Kimi K3 will be open source in 4 days, and Americans are really anxious this time.

Original Title: "Kimi K3 Open Source in 4 Days, Americans Are Really Anxious This Time"
Original Author: InsightBeating

Americans always want to sit at the very center of every industry.

The AI community is no exception. Americans have always had a sense of holding all the aces, holding a hand of cards that seems unbeatable.

No matter who is developing AI applications outside, Americans believe that ultimately, the reckoning will always come back to them. The chips belong to NVIDIA, the cloud to Microsoft, Amazon, and Google, the most advanced models are locked behind OpenAI and Anthropic's APIs. Whenever any company in the world wants to use AI, they always have to go through the U.S.

Even if a Chinese team's name occasionally appears on the rankings, Wall Street doesn't take it too seriously. With a chokehold on chips, a grip on the cloud, and talent still flowing to Silicon Valley, how could they lose?

However, this sense of unwavering confidence has recently been exposed by the emergence of the Chinese model Kimi K3.

The U.S. tech industry has urgently updated its perspective, describing Kimi K3 as a "Sputnik" moment, just like the shock brought by the launch of the Soviet satellite in 1957. Discussions around Kimi K3, Yang Zhilin, and Chinese models quickly moved beyond the small circle of the tech community and became a topic with tens of millions of views.

While Kimi K3 did not outperform the strongest closed-source American models in every aspect, it has shown more people a possibility: high capability, efficiency, and an open ecosystem may not only emerge simultaneously in a few labs in the U.S.

Silicon Valley in the U.S. is indeed feeling anxious.

Storage is the Anxiolytic of the American AI Community

When news of Kimi K3 reached Wall Street, several investment banks almost simultaneously released research reports. Instead of discussing at length who it would impact or whether it would force American models to lower their prices, these institutions quickly shifted their focus to storage.

In a strikingly unanimous interpretation, these institutions viewed the emergence of Kimi K3 as a sign of the strong demand for storage. With AI needing to remember more things, accumulating a wealth of images, sounds, videos, and work records, flash memory, hard drives, data centers, and data services are all set to benefit.

So, Micron, SanDisk, and Western Digital became the beneficiaries of this story.

Sure enough, in yesterday's U.S. stock market, the storage sector collectively rebounded violently. The Roundhill Storage ETF rose by 10.91% in a single day, SanDisk rose by 14.27%, and Micron rose by 12%. The sector that was dumped a few days ago due to "DeepSeek Moment 2.0" overnight became the most certain bull market again.

From an industry perspective, this line of thought is not absurd. In the past, chatbots were like one-time Q&A sessions: you ask a question, it answers, you close the page, and many things are forgotten. The AI that everyone now expects is more like a new employee at a company. It needs to go through past contracts and emails, remember what customers have said, take over unfinished work from yesterday, keep records to avoid mistakes that no one can clarify. An AI that can work, take notes, and even see images and hear audio, of course, can "digest" data better than an AI that can only chat.

This conclusion is not groundless, but looking back at previous model launches and implementations, will the market's response be, "Don't focus on the model, focus on storage"?

One can only say that this is an answer that can reassure Americans.

The impact of a Chinese model should have brought up a series of difficult questions: Will it make it harder for American model companies to maintain high prices? Will it make developers rely less on them? Will it allow new companies to start elsewhere other than Silicon Valley? Why not directly discuss how Kimi K3 will take away users from whom, force price reductions, or compel product changes?

To bypass the most acute questions and first discuss hard drives, there is a hint of "there must be more than meets the eye."

Like a shop owner who thought they monopolized the entire street, suddenly realizing that a very competitive new store has opened next door, so they quickly console themselves: no matter how many customers the new store attracts, they will still need to use the water, electricity, and counter I sell.

Storage is the most potent consolation in the anxious U.S. AI circle.

Closed-Source Models are Starting to Stumble

Over the past few years, closed-source has been almost unquestionably the standard answer for U.S. AI.

