Moonshot AI and DeepSeek Drive China’s AI Valuation Surge with High P/ARR Multiples

Author: MD

Produced by Bright Company

Earlier, MoonShot AI released its latest model, Kimi K3, which has garnered significant attention from the capital market due to its leading performance across various benchmarks. According to MoonShot AI’s post on X, K3 is a multimodal model with 2.8 trillion parameters, supporting a context length of 1 million tokens, featuring native visual capabilities, and incorporating multiple innovative technologies.

Artificial Analysis reports that Kimi K3 scores 57 on the Artificial Analysis Intelligence Index. Its intelligence level is comparable to Opus 4.8 and GPT-5.5, but still lags behind Fable 5 and GPT-5.6 Sol. Moonshot AI has indicated plans to release the weights of this 2.8T parameter model, which would make it a leading open-source model.

On the AI Arena Code leaderboard, K3 even outperforms Anthropic’s Fable 5, Gpt5.6-sol, and Zhipu’s GLM-5.2 (max). Following its most recent funding round, Moonshot AI’s post-money valuation reached approximately $31.5 billion; however, based solely on performance rankings, this valuation is significantly lower than those of Anthropic and OpenAI.

Another Chinese AI company attracting significant attention in the primary market is DeepSeek. According to The Information and Bloomberg, citing sources familiar with the matter, DeepSeek is advancing its second funding round at a $74 billion valuation, just one month after its first round was priced at a $54.3 billion pre-money valuation.

From a valuation perspective, DeepSeek and Moonshot AI, two private-market companies, are approaching or even surpassing comparable public-market counterparts. Perhaps due to constraints in private-market liquidity and information transparency, Zhipu and MiniMax have undergone over half a year of continuous public-market pricing, resulting in a well-established narrative consensus—whether rewarded or penalized by the market. In contrast, while DeepSeek and Moonshot AI’s valuations are more aggressive, the narratives surrounding them have not yet fully crystallized.

Kimi-K3 is currently ranked first on the Arena AI Code leaderboard (source: Arena AI)

The listings of Zhipu (02513.HK) and MiniMax (00100.HK) have established a valuation benchmark for China’s independent model vendors. The clarity of their narrative and their ARR growth rate determine the premium level when evaluated using P/ARR.

Since its Hong Kong stock market listing in January, Zhipu's stock price has surged over 1,000%, with a market capitalization of approximately $92 billion. This valuation is supported by an exceptionally steep revenue curve: ARR increased from $67 million in January to $1 billion in July—a 15-fold growth in seven months, achieving its annual target six months ahead of schedule. Multiple investment banks have noted that this growth rate exceeds Anthropic’s previous record of growing ARR from $100 million to $1 billion over 15 months. According to the latest information from 36Kr, Zhipu’s ARR has already reached $1 billion, far outpacing the earlier expectation of reaching $1 billion by year-end.

MiniMax's market capitalization has declined by approximately 75% from its IPO peak of around HK$410 billion to about HK$81 billion, implying a P/ARR multiple of roughly 13x. “Bright Company” has recently discussed the challenges surrounding MiniMax’s valuation narrative: as the industry’s focus shifts toward coding and agents, MiniMax’s multi-pronged strategy—centered on “multimodal capabilities and the C+B dual engine”—has become a target of market skepticism.

DeepSeek's seed round was signed at the end of May: exceeding RMB 50 billion (approximately $7.4 billion), marking the largest single round in Chinese AI large model history.

According to The Information, the company clearly informed investors during its initial roadshow that it had no intention of commercializing the model, was focused on cutting-edge research and development, and had no defined IPO timeline. Before the first round of funding was officially completed, Liang Wenheng had already begun planning the second round. The terms of this round underwent a fundamental shift: the valuation rose to $74 billion, an increase of approximately 36% from the first round.

The Information also reported that DeepSeek plans to file its application this year and go public on the Shanghai Stock Exchange's Science and Technology Innovation Board in 2027. The primary driver behind this move is the massive computing costs required for model development.

The fundraising pace of Moonshot is equally extraordinary: its post-money valuation was only $4.3 billion by the end of 2025; in May 2026, it raised $2 billion, bringing its post-money valuation to $20 billion; and on June 30, it launched a new round with a pre-money valuation of $31.5 billion—more than a sevenfold increase in valuation within six months.

Both DeepSeek and Moonshot have achieved high valuations in the primary market, but the underlying logic supporting these valuations appears to differ. DeepSeek relies more on usage volume, technical influence, and efficiency engineering, while Moonshot seems to be replicating Anthropic’s early revenue trajectory.

According to sources familiar with the matter who spoke to The Information, DeepSeek’s recent annualized revenue has reached $400 million to $500 million, primarily from API services; at the upper end of $500 million, a $74 billion valuation equates to approximately 148x P/ARR—making it the most expensive among China’s four leading companies and significantly surpassing many U.S. AI startups.

Analyzing this, the reasons supporting this multiple may stem from three aspects.

First is the number of requests.

According to a report by China Merchants International on July 3, as of the week ending June 22, DeepSeek ranked first globally on the OpenRouter platform with 6.7 trillion tokens, surpassing Anthropic’s 4.25 trillion. In the second quarter of 2026, total API calls on OpenRouter increased from approximately 21 trillion to 46.66 trillion tokens, doubling quarter-over-quarter, with growth primarily driven by Chinese open-source models, as combined API usage by Chinese vendors has now exceeded that of U.S. vendors.

Source: OpenRouter

According to the latest weekly market share rankings from OpenRouter, DeepSeek ranks second in market share, behind only Google.

Second is efficiency engineering. On million-token context scenarios, V4 requires only 1/20 the computation per token compared to the previous generation. V4-Pro delivers performance comparable to top-tier proprietary models, with China Merchants Securities evaluating it as approaching Claude’s non-reasoning mode level.

Third is the validated UE. According to Bloomberg, despite its API pricing being only a fraction of OpenAI’s and Anthropic’s, V4 has maintained a gross profit margin above 50%—proving that low pricing does not equate to low margins, a key data point validating the “compute efficiency” narrative financially.

In addition, on the strategic front, the company launched V4 officially in mid-July and introduced a peak-valley pricing mechanism—its first pricing initiative with clear commercial intent. V4 is deeply optimized for domestic computing power platforms such as Huawei Ascend and Cambricon, and combined with direct equity stakes from China’s national AI fund, its strategic positioning as a cornerstone of a domestic computing power closed-loop model layer is clearly evident.

Now let’s look at the dark side of the moon.

Its pre-money valuation of $31.5 billion corresponds to approximately 105x P/ARR; using the $400 million to $500 million ARR range cited by experts on the Nomura conference call, it equates to approximately 70x. With the release of Kimi K3, the market generally anticipates that the K3 model, scheduled for release in the third and fourth quarters, will open new pricing potential.

Several brokerages have compared Moonshot’s revenue trajectory to Anthropic’s early stage: increased developer usage, rising API share, overseas paid growth, and model iterations driving upward pricing pressure.

The transcript of Nomura's expert conference call on July 6 indicates that the expert expects its ARR to exceed $1 billion by year-end. However, according to publicly available information, Yang ZhiLin has never mentioned a target of "$1 billion ARR by year-end."

Compared to Zhipu’s approximately 100-fold and MiniMax’s approximately 13-fold public market valuations, the premiums for these two private companies stem from entirely different sources.

From previous valuations, Moonshot's premium appears to be more of a "financial premium" and a "performance premium."

If ARR reaches $1 billion by year-end, the 31.5 billion valuation implies a forward multiple of only about 30x—cheaper than Zhipu’s current valuation; investors are also betting on a re-rating through an IPO following Zhipu’s path. Zhipu’s last private round valuation was $13.7 billion, and its market cap once exceeded $100 billion after listing.

This premium is falsifiable and trackable, with monthly ARR data continuously validating it.

In contrast to Moonshot, DeepSeek’s premium includes non-financial components: technological influence premiums from usage volume and architectural control rights, geopolitical option value stemming from national strategic scarcity, and the “founder belief” backing provided by Liang Wenheng’s personal investment of 20 billion yuan, among others. In particular, the impact of the “DeepSeek Moment” on the revaluation of Chinese assets remains unmatched by any other vendor to date.

A static multiple of 148x cannot be justified by any cash flow model; investors are buying into the belief that it has the potential to generate Anthropic-level revenue once commercialization is fully activated.

However, as these two companies move toward an IPO, they will confront the continuous pricing mechanism of public markets, which may subject their valuation logic to changes—or even pressure.

First, test the quarterly growth rate. The primary market has been willing to pay for "narratives," while the secondary market is increasingly paying for "execution." Initially, MiniMax was valued at approximately 65x P/ARR, but as the industry narrative shifted toward agents and its monetization path came under scrutiny, its valuation was reduced by three-quarters.

DeepSeek, positioned in the mid-tier price segment, experienced a 95% deflation in inference pricing during the second quarter. According to China Merchants International data, the optimal invocation price for the intelligence index range of 40–50 dropped from $1.2 per million tokens in March to $0.058 in June. Whether the leading invocation volume can continue to translate into revenue growth will be a mandatory question in every quarterly report.

The second aspect is the lock-up expiration pressure on the supply side. Zhipu and MiniMax saw price declines of 8.5% and 22.5%, respectively, during their respective lock-up expiration weeks, followed by aggregate placements exceeding HK$40 billion—this “lock-up expiration–placement” supply shock is likewise awaiting DeepSeek and Moonshot AI upon their future listings.

Finally, there is the unification of valuation metrics—for example, "token monetization rate" will become a directly comparable metric.

Source: Compiled from public information; Bright Company (valuation and market capitalization as of close on July 16, excluding the latest rumored ARR from Zhipu)

In addition, at the narrative level, a key benchmark for Chinese model companies at this stage is Anthropic.

Anthropic completed a $6.5 billion Series H round on May 28 with a post-money valuation of $965 billion and has confidentially filed for an IPO. Its ARR increased from $9 billion at the end of 2025 to $47 billion in May 2026.

According to Jefferies, citing SemiAnalysis, its July ARR will exceed $60 billion, with a net revenue retention rate of approximately 500%, an API gross margin exceeding 80%, and an expected GAAP operating profit of over $1 billion in the third quarter.

At a $47 billion ARR, the $965 billion valuation implies a roughly 20x P/ARR. OpenAI has also confidentially filed for an IPO, with an ARR of approximately $25 billion—once the undisputed leader, its revenue scale has now been surpassed by Anthropic.

Anthropic’s valuation multiple has compressed from 184x at $100 million in ARR to approximately 20x today. This means DeepSeek’s 148x and Moonshot’s 105x multiples are essentially prepaying for revenue growth that has not yet materialized—downward multiple pressure is inevitable; the only variable is whether ARR growth can outpace multiple contraction. After Zhipu reached $1 billion in ARR in July, its current multiple has automatically adjusted significantly from its year-beginning level.

Additionally, on the geopolitical competition front, Jefferies’ report on July 13 noted that as U.S. manufacturers gain access to more next-generation computing power in the second half of the year, combined with anti-distillation mechanisms, the capability gap between Chinese and U.S. models could widen again. The entire valuation logic for Chinese models—ranging from Zhipu’s 100x to DeepSeek’s 148x—implicitly assumes that the capability gap will continue to narrow; at least so far, the release of K3 has somewhat offset the assessment that the capability gap is widening.

Currently, both Moonshot and DeepSeek are showing promising results in terms of funding valuation and model performance.