52 billion.

Just one month later, market news emerged that the company was in discussions with potential investors for a new round of financing, with the proposed pre-money valuation raised to $71 billion.

This company, once renowned for being "backed by High-Flyer, having sufficient self-owned funds, and long rejecting external financing," has now consecutively secured financings and is advancing rapidly toward the capital market. This reversal raises three core questions that run through the entire industry:

Why does DeepSeek suddenly need massive capital?

How will its capitalization path reshape the competitive fates of the "Four AI Rising Stars": Zhipu AI, MiniMax, Kimi, and StepFun?

Will DeepSeek's entry lift the valuation ceiling for Chinese AI, or burst the current valuation bubble in the large model sector?

DeepSeek: From Never Short of Money to Embracing Capital

This is the first time DeepSeek has widely opened its doors to large-scale financing in the three years since its founding.

The $7 billion raised in the first round far exceeds the single-round financing size of the vast majority of AI startups.

Notably, founder Liang Wenfeng reportedly invested approximately 20 billion RMB, making him the largest single contributor in this round. Industrial capital players including Tencent, CATL, JD.com, and NetEase participated in the investment, and the National Artificial Intelligence Industry Investment Fund also became a minority shareholder.

This means this round of financing is not a cash-out exit for the founder, but a move to introduce long-term industrial capital and state capital to the company, building a more stable foundation for capital and resources.

Launching the next round quickly before the first round of financing is fully digested reflects that DeepSeek's competitive logic is changing, with capital demands concentrated in four areas:

First, competition in frontier models has shifted to large-scale Agent operation scenarios, where total computing power consumption is rising instead of falling.

Second, the construction of self-owned data centers has been put on the agenda.

According to reports from the Financial Times, the new round of financing is directly tied to DeepSeek's expansion of capital expenditures, which primarily includes building self-owned data centers and purchasing AI chips in bulk.

Third, the layout of self-developed inference chips has been initiated.

The market has repeatedly reported that DeepSeek is developing AI chips tailored for inference scenarios, aiming to reduce reliance on overseas and third-party chips.

This marks that DeepSeek is evolving from a pure model lab to a system-level company integrating "models, computing power, and chips", but it also means longer R&D cycles, higher upfront investments, tape-out risks, and greater supply chain costs.

Fourth, full-scale expansion of the team and commercialization system.

DeepSeek has publicly stated plans to at least double the headcount of employees across all departments.

The organizational model that used to rely on a small team and research-driven operations is also expanding toward a more complete system covering R&D, product development, commercialization, and listing governance.

Overall, DeepSeek has entered a new stage of competition in full-stack AI infrastructure, where the cost structure it faces has changed, and the demand for large-scale capital has naturally increased accordingly.

Large Model Companies Are Being Rearranged

The most direct impact of DeepSeek joining the IPO queue is that it will completely rewrite the capitalization narrative of China's large model industry, with the "Four AI Rising Stars" being the first to bear the brunt.

Previously, Zhipu AI and MiniMax enjoyed a premium as "scarce targets among the first batch of listed large model companies", while Kimi and StepFun competed for the position of "the third listed large model company".

When DeepSeek, a widely recognized top player in technology, enters the market, the core question will immediately become: apart from DeepSeek, what justifies the current valuations of other companies?

Zhipu AI: Benefiting from the Momentum in the Short Term, Facing Direct Valuation Scrutiny in the Long Run

In the short term, DeepSeek's preparation for listing may be beneficial to Zhipu AI.

Both companies develop general foundational models, emphasize code capabilities, Agent applications, enterprise APIs, and open-source ecosystems, and are both regarded as core players in China that benchmark against OpenAI.

DeepSeek has not yet entered the secondary market. In the absence of direct trading targets, Zhipu AI, which has already listed on the Hong Kong Stock Exchange and has a business structure most similar to DeepSeek, is likely to become an alternative trading entry point.

But in the long run, Zhipu AI is also the most vulnerable to the impact from DeepSeek.