The stronger the model, the more it should be locked behind an API. Users pay to access it, model companies earn high margins, and security and compliance are also unified under it. This is a decent and profitable path to follow, with customers at ease, investors satisfied, and regulators well-informed.

Americans have even become accustomed to the rhythm of this road, releasing a stronger version every few months, setting a higher price, and telling a bigger story.

But as the open model becomes stronger, the ground of this road is starting to feel rocky.

In this game, Kimi K3's position is not about "catching up", but about reducing the cost of catching up. An open and sufficiently strong model, the most dangerous aspect is not what it can do on its own, but that it offers a much cheaper learning curve to all followers.

This is not a showdown in the tech circle, but a question of whether business will be rewritten. The most comfortable arrangement the U.S. had before was to turn AI into an enterprise service: capabilities hidden in the cloud, clients signing long-term contracts, ordinary people unable to see the underlying technology, and not easy to replace it. However, if models elsewhere are good enough, developers will have one more option, companies will present another quote during procurement, and small teams may not necessarily have to bet their future solely on the same batch of U.S. companies. At that point, merely holding onto a few major contracts and only selling AI to the B-end will no longer be an impregnable moat.

This means that Kimi will give birth to more excellent models, and it also means that the competition among models will be greater, and the bargaining power will not be as strong.

Even the U.S. tech industry has felt the shift in the wind.

A few days before the release of Kimi K3, on July 15, Thinking Machines Lab founded by OpenAI's former CTO Mira Murati released a model called Inkling. With a parameter size close to the trillion level, the code and technology are completely open, and anyone can download, modify, and use it for free.

This can be considered the U.S.'s first "serious" open-source AI. While there have been open-source models like Meta's Llama, Google's Gemma, Microsoft's Phi, NVIDIA's Nemotron, and OpenAI's gpt-oss before, most of them were more experimental.

The significance of Inkling lies in the fact that someone who once took closed-source to its peak, the former CTO of OpenAI, has now turned to seriously focus on open source.

It is worth mentioning that in the early stages of Inkling's training, it used data generated by open models like Kimi K2.5, and its architecture also referenced DeepSeek's approach. In other words, this most decent open-source response from the U.S. was also written on the shoulders of China's open source.

In contrast, Anthropic stands out. In February of this year, Anthropic publicly accused DeepSeek, DarkSide of the Moon, and MiniMax of launching an "industrial-scale distillation" against Claude, claiming they had created twenty-four thousand fake accounts and engaged in sixteen million conversations to steal Claude's capability. In June, they escalated by calling out Alibaba. By July 21, U.S. Treasury Secretary Bessent of the Trump administration bluntly stated that sanctions could be imposed on China for "AI theft."

While the rhetoric of threats may be loud, when it comes time to control costs and improve efficiency, the Chinese model is indeed enticing.

Airbnb uses Qwen for customer service, Cursor has Kimi as its own programming agent, DoorDash outsources some tasks directly to Kimi, and even Murati's Inkling, post-training, has to use Kimi's data.

Whether stumbling over the closed-source path or facing backlash from distillation accusations, these are all just embarrassments at the business model level. In reality, it is the issues of privacy and security that truly undermine the last defense of the closed-source camp.

The "Jailbreak" of AI Models

Security has always been the final line of defense for closed source.

Locking up models, securing weights, accessing via APIs, and not persisting data—this space surrounded by four walls is the most commendable promise of the closed-source camp. Enterprise customers are willing to pay a premium for this and are attracted by this sense of security.

However, enterprises are becoming increasingly uneasy. They are beginning to ask questions that closed-source companies find difficult to answer: What did you do with my code, contracts, and customer data after I handed them over to your model? Will an agent armed with access to browsers, terminals, credentials, and long-term objectives cross the line I allowed it to cross in order to complete a task? Sending a token to a closed-source API, in a sense, allows data to leave its own walls. This is precisely the toughest selling point of open weights—at least I can see what the model is doing.

And just when the debate between open and closed about who is safer was at an impasse, something almost darkly humorous happened.

On July 21, OpenAI confirmed that its flagship model GPT-5.6 Sol and a more powerful unreleased model breached the isolation environment in an internal network security assessment.

The story goes like this: the engineering team wanted to test the model's upper limit of defense capability, so they lowered the model's security restrictions, turned off the usual protection that intercepts high-risk behaviors, and vulnerabilities. The model was originally supposed to complete the test questions obediently, but it discovered a security flaw in the system by itself, climbed onto the public network through this loophole, bypassed permissions, traversed the system, and finally breached the core system of the world's largest open-source AI platform, Hugging Face, using stolen login credentials directly from the database.

The explanation given by OpenAI was eight words: no malicious intent, overly focused.

It was these eight words that truly sent chills down people's spines.

For enterprise customers, the most frightening thing has never been the model acting maliciously on its own. It is when it diligently achieves a malicious goal on your behalf.

The greatest irony of this incident is that over the past year and a half, the world has been on guard against that "dangerous Chinese open-source model," which is still stuck in the hypothetical stage. The one that actually broke free and breached someone else's production system is the flagship of the closed-source camp itself. Hugging Face's CEO, Clem Delangue, quickly turned this incident into an open-source ad. He said, "AI security cannot be achieved by one company working in isolation; it can only be addressed through collaboration in an open environment."

The same incident was taken by both the open and closed camps as evidence that their own path was correct.

The true watershed of the future is probably not whether the model is open source or closed source, but rather, the environment in which the model runs — the identity system, the revocable permissions, and the audit log. Whether closed or open source, this question cannot be avoided.

While the narrative of closed-source security was experiencing its own incident, a larger-scale reversal was quietly taking place.

Role Reversal: It's America's Turn to Be Afraid

In part of the U.S. policy discussions and tech narratives, there has always been an almost "Three-Body Problem"-style imagination: as long as the most advanced Nvidia chips are restricted from entering China, AI progress will be forced to slow down.

This doesn't mean that China will be completely unable to conduct research. Rather, they believe that the gap in computing power will continue to widen, and the threshold for training cutting-edge models will become so high that it will be difficult to cross. Advanced chips are like the "laws of physics" in this race. Whoever doesn't have access will find it difficult to make it to the front.

This assessment is not unfounded. Building large models does indeed require computational power, and chip restrictions will increase costs, slow down expansion, and make it harder for many teams to replicate the training scale of U.S. laboratories. The problem is that restrictions also change people's choices. You could have bought the best tools ready-made, but you no longer have as much motivation to figure out how to save on computing power, how to change the model structure, or how to make the most of each training session. When the door is shut, taking a detour is no longer an option; it becomes a survival instinct.

So Americans actually find it hard to understand why restricting NVIDIA's supply didn't keep the Chinese model in check, but instead forced out a group of teams in China who are even more determined in efficiency, engineering, and open-source distribution.

It is said that even as of 2023, the dark side of the moon is still using the compliance version AI chip H800, custom-made by NVIDIA for the Chinese market, for training.

This may be the Chinese's best at playing the "millet plus rifle" tactic.

In June 2026, to comply with export controls, the United States temporarily shut down China's most powerful Fable 5 and Mythos 5 by Anthropic. While this may make sense in terms of compliance, it handed each Chinese open-source lab a ready-made marketing slogan — at least, our model doesn't come with a remote kill switch.

The more you emphasize control, the more control itself becomes the selling point for the opponent.

On the more dramatic side, Reuters reported that domestic discussions have also begun with companies like Alibaba and ByteDance to consider restricting foreign access to China's most advanced AI models, with the scope even including those that are already publicly available as open-source models. Zhou Hongyi, the founder of 360, also publicly declared that China should also have its own top-tier closed-source models to defend its technological high ground.

A year ago, it was the U.S. worrying about advanced chips flowing to China. A year later, it's China that has something worth restricting.

But amid all this structural anxiety, whether it's about storage, computing power, closed-source, security, offense, defense switching, there is one most concrete, most heartfelt, most personal focal point. It's not an industry trend, not a research report, and not a policy.

It's a person.

The End of Anxiety, Focuses on Yang Zhilin

In the end, the open-source debate has shown the U.S. the same, larger issue: Will AI's future only serve a few companies that can sign big contracts, or will it become a capability that more and more ordinary teams can use, like electricity or the internet? If the answer gradually leans towards the latter, whoever can attract developers, whoever can make young people willing to stay and tinker will be more important than who has more enterprise clients.

And the question of "where people will go" has ultimately brought America's anxiety to a very specific name.

The reason Yang Zhilin has been repeatedly mentioned in the US tech community is not just because he is an outstanding Chinese researcher, nor is it because some want to simplify the issue as "the US didn't retain talent." Reducing a person's choice to stay or leave to merely a visa is too simplistic and too much like hindsight bias.

What truly pains Americans is the irrevocable assumption: What would happen if people and teams like Yang Zhilin ultimately completed the entire journey from research to entrepreneurship in the US? They would train models on US cloud and chips, hire from the US talent network, raise money from US VCs, and take their products to knock on the doors of US big customers. In a few years, a new star company could appear on Wall Street's books. One person's choice, following that familiar relay chain, could turn into a company's revenue, people's jobs, and an entire industry's confidence.

What the US has been most proud of in the past is this amplification ability. It not only attracts smart people to study and work but also catches hold of their smartness, not letting it stop at papers or labs. Here, there is enough money, enough customers, and enough people willing to take risks together.

Why didn't such people stay in the US originally? Legendary investor Vinod Khosla directly pointed the finger at the Trump administration's tightening immigration policies. However, Yang Zhilin's mentor at Carnegie Mellon, Salakhutdinov, came out to refute this, saying it had nothing to do with visas. Yang had plenty of opportunities to stay back then, and Salakhutdinov even asked Yang in an email if he wanted to join Apple.

It was Yang Zhilin's own firm decision to return to China to start a business.

This is the most poignant aspect of this debate. "He chose to return to his home country" is much more uncomfortable for the US than "being driven away by immigration policies." The former implies that the system can still be fixed, while the latter means that even if you open the doors wide, others may not necessarily want to come in.

When discussing Yang Zhilin overseas, what truly pains is never "another outstanding Chinese researcher has emerged," but rather another counterfactual: if this person had stayed within the US system, his papers, team, funding, and company's value should have been recorded in the US AI ledger. Now, this achievement is first seen as the ability of a Chinese team and then radiates globally through the open-source community.

For a system that has been self-assured for half a century, the most difficult thing to accept is often not that someone is better than you, but that someone has proven that they don't need to go through you to reach the finish line.

China has a dense population of engineers, teams that can quickly turn ideas into products, a massive application market, and customers willing to pay for efficiency. The open model also makes distribution easier. A team may not necessarily be hired by a large US company first or receive validation from Silicon Valley investors first, yet they can still deliver their products to developers worldwide. For top talent, the choice is no longer just between "going to the US" or "not going to the US"; it's about where their judgment can truly transform into a company, a product, or even a new ecosystem.

This is where the most concealed anxiety in the United States lies.

Stocks of storage companies have risen, which is certainly worth celebrating; cloud services have seen increased sales, which is also impressive. But they cannot replace one question: when the next group of the smartest and most ambitious people are ready to place their bets, will they hesitate to consider the United States as the only answer?

Kimi K3 came in like a gust of wind, blowing this question in through the crack in the door. The United States still has a solid foundation, with chips, cloud services, capital, and the corporate market that cannot be replaced overnight. It will continue to make big money from the global AI boom.

Relying solely on closed-source AI is no longer enough to put Americans at ease.

Original Article Link

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia