Kimi K3 has not yet been open-sourced, prompting an overseas reassessment of Chinese AI

TL;DR

• After the release of Kimi K3, overseas discussions shifted from model capabilities to Yang Zhilin returning to China to start a business and the attractiveness of the US to talent.
• Vinod Khosla pointed to US immigration policy, while Russ Salakhutdinov stated that Yang Zhilin had opportunities to stay in the US but actively chose to return to China to start a business.
• Related targets: Alibaba, Tencent, Microsoft, Nvidia. Indirectly related to OpenAI, Anthropic, Hugging Face, and AI narrative assets.

After the release of Kimi K3 in mid-July, overseas tech circles quickly shifted discussions from parameters and leaderboards to a more easily disseminated question: Why didn't the US retain Yang Zhilin?

This question attracts market attention, not only because of the founder's typical background. Kimi K3 announced it would take an open-weight route, with the official team stating that the complete model weights would be opened before July 27, 2026. For enterprises, this relates to the ability to deploy models in private environments and control data and costs. For investors, it challenges the commercial moats of closed-source model companies like OpenAI and Anthropic.

The controversy heated up when two named parties provided different explanations. Vinod Khosla, founder of Khosla Ventures, used the success of Kimi K3 to criticize US immigration policy, arguing that America is scaring away top global AI talent. Yang Zhilin's advisor at CMU, Russ Salakhutdinov, subsequently countered on X, stating that Yang had many opportunities to stay in the US, including contact from Apple executives, but ultimately chose to return to China to start a business.

The value of Kimi K3 lies not in proving that Chinese AI has comprehensively taken the lead, nor in proving that the US has lost the talent war. It brings a more practical question to the surface: open models, entrepreneurial environments, and talent choices are changing the pricing benchmark for global AI.

K3 Makes Overseas Markets Take Chinese Open Models Seriously

Kimi K3 initially attracted attention because it achieved visible results on high-value tasks. According to the Kimi official blog, K3 uses a MoE (Mixture of Experts) architecture with 2.8 trillion parameters, activating 16 out of 896 experts per token. Simply put, the model scale is large, but only a portion of its capabilities are invoked each time to balance performance and inference costs.

The competition in large models is no longer just about chat experience. The market cares more about whether AI can write code, break down multi-step tasks, plan for the long term, and correct errors – essentially, agent tasks. These scenarios are closer to software development, enterprise automation, and future workflow gateways.

Kimi officially stated that K3 supports million-token contexts and native vision capabilities, performing competitively in some coding and agent benchmarks. However, boundaries must be maintained. The company's own statements mentioned that K3 still overall lags behind some of the most advanced overseas closed-source models, and some benchmark results require more independent replication.

Even so, K3 has been sufficient to prompt overseas developers and investment circles to reassess Chinese open models. In the past, many Chinese models were viewed as low-cost alternatives. Now, the market is beginning to discuss whether they can rank highly in scenarios with greater commercial value, such as programming, agents, and long contexts.

Open weights amplify this signal. Closed-source models are more like cloud services, accessed via API calls. Open weights allow enterprises to obtain the weights and deploy and adapt the model locally or in private environments. For companies concerned about data leaks, cost overruns, or vendor lock-in, this is primarily a procurement choice, not a technological stance.

Yang Zhilin's Case Cannot Be Simplified as America Losing Him

Yang Zhilin's story is easily placed within the narrative of American talent anxiety. Public information shows he completed his undergraduate degree at Tsinghua University, earned his PhD from CMU, worked at Google Brain and Meta AI, and after returning to China, founded Moonshot AI, with support from Alibaba, Tencent, and others.

Khosla's assessment touches on a long-standing pain point in the US tech ecosystem. H-1B visas, green cards, and geopolitical restrictions make the path to staying in the US more uncertain for international students. Top labs are highly dependent on global talent; the more friction in US policies, the more likely marginal talent flows will change.

However, Russ Salakhutdinov's statement altered the causal chain in this specific case. As Yang's advisor, he provided a perspective close to the individual involved: Yang Zhilin did not lack opportunities to stay in the US but felt he would regret it for life if he didn't return to China to start a business. Attributing Kimi K3 directly to a brain drain caused by US immigration policy lacks sufficient evidence.

A more fitting explanation is that two forces exist simultaneously. The US immigration and international exchange environment indeed imposes more uncertainty on foreign talent. Simultaneously, entrepreneurial opportunities in China, capital support, and the open model route are creating a sufficiently strong pull. The former explains America's anxiety, while the latter explains why this choice isn't necessarily just a passive departure.

For investors, a single case cannot prove that America is systemically losing the talent war. But it is sufficient to show that the optimal choice for top AI talent is no longer defaulting to staying in a big US tech company or starting a business within the US university system.

Open Weights Challenge the Commercial Control of Closed-Source Models

The greater significance of Kimi K3 is that it connects the talent narrative with the model strategy narrative. A returning entrepreneur created an open model with global relevance, so overseas discussions naturally shift from who trained the model to who controls the future distribution and deployment of AI.

The core advantages of closed-source model companies remain their strongest model capabilities, unified APIs, developer ecosystems, and enterprise client relationships. Once open-weight models approach competitive levels in scenarios like coding, agents, and long contexts, they will change the negotiation structure for enterprises. Companies may not completely migrate away from closed-source models, but they can use open models to replace some workloads, lower costs, and retain more data sovereignty.

Tencent's Hunyuan Hy3, Alibaba's Qwen, and overseas open models are being viewed along the same thread because enterprises are searching for a second infrastructure stack. Tencent's Q1 report mentioned that Hy3 preview has been one of the most widely used models on OpenRouter since April 28, 2026.

These cases do not prove that open weights have defeated closed source, but they do illustrate the growing concern among developers and enterprises about data control rights. Microsoft CEO Satya Nadella has warned enterprises about the risks of providing data to proprietary models, while Hugging Face has long advocated that open models can reduce power centralization. They are not endorsing a specific Chinese model, but reflecting the genuine concerns of enterprise clients.

For asset pricing, this will affect two types of narratives. Companies like Alibaba and Tencent, which have models, cloud services, and application distribution, may benefit from the expansion of the open model ecosystem. Closed-source frontier companies like OpenAI and Anthropic still possess capability advantages, but their valuation narrative needs to continuously prove that their leading edge, enterprise stickiness, and security/compliance capabilities are sufficient to offset the catch-up by open models.

Adoption Rates Will Determine the Slope of Re-evaluation

Kimi K3 currently provides a strong early signal, not yet a realized industry inflection point. Leading scores on specific benchmarks, a shift in overseas public opinion, and the announcement of an open-weight plan indicate that the market is beginning to search for a new frame of reference. However, these do not yet equate to enterprise revenue, long-term retention, or production-grade replacement.

What needs verification is whether open-weight models can transition from developer trials to enterprise deployment. Whether enterprises are willing to migrate core workflows to models like Kimi K3, Hy3, or Qwen depends on stability, inference costs, security audits, toolchain compatibility, and service support, not just leaderboard rankings.

The same applies to talent. Yang Zhilin has proven that returning to China to start a business can yield globally visible results. However, whether this will lead to a larger-scale talent migration depends on whether US policies adjust, whether Chinese companies can continuously provide world-class research environments, and the real choices of the next batch of top PhDs.

The most important aspect of Kimi K3 for investors is that it makes old assumptions less stable. Top-tier AI capabilities may not be defined solely by closed-source giants, and top-tier talent may not achieve their maximum potential only within the US system. Leaderboard hype will fade, but whether the complete weights are released as planned, whether enterprises truly adopt the model, and whether developers stay, will determine how far this re-evaluation can go.