After Zhipu AI listed on the Hong Kong Stock Exchange in January 2026, its stock price once hit HK$2980 intraday on June 22, about 25.6 times its issue price, with a total market cap exceeding HK$1 trillion, equivalent to around $128 billion.

Calculated based on its audited 2025 revenue of 724 million RMB, its static price-to-sales ratio exceeds 1200 times.

Although private valuations are not fully comparable to secondary market market caps, DeepSeek's proposed pre-money valuation for the new round is only $71 billion, which raises a sharp market question:

Why is DeepSeek, which has stronger global technological influence and a more prominent open-source brand, valued lower in private financing than Zhipu AI's market cap in the secondary market?

Once DeepSeek formally submits its listing application and releases its prospectus, investors will for the first time obtain complete financial and operational data that can be compared horizontally.

If DeepSeek has larger revenue, lower costs, and stronger technological influence, but its IPO valuation is lower than Zhipu AI's, Zhipu AI's scarcity premium may be compressed.

If DeepSeek ultimately lists at a higher valuation, it may in turn raise Zhipu AI's valuation ceiling.

Therefore, Zhipu AI is likely to be the company that benefits the most from DeepSeek's IPO expectations in the short term, and is most easily revalued by DeepSeek in the long run.

MiniMax: Differentiated Competition Remains, Commercialization Assessment Is Upgraded

MiniMax faces relatively less direct impact, as it is not an investment target identical to DeepSeek.

DeepSeek focuses more on foundational models, open-source ecosystems, code capabilities, and developer platforms;

MiniMax, on the other hand, has consumer products such as Talkie, Conch AI, and audio/video generation tools. In 2025, over 70% of MiniMax's revenue came from overseas markets.

It can fully present a different story to investors: investing in DeepSeek means investing in China's foundational model leader, while investing in MiniMax means investing in a global AI-native application platform.

But DeepSeek's capitalization moves will still indirectly raise the market's requirements for MiniMax.

In July this year, MiniMax announced plans to raise approximately $2.05 billion through private placement of new shares and issuance of convertible bonds, with the funds intended for R&D, commercialization, working capital, and general corporate purposes.

After the news was released, its stock price once fell by around 12%, as the market worried about potential equity dilution and subsequent capital expenditures.

This indicates that the capital market's requirements for large model companies have changed.

In the past, as long as a company improved its model capabilities, grew its user base, and expanded its products into overseas markets, it could receive a valuation premium.

Now investors will further ask whether revenue can cover model R&D expenses, whether customer acquisition costs can be reduced, and whether gross margins can continue to improve.

Kimi: Valuation Has a Defined Ceiling and a Reference Point

For Kimi, DeepSeek's listing may either expand its valuation space or expose its valuation shortcomings.

According to media reports, Kimi is in the preparation stage for a Hong Kong IPO. It completed a financing of around $2 billion in May 2026, with a post-money valuation exceeding $20 billion, and its annual recurring revenue (ARR) in April of the same year reportedly surpassed $200 million.

On one hand, DeepSeek's high valuation has opened up the valuation ceiling for the entire industry.

If DeepSeek ultimately lists at a valuation of over $100 billion, even if Kimi only obtains 30%-40% of DeepSeek's valuation multiple, it can still support a market cap of $30 billion to $40 billion, which still leaves room for growth compared to its current primary market valuation.

On the other hand, the pressure is equally tangible.

Investors will not simply price Kimi by applying a proportional discount to DeepSeek's valuation. Instead, they will raise more pointed questions, such as:

What justifies Kimi's foundational model capabilities being worth 30% to 40% of DeepSeek's? Can its user base be stably converted into subscription revenue? Can its Agent products effectively increase ARPU? Is the commercial value of its consumer-facing entry point really higher than that of an open-source model ecosystem?

The most rational choice for Kimi is to actively differentiate itself from DeepSeek.

Only when Kimi positions itself as the strongest consumer-facing AI entry point, Agent, and productivity platform, rather than being regarded as a "miniature DeepSeek" by investors, can it be more likely to obtain an additional premium at the application layer.

StepFun: Facing the Most Direct Impact, with Dual Pressure on Positioning and Listing Window

Among the "Four AI Rising Stars", StepFun is likely the unlisted company most severely impacted by DeepSeek's IPO.

Previously, market rumors claimed that StepFun had completed shareholding restructuring, dismantled its offshore structure, and was preparing for a Hong Kong IPO, with a target valuation of around $10 billion.

But its core dilemma lies in unclear identity:

It competes directly with DeepSeek in foundational models and Agent applications, but its technological brand power is inferior to DeepSeek's; its C-end product recognition is not as strong as Kimi's; its secondary market influence lags behind Zhipu AI's; and its overseas application revenue is less than MiniMax's.

When DeepSeek joins the IPO track, investors' questions will be extremely direct:

Where exactly is StepFun's leading edge in models? If its model capabilities cannot rank first, why does it still maintain full-scale foundational model R&D investments? Are its collaborations with manufacturers such as OPPO and Geely confirmed revenue orders, or are they still at the level of ecosystem partnerships?

DeepSeek has also disrupted StepFun's original listing rhythm.

It could have smoothly gone public as "the third listed large model company" following Zhipu AI and MiniMax, but this narrative has now lost its appeal.

StepFun is faced with a dilemma:

Going public before DeepSeek may expose the problem of immature financial data and business models;

Waiting to list after DeepSeek's IPO will result in its valuation being fully anchored by DeepSeek's prospectus, drastically narrowing its bargaining space.

Financial Market Restructuring: Valuations Will Be Discounted Against DeepSeek

The most far-reaching impact after DeepSeek enters the capital market may be that China's large model industry will for the first time have a local valuation benchmark.

Prior to this, when pricing Chinese large model companies, investors could only refer to overseas players such as OpenAI and Anthropic, and then apply different proportional discounts based on China's market environment, regulatory restrictions, internationalization level, and user payment capacity.

This valuation method is highly dependent on market narratives and sentiment, with extremely high flexibility.

After DeepSeek lists, the situation will change.

It operates in a market environment similar to Zhipu AI, MiniMax, Kimi, and StepFun, faces comparable chip restrictions, talent costs, regulatory systems, and user payment capacities, while also possessing technological influence recognized by the global market.

Once its financial data is made public, investors can directly calculate: how many times is Zhipu AI's valuation relative to DeepSeek's? What percentage of DeepSeek's valuation does Kimi deserve? Can MiniMax's application business obtain an additional premium beyond its model value?

China's large model valuation system will officially shift from being discounted against OpenAI to being discounted against DeepSeek.

In the long run, this differentiation will become more brutal.

Companies that are inferior to DeepSeek in technological capabilities, have revenue growth below expectations, low gross margins, faster cash burn, but valuations close to or even exceeding DeepSeek's will all face valuation compression.

This is not a general rise or fall across the board, but a process that first lifts the valuation ceiling of the entire industry, then squeezes out the scarcity premium that lacks fundamental support.

Currently, widespread market rumors state that DeepSeek is targeting the domestic capital market. If it ultimately chooses to list on the A-share market, DeepSeek will become the purest foundational model target on the A-share market.

It will not only set pricing standards for model companies, but also gradually become a critical entry point for public funds and index funds to allocate positions in China's AI track, and may even act as a valuation anchor for the entire AI industry chain covering domestic computing power, self-developed chips, and data centers.

But it is worth noting that the total amount of market capital is not unlimited. DeepSeek's large-scale financing may create a capital siphon effect on other AI companies, raising the difficulty of IPO pricing and issuance for small and medium-sized AI enterprises.

Therefore, what DeepSeek's IPO truly changes is not just its own equity structure, but the entire evaluation system of China's AI industry.

All companies need to prove their technological strength, revenue quality, cost control capabilities, and capital efficiency.

At this point, the competition among China's large model companies has officially entered the stage of every company's balance sheet.

This article is sourced from the WeChat Official Account "World Model Workshop", authored by World Model Workshop, and published by 36Kr with authorization.