TL;DR

• After the release of Kimi K3, overseas discussions shifted from model capabilities to Yang Zhiling's return to China to start a business and the US's talent attraction.
• Vinod Khosla pointed to US immigration policies, while Russ Salakhutdinov claimed that Yang Zhiling had opportunities to stay in the US, but chose to return home to start a business.
• Related entities: Alibaba, Tencent, Microsoft, Nvidia. Indirectly related to OpenAI, Anthropic, Hugging Face, and AI narrative assets.

After the release of Kimi K3 in mid-July, the overseas tech community quickly shifted discussions from parameters and rankings to a more easily spread question: Why didn’t Yang Zhiling stay in the US?

This question will draw attention from the market, not only because the founder's background is typical. Kimi K3 announced a move towards open-weight models, with the official statement claiming that the complete model weights will be made available before July 27, 2026. For enterprises, this is about whether they can deploy models, control data, and manage costs in private environments. For investors, it will touch on the commercial barriers of closed-source model companies like OpenAI and Anthropic.

The controversy heated up because two named parties provided different explanations. Vinod Khosla, founder of Khosla Ventures, criticized US immigration policies in light of Kimi K3's success, believing that the US is scaring away top AI talent globally. Yang Zhiling's mentor at CMU, Russ Salakhutdinov, countered on X that Yang Zhiling had many opportunities to stay in the US, including contacts with Apple executives, and ultimately chose to return home to start a business.

The value of Kimi K3 lies not in proving that Chinese AI has fully taken the lead, nor in proving that the US is losing the talent war. It brings a more realistic issue to the forefront: the open model, entrepreneurial environment, and talent choices are changing the global pricing references for AI.

K3 Makes the Overseas Community Take China’s Open Models Seriously

Kimi K3 first gained attention because it delivered visible results on high-value tasks. The official Kimi blog stated that K3 adopts a 28 trillion parameter MoE (Mixture of Experts) architecture, activating 16 out of 896 experts at a time. In simple terms, the model is very large, but only a portion of its capabilities is called upon at any given time, balancing performance and inference costs.

The competition in large models is no longer solely focused on chat experiences. The market is more concerned with whether AI can write code, break down multi-step tasks, plan long-term, and correct errors, which represents Agent tasks (autonomously completing multi-step work). These scenarios are closer to software development, enterprise automation, and future workflow portals.

Kimi stated that K3 supports millions of contexts and native visual capabilities, performing well in some coding and Agent evaluations. Boundaries must also be retained. The company's own statements mention that K3 overall still lags behind some of the most advanced closed-source models overseas, and some ranking results need more independent verification.

Even so, K3 is already sufficient for overseas developers and investors to re-evaluate China's open models. In the past, many Chinese models were viewed as low-cost substitutes. Now, the market is starting to discuss whether they can rank highly in scenarios with higher business value, such as programming, Agents, and long-context tasks.

Open weights will amplify this signal. Closed-source models are more like cloud services, where users call through interfaces. Open weights allow companies to deploy and adapt models in local or private environments after obtaining the weights. For businesses concerned about data leakage, uncontrolled costs, or vendor lock-in, this is primarily a procurement choice, not a technical stance.

Yang Zhiling’s Case Cannot Be Simplified to the US Losing Him

Yang Zhiling's experience is easily placed within the context of US talent anxiety. Public information shows that he graduated from Tsinghua, obtained a Ph.D. from CMU, had worked at Google Brain and Meta AI, and after returning to China, founded Moonshot AI, receiving support from Alibaba, Tencent, and others.

Khosla's judgment captures a long-standing pain point in the US tech community. H-1B visas, green cards, and geopolitical restrictions have made the path for international students to remain in the US more uncertain. Top laboratories rely heavily on global talent, and the more friction there is in US policies, the more likely the marginal talent flow will change.

However, Russ Salakhutdinov's remarks change the causal chain of this case. As Yang Zhiling's mentor, he provides a perspective close to the individual involved: Yang Zhiling did not lack opportunities to stay in the US, but believed that if he did not return to start a business, he would regret it for life. To directly attribute the situation of Kimi K3 to US immigration policies causing talent loss is not supported by enough evidence.

A more appropriate explanation is that both forces exist simultaneously. The US immigration and international exchange environment indeed makes foreign talent bear more uncertainty. The local entrepreneurial opportunities, capital support, and open model paths in China also create a sufficiently strong attraction. The former explains why the US is anxious, while the latter explains why this choice may not just be a passive departure.

For investors, a single case cannot conclude that the US is systematically losing the talent war. However, it suffices to illustrate that the optimal choices for top AI talent no longer automatically default to staying at major companies in the US or starting businesses within the US academic system.

Open Weights Challenge Closed Source Commercial Control

The larger significance of Kimi K3 is that it connects the narratives of talent and model routes. A returnee entrepreneur has created an open model with global discussion significance, naturally leading overseas discussions from who trained the model to who controls the future distribution and deployment of AI.

The core advantage of closed-source model companies remains the strongest model capabilities, unified APIs, developer ecosystems, and relationships with enterprise clients. Once open weight models approach in scenarios such as programming, Agents, and long contexts, it will change the negotiation structure for enterprises. Companies may not entirely migrate away from closed-source models, but they can replace parts of the workload with open models, reduce costs, and retain more data sovereignty.

Tencent's Hongyuan Hy3, Alibaba's Qianwen, and overseas open models are being placed on the same thread because companies are looking for a second set of infrastructure. Tencent once mentioned in its quarterly report that the Hy3 preview will become one of the most widely used models on OpenRouter starting from April 28, 2026.

These cases cannot prove that open weights have defeated closed sources, but they can indicate that developers and companies' concerns about data control are rising. Microsoft CEO Satya Nadella has warned companies about the risks of providing data to proprietary models, while Hugging Face has long advocated that open models can reduce power concentration. They are not endorsing a specific Chinese model, but reflecting the genuine concerns of enterprise clients.

For asset pricing, this will influence two types of narratives. Companies like Alibaba and Tencent, which own models, cloud, and application distribution, may benefit from the expansion of the open model ecosystem. Closed-source frontier companies like OpenAI and Anthropic still have competitive advantages, but their valuation narratives need to be continuously proven, with leadership margins, enterprise stickiness, and security compliance capabilities being sufficient to offset the chase of open models.

Adoption Rates Will Determine Revaluation Slopes

Kimi K3 currently provides a strong early signal, but it is not yet an industry inflection point that has been realized. Specific evaluation leads, shifts in overseas public opinion, and the release of open weight plans indicate that the market is beginning to seek new reference points, but these cannot equate to enterprise revenues, long-term retention, and production-grade alternatives.

What needs to be verified is whether open weight models can transition from developer trials to enterprise deployment. Whether companies are willing to migrate core workflows to models like Kimi K3, Hy3, or Qianwen depends on stability, inference costs, security audits, toolchain adaptability, and service support, rather than just rankings.

The talent side is similarly affected. Yang Zhiling has demonstrated that returning to start a business can yield globally visible results, but whether it results in a larger-scale talent migration will depend on whether US policies adjust, whether Chinese companies can continue to provide world-class research environments, and the real choices of the next generation of top doctoral candidates.

The aspect of Kimi K3 that investors should pay the most attention to is that it makes previous judgments less solid. Top AI capabilities are not necessarily defined only by closed-source giants, and top talent may not only maximize their potential within the US system. As interest in rankings wanes, whether the complete weights can be opened as planned, whether enterprises genuinely adopt them, and whether developers continue to stay will determine how far this revaluation can go.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。