DeepSeek funding, rewriting everyone's valuation

$10 billion, $45 billion, $51.5 billion.

In 23 days, DeepSeek’s valuation increased nearly fivefold. According to The Information, DeepSeek’s latest funding round raised up to 50 billion yuan RMB, with founder Liang Wenfeng personally investing 20 billion yuan, accounting for about 40% of the round.

What’s truly worth watching about this is not how much DeepSeek is worth, but that it has begun to answer another question: a company that originally emphasized “no fundraising, no commercialization, no roadshows,” once it starts raising funds, pricing, and considering exits, is rewriting not just its own path but also the valuation logic of the entire large model industry.

Because once DeepSeek enters a narrative of commercialization and IPO, the market has to reconsider one thing: do large model companies have the ongoing power to raise prices?

This is the deepest impact it has on the industry.

Why does DeepSeek have to commercialize?

On the surface, DeepSeek’s fundraising is to supplement computing power, infrastructure, and talent; fundamentally, it is shifting from a “research organization” to a “company organization.”

There are three accounts behind this, which cannot be avoided.

The first is the talent account.

In the past, DeepSeek could retain people through technological ideals and organizational temperament; but as the large model industry enters a phase of intense talent poaching, vision alone is no longer enough. What truly retains core talent are the pricing system, option expectations, and exit channels.

Without pricing, it’s hard to give binding equity incentives; without exit expectations, organizational stability will be weakened. For DeepSeek, fundraising is not just about lacking money, but about the company governance reaching a point where capitalization is necessary.

The second is the expansion account.

Large model companies have never only burned money on computing power. Productization, delivery, sales, and ecosystem building are all subsequent heavier investments. Model capabilities can generate buzz, but they don’t automatically turn into revenue. If DeepSeek wants to transform from a “high-capability model company” into a true platform AI company, it must turn technology into products and products into cash flow.

The third is the exit account.

Since external capital is introduced, an exit must be faced. Buybacks are unrealistic; transfer of old shares can only solve partial liquidity issues. The only truly matching scale and narrative for DeepSeek is an IPO. And the premise for an IPO has never been “the model is very strong,” but “the company is commercially viable.”

Therefore, the key change for DeepSeek moving forward is not whether it will continue to develop models, but whether it can prove that its technological advantages can translate into sustainable income—that is, commercialization.

Small profits but quick turnover

This wave of “raising lobsters” has given large model manufacturers a clearer path to commercialization.

In the past, the biggest doubt about large model commercialization was that, although model capabilities were getting stronger, questions remained about who would pay, what value they would pay for, and whether the payments could cover costs. These issues had never been truly resolved.

But companies like OpenClaw have changed this narrative.

Because users are no longer just buying one-time Q&A, but a continuous digital workforce. Token consumption is no longer just a “chatting cost,” but becomes a production cost that enterprises are willing to pay for efficiency improvements.

Recently, many companies have begun ending free public testing and shifting to paid, on-demand commercial services, even signaling price increases.

The key point is that the market has already begun accepting the right to charge for high-quality tokens, rather than “who raised the price how much.” This is more important than simply increasing call volume.

For example: On February 12, Zhipu released the GLM-5 and increased the Coding Plan package price by at least 30%. On March 16, with the release of GLM-5-Turbo, prices rose another 20%, for a total increase of 83% compared to GLM-4.7. Even so, demand remains strong, with call volume increasing by 400%.

Previously, DeepSeek’s commercialization was very limited (only providing API calls), which gave room for price competition among major model providers, such as the Coding Plan service.

However, after this round of funding, DeepSeek will also commercialize, aiming to usher users into an “inclusive era.”

A review of Deep Seek’s recruitment info shows that it has released seven detailed product manager roles for different model strategies (search evaluation, search reply, multimodal, safety, literary/function writing, Agent), trying to outline a path from technical capability to commercial implementation.

This overlaps quite a bit with most application scenarios of large model vendors on the market, implying that products like Coding Plan services could shift from “price hikes” back to “universal affordability.”

According to official information from DeepSeek, due to limited high-end computing power, the current throughput of the Pro service is very limited. After the Ascend 950 supernodes are mass-produced in the second half of the year, the Pro price will be significantly reduced.

The latest main models’ total parameters, activation parameters, context input, cache hit, output prices are as follows:
DeepSeekV4-Flash 51.5B 150M 1M / 384K out 1 yuan 0.2 yuan 2 yuan
Deep SeekV4-Pro 1.6T 49B 1M / 384K out 12 yuan 1 yuan 24 yuan
KimiKimi K2.6 51.5B 5.17B 256K about 6.9 yuan about 1.16 yuan about 29 yuan
Zhipu / Z.AI GLM-5 1744B 40B about 200K 0-32K 6 yuan; 32K-200K 8 yuan with cache discount 0-32K 24 yuan; 32K-200K 28 yuan
MiniMax MiniMax-M 2.7 20B 10B 204.8K 2.1 yuan
Cache hit discount 10% 8.4 yuan
Alibaba / Qwen Qwen 3.6-Plus not disclosed not disclosed 1M ≤ 256K 2 yuan; 256K-1M 8 yuan with context cache discount ≤ 256K 12 yuan; 256K-1M 48 yuan
Xiaomi MiMo-V2-Pro >1T 42B 1M / 131K out ≤ 256K about 7.25 yuan; 256K-1M about 14.5 yuan ≤ 256K about 1.45 yuan; 256K-1M about 2.9 yuan ≤ 256K about 21.75 yuan; 256K-1M about 43.5 yuan

Compared to the current prices of major model providers, considering that DeepSeek-Pro may further lower prices, DeepSeek will put other large model companies in a position where they dare not raise prices.

Because DeepSeek’s low prices are not just a sales strategy but could become a new industry price anchor. It will force peers to face a reality: model upgrades may not necessarily lead to price increases; they might even lead to price reductions.

If this expectation materializes, the entire industry’s growth narrative based on “capability enhancement—unit price increase—revenue growth” will be interrupted.

Therefore, Miantou believes that the most dangerous aspect of DeepSeek’s commercialization is not how many customers it will steal; but that it redefines large models not as a business that profits from technical premium, but ultimately as a scale, cost, and delivery efficiency business.

Premiums must be supported by performance

The return to a business model will inevitably trigger changes in valuation logic.

After DeepSeek V4 was released on April 24, MiniMax’s stock price halved from its record high of HKD 1,330; Zhipu also dropped over 9% in a single day, falling below HKD 1,000.

Goldman Sachs and other institutions attribute this sharp reaction to the fact that the V4 series API’s “cheap price” directly shattered the profit improvement expectations based on continuous price hikes for Zhipu and others, forcing the entire industry to reassess token value.

Currently, the market’s valuation of Zhipu and MiniMax heavily relies on P/S ratios and ARR multiples.

For example: Based on Minimax’s projected 2026 ARR of $150 million, the market assigns a valuation of about 200 times P/S.

This valuation is already very high, requiring explosive performance growth in the coming years to justify it.

DeepSeek has created the greatest uncertainty: model upgrades can not only raise prices but also significantly lower them. This will directly impact Minimax and Zhipu’s “volume and price increase” growth logic.

Some investors also need to see that after their market caps surged 5-8 times, their valuation factors may have already embedded overly optimistic growth expectations, making short-term volatility more intense.

It’s not just growth potential; the market also worries about the “premium” derived from scarcity.

But this window won’t stay open forever.

Post-commercialization, DeepSeek is likely to go public to retain core talent. Kimi, Jieyue Xingchen, and others have IPO plans. More listings will disperse capital focus from Zhipu and MiniMax, and their “premiums” will be discounted.

Once DeepSeek completes a $51.5 billion valuation round of financing with backing from national funds, its “must-buy” status in capital narratives will be solidified, further compressing valuation elasticity for Zhipu and MiniMax.

In the short term, Minimax and Zhipu’s growth expectations based on token price hikes face pressure from DeepSeek; in the medium term, their “valuation premiums” due to scarcity will also be discounted. The remaining path for Minimax and Zhipu is to prove their growth through performance.

Until new performance guidance emerges, Miantou believes that the valuations of Minimax and Zhipu will be suppressed, and the key disruptive factor is DeepSeek.