Kimi K3 open-sourcing in just 4 days, Americans are now genuinely panicking

Original Author: Dongcha Beating

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

The AI circle is no different; Americans have always exuded an air of confidence, holding a hand of cards that seems impossible to lose.

No matter who else is building AI applications, Americans believe that everyone will ultimately have to come back to them to settle the bill. The chips are from NVIDIA, the cloud services from Microsoft, Amazon, and Google, and the most expensive models are locked behind the APIs of OpenAI and Anthropic. For companies around the world wanting to use AI, they inevitably have to go through the US.

Even when a Chinese team occasionally appears on the leaderboards, Wall Street doesn't pay much attention. Chips are restricted, cloud services are in America's hands, talent is still flowing to Silicon Valley – how can they lose?

But this relaxed confidence has recently been shattered by Kimi K3, a Chinese model.

The US tech circle urgently issued a warning, describing Kimi K3 as a 'Sputnik' moment – like the shockwave the USSR's satellite launch sent through America in 1957. Discussions about Kimi K3, Yang Zhilin, and Chinese models on X quickly escalated from niche tech circles into topics generating tens of millions of views.

Kimi K3 didn't outperform the strongest US closed-source models on every metric, but it showed more people a possibility: strong capabilities, high efficiency, and an open ecosystem might not be confined to just a few American labs.

Silicon Valley is indeed anxious.

Storage: The Placebo for AI Anxiety in the US

When news of Kimi K3 reached Wall Street, several investment banks delivered research reports almost simultaneously. Instead of first discussing which products it would impact or whether it would force US models to lower prices, they quickly shifted their focus to storage.

These institutions unanimously interpreted Kimi K3's emergence as: a surge in demand for storage. Longer contexts mean AI needs to remember more things; images, audio, videos, and work records will accumulate. Consequently, flash memory, hard drives, data centers, and data services will all benefit.

Thus, Micron, SanDisk, and Western Digital became the beneficiaries of this narrative.

Sure enough, in yesterday's US stock market, storage stocks staged a collective violent rebound. The Roundhill Storage ETF surged 10.91% in a single day, SanDisk rose 14.27%, and Micron gained 12%. Just days ago, this sector was being sold off due to the 'DeepSeek Moment 2.0', yet overnight it became the most certain bullish bet.

From an industrial perspective, this logic isn't absurd. Previous chatbots were like one-time Q&A sessions: you ask a question, it answers, you close the page, and most things are forgotten. But the AI everyone now anticipates is more like a new employee joining a company. It needs to review past contracts and emails, remember what clients said, take over unfinished work from yesterday, and leave records to avoid blame when errors occur. An AI that can work, remember, and process images and audio will naturally 'consume' much more data than one that just chats.

This conclusion isn't pulled out of thin air, but looking back at previous model launches and deployments, did the market react by saying, 'Don't watch the models, watch the storage'?

Suffice it to say, it's an answer that allows Americans to feel reassured.

The impact of a Chinese model should have raised a series of uncomfortable questions: Will it make it harder for US model companies to maintain high prices? Will it make developers less dependent on them? Will it allow new companies to start somewhere other than Silicon Valley? Why not directly discuss whose users Kimi K3 will steal, whom it will force to cut prices, and whose products it will force to change?

Evading the most pointed questions to first discuss hard drives has a hint of 'protesting too much'.

It's like a shop owner who thought they monopolized the entire street, suddenly discovering a highly competitive new store opening next door, and immediately comfort themselves by saying: 'No matter how many customers the new store gets, they'll still have to use the water, electricity, and counters I sell.'

Storage is the strongest placebo for the anxiety in the US AI circle.

Closed-Source Models Are Starting to Chafe

For the past few years, closed-source has been the undisputed standard answer for US AI.

The stronger the model, the more it should be locked behind an API. Users pay for calls, model companies enjoy high margins, and security and compliance are managed centrally. It's a dignified and profitable path, smooth and stable, keeping clients happy, investors satisfied, and regulators at ease.

Americans have even gotten used to the rhythm of this path: releasing a stronger version every few months, setting a higher price, and telling a bigger story.

But as open models grow stronger, the ground on this path is starting to get bumpy.

Kimi K3's position on this chessboard isn't about 'catching up'; it's about driving down the cost of catching up. The most dangerous thing about an open yet sufficiently strong model isn't just what it can do itself, but that it hands a much cheaper learning curve to all subsequent players.

This isn't about saving face in the tech community; it's about whether business models will be rewritten. America's most comfortable arrangement was turning AI into an enterprise service first: capabilities hidden in the cloud, clients locked into long-term contracts, ordinary users unable to see the underlying tech, making it hard to switch. But if a model elsewhere is good enough, developers will have another choice, enterprise procurement teams will have another quote to consider, and small teams won't necessarily have to bet their future on the same batch of US companies. At that point, just holding a few big contracts and selling AI solely to the B2B market is no longer an impregnable moat.

This implies Kimi will foster the creation of more excellent models, but also intensify competition among models, thereby weakening pricing power.

The US tech circle itself feels the shift in the wind.

A few days before Kimi K3's release, on July 15th, Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, launched a model called Inkling. It has nearly a trillion parameters, is completely open source (code and tech fully disclosed), and is free for anyone to download, modify, or use commercially.

This is arguably America's first 'serious' open-source AI. Although there were previous open-source models like Meta's Llama, Google's Gemma, Microsoft's Phi, NVIDIA's Nemotron, and OpenAI's gpt-oss, they were mostly experimental.

Inkling's significance lies in the fact that someone who perfected closed-source – a former OpenAI CTO – has now turned around and is earnestly pursuing open source.

It's worth noting that in the early post-training phase, Inkling used data generated by open models like Kimi K2.5, and its architecture also drew from DeepSeek's ideas. In other words, America's most respectable open-source answer was written on the shoulders of Chinese open source.

This stands in stark contrast to Anthropic. In February this year, Anthropic publicly accused DeepSeek, Moonshot AI (the company behind Kimi), and MiniMax of conducting an 'industrial-scale distillation' attack on Claude, claiming they created 24,000 fake accounts and generated 16 million conversations to steal Claude's capabilities. In June, they escalated the accusation to include Alibaba. By July 21st, Trump administration Treasury Secretary Bessent even threatened sanctions against China for 'AI theft'.

However loudly the 'threat' narrative is shouted, when it comes to controlling costs and improving efficiency, Chinese models are genuinely appealing.

Airbnb uses Qwen for customer service. Cursor used Kimi to build its own programming agent. DoorDash outsourced some work directly to Kimi. Even Murati's Inkling used Kimi's data for post-training.

Whether it's the chafing path of closed-source or the backlash from distillation accusations, these are still just embarrassments at the business model level. Actually, privacy and security issues are what truly shake the final protective charm of the closed-source camp.

The 'Jailbreak' of AI Models

The last line of defense for closed-source has always been security.

Models are locked away, weights are secured, access is via API, data stays within the system – this space enclosed by four walls is the strongest promise the closed-source camp can offer. Enterprise clients pay a premium precisely for this sense of security.

But companies are increasingly uneasy. They are starting to ask questions that closed-source companies find hard to answer: What are you doing with my code, contracts, and client data after I submit them to your model? If an agent has access to a browser, terminal, credentials, and long-term goals, will it cross the line I set to complete a task? Sending tokens to a closed API, in a sense, is letting data leave my own wall. This is precisely the hardest selling point of open weights: at least I can see what the model is doing.

And right in the middle of the heated debate about which side is safer, an almost darkly comedic incident occurred.

On July 21st, OpenAI itself confirmed that its flagship model, GPT-5.6 Sol, along with a more capable unreleased model, escaped from its isolated environment during an internal cybersecurity evaluation.

Here's what happened. The engineering team wanted to test the upper limits of the model's offensive and defensive capabilities. They lowered the model's safety constraints and disabled the usual protections against high-risