Kimi K3’s release has intensified global attention on China’s artificial intelligence sector, drawing more than five million views in online discussions across U.S. technology circles and renewing debate over whether Chinese AI labs are narrowing the gap with American frontier-model companies.
The strongest reaction has centered on two founders whose different strategies now define much of China’s AI race: Liang Wenfeng, the founder of DeepSeek, and Yang Zhilin, the founder of Moonshot AI. Liang shocked global markets in early 2025 with DeepSeek’s R1 reasoning model, while Yang pushed Moonshot’s Kimi series into mainstream use through long-context language tools and, more recently, the Kimi K3 release.
Together, their companies have changed expectations around model cost, open-source development, talent competition and the future of AI-linked digital markets. Once seen mainly as followers of U.S. labs, Chinese model developers are now forcing global technology firms to respond on price, speed and accessibility.
The release of Kimi K3 has also attracted attention beyond China. Former U.S. government technology advisers have said publicly that they use Kimi tools in daily work, while technology veterans have connected the model’s rise to broader talent migration trends. Those comments have added political weight to an already intense industry debate over how easily AI expertise can move across borders and how quickly national advantages can be challenged.
A new phase in the global AI race
The latest wave of attention comes after a year and a half of abrupt shifts in AI pricing and product strategy. DeepSeek’s R1 model, released in January 2025, was widely viewed as a turning point because it delivered strong reasoning performance at a much lower cost than many traders and technology executives had expected.
The market reaction was immediate. Nvidia’s market value fell by nearly $600 billion in a single day after the R1 launch, a selloff some observers described as a modern “Sputnik moment” for the technology industry. The comparison reflected anxiety that the assumed cost barriers around frontier AI models may have been less secure than previously believed.
Moonshot AI’s Kimi series followed a different path. Rather than first becoming known for a shock to chip markets, Kimi gained traction through long-context capabilities that allowed users to process and analyze unusually large amounts of text. That function helped make Kimi one of China’s most widely used AI products and positioned Moonshot as one of the country’s most closely watched AI start-ups.
By the time Kimi K3 drew major attention in the United States, the competition between Chinese AI labs had already moved beyond benchmark scores. Companies were competing over inference costs, open-source strategy, model reliability, developer adoption and access to scarce research talent.
Liang’s trading roots and DeepSeek’s unusual structure
Liang Wenfeng’s route into artificial intelligence was unusual compared with many AI founders. Born in 1985 in Wuchuan, Guangdong, he spent about 15 years in quantitative trading before building DeepSeek into an independent AI research organization.
That background shaped DeepSeek’s early development. Instead of relying heavily on outside financiers, Liang used trading revenue to fund large computing clusters and internal research. By 2019, he had reportedly spent nearly 2 billion yuan on computing resources. By 2021, that commitment had risen to more than 10 billion yuan.
DeepSeek became independent in 2023 and kept a structure that differed from many venture-backed technology companies. It operated with a lean team of roughly 140 people, avoided outside funding and maintained a relatively flat internal organization. That approach gave researchers more freedom, but it also placed unusual pressure on the founder to keep financing the company’s infrastructure and talent needs.
The release of R1 made DeepSeek one of the most closely studied AI labs in the world. Its free and open-source approach challenged the pricing models of larger technology groups. After R1 entered the market, major Chinese technology companies reduced application programming interface fees sharply, with some cutting prices to about one yuan per million tokens.
That rapid decline in cost changed the economics of AI access. High-end reasoning systems, once limited mainly to large companies with deep budgets, became easier for independent developers and small teams to use. In some cases, advanced open-source model access fell to only a few cents per million input tokens, weakening the financial barriers that had previously protected large AI providers.
Yang’s academic path and Moonshot’s rise
Yang Zhilin followed a more traditional academic path into AI. Born in 1992 in Shantou, he studied at Tsinghua University, where he ranked at the top of his class, before earning a Ph.D. at Carnegie Mellon University with a focus on language models.
His research on long-context memory became central to Moonshot AI’s product direction. Papers associated with his work drew about 20,000 citations, helping establish the technical foundation for Kimi’s ability to handle large quantities of information in a single interaction.
Yang returned to China in 2019 and later co-founded Moonshot AI in early 2023. The company secured roughly $60 million in seed funding and quickly became one of China’s highest-profile AI start-ups. As domestic technology groups joined later financing rounds, Moonshot’s valuation rose into the multibillion-dollar range.
Kimi’s early popularity reflected strong demand for tools that could summarize documents, assist with research, support coding tasks and manage complex written information. Its long-context design gave it a clear identity in a crowded market, allowing Moonshot to stand out even as larger companies launched competing AI assistants.
But rapid growth came with high costs. In October 2024, Moonshot reportedly spent 2.2 billion yuan on advertising. In November, it spent another 2 billion yuan to maintain growth momentum. The marketing push expanded public visibility, but it also raised questions about whether user acquisition costs were sustainable.
Those concerns deepened after a shareholder arbitration tied to Yang’s earlier start-up surfaced in Hong Kong. The legal dispute added pressure at a time when China’s AI market was already becoming more competitive and far more price-sensitive.
DeepSeek’s price shock forces a strategy reset
DeepSeek’s R1 launch changed the environment for Moonshot and other Chinese AI companies almost overnight. By releasing a powerful model at no cost and opening it to developers, DeepSeek made it harder for rivals to justify expensive closed systems or marketing-heavy growth strategies.
Moonshot responded by reducing large-scale marketing and narrowing its focus to core model development. It also shifted more strongly toward open-source releases, a move that aligned the company with the broader direction of the Chinese AI sector after R1.
That reset appeared to gain traction. In July 2025, Moonshot released K2, a model with one trillion parameters. The company said it achieved strong benchmark scores, including better results than GPT-5 in several agent-related tasks. The release helped restore confidence in Moonshot’s technical direction and signaled that the company could compete through model performance rather than advertising alone.
By December 2025, Moonshot had completed a $500 million Series C funding round at a valuation of $4.3 billion. The company also reported that monthly subscription revenue had increased more than eightyfold after the release of K2.5.
Kimi K3 has now extended that recovery narrative. Its reception in U.S. technology communities suggests Moonshot is no longer being assessed only as a Chinese consumer AI product, but as a model developer with potential relevance in global professional workflows.
Talent pressure rises inside DeepSeek
DeepSeek’s success has also created internal pressures. Between 2025 and 2026, several key researchers left the company for senior roles elsewhere. That movement highlighted one of the biggest challenges facing frontier AI labs in China and abroad: keeping elite technical teams intact once their work becomes globally visible.
After years of avoiding outside capital, Liang began meeting financiers for the first time. Initial funding proposals reportedly started at 50 billion yuan before later being reduced to 15 billion yuan. Discussions were said to focus less on valuation and more on retaining staff, protecting the laboratory’s research culture and preserving organizational integrity.
That shift marked a major change for DeepSeek. The company had built much of its identity on independence, efficiency and freedom from conventional capital-market pressure. But success increased the cost of staying independent. Once a lab becomes strategically important, rivals, financiers and large technology firms all begin competing for its people.
The result is a difficult balance. DeepSeek must preserve the culture that helped produce R1 while building the financial and organizational systems needed to support a larger research agenda. That is a familiar problem for fast-growing technology companies, but it is especially acute in AI, where a small number of researchers can determine a model’s direction and commercial value.
Lower AI costs reshape digital asset markets
The collapse in AI processing costs is also influencing digital asset markets, especially sectors connected to automation, data infrastructure and decentralized computing.
Cheaper access to reasoning models means independent builders can create sophisticated automated systems without spending large sums on server hosting or proprietary AI interfaces. Small development teams can now test trading tools, market scanners, research agents and smart-contract monitoring systems from low-cost environments that were not practical a few years ago.
For cryptocurrency traders, this changes daily risk management. Machine-driven systems already account for a large share of activity across digital asset markets, and rapid access to language models and market data tools may increase that share further. Manual price tracking is becoming less competitive in markets where automated programs can scan order books, news feeds, social platforms and on-chain data in seconds.
That does not mean automation removes risk. In fact, faster tools can intensify volatility when many systems react to the same headline or liquidity signal at once. Traders who use automation still face execution risk, model error, poor data quality and unexpected market gaps. Tighter controls, clearer stop-loss rules and better monitoring are becoming more important as AI systems become easier to deploy.
The cost shift has also encouraged attention toward decentralized networks that provide machine learning hardware, shared graphics processing capacity and distributed compute markets. By mid-2025, tokens connected to decentralized GPU sharing and AI infrastructure had grown into a sector valued at about $27 billion. That growth reflected demand for alternatives to centralized cloud computing, though the sector remains highly speculative and sensitive to changes in AI demand.
Open-source acceleration changes the competitive map
Open-source development has become one of the defining forces in China’s AI sector. DeepSeek’s R1 demonstrated that open releases can rapidly reshape pricing expectations. Moonshot’s later pivot showed that companies under competitive pressure can use open frameworks to regain momentum and developer trust.
The effect is broader than China. When advanced models become cheaper and easier to adapt, developers around the world can build products faster. Corporate software teams, independent builders and research groups can experiment with systems that previously required major budgets. That opens the door to more innovation, but it also reduces the advantage once held by companies that relied mainly on controlling access to expensive models.
For large technology firms, lower pricing creates a strategic dilemma. Cutting fees can expand adoption, but it may also compress margins. Keeping prices high can protect revenue, but it risks losing developers to open-source alternatives. This tension is likely to shape AI business models through 2026.
For Chinese AI companies, the challenge is even sharper. Domestic competition is intense, and the same price cuts that attract users can make it harder to fund expensive training runs, hire top researchers and maintain infrastructure. Companies must prove that they can convert technical strength into durable revenue.
Two paths begin to converge
By mid-2026, Liang and Yang appeared to be moving toward a similar strategic middle ground despite starting from opposite directions.
Liang built DeepSeek through liquidity generated from quantitative trading and avoided outside funding for as long as possible. His model questioned the assumption that frontier AI must require enormous external capital from the beginning. But DeepSeek’s success has now pushed him toward capital markets as he tries to retain staff and sustain the lab’s ambitions.
Yang built Moonshot through academic credibility, venture-style financing and rapid product expansion. His company spent heavily to grow, then faced pressure after DeepSeek changed pricing norms. Moonshot’s response was to reduce marketing, focus on core models, embrace open-source releases and push harder on efficiency.
The two strategies now appear to be converging around the same realities: models must be cheaper to run, teams must stay small enough to move quickly, open-source ecosystems matter, and capital is still needed to compete at the frontier.
Their trajectories show that China’s AI race is not only about chips, data or government policy. It is also about whether founders can reconcile research culture, organizational discipline and financial pressure. DeepSeek proved that cost assumptions could be challenged. Moonshot showed that a company could recover from rapid spending and legal distractions by refocusing on product and technical execution.
Kimi K3’s reception suggests that the global AI race is entering a more complicated phase. The leading question is no longer simply which country has the largest models or the most chips. It is which organizations can build useful systems quickly, distribute them cheaply, retain rare talent and adapt when the economics of intelligence change overnight.
Explore how algorithmic strategies mirror China’s AI shift in our guide to TradFi vs DeFi for tech‑driven investors.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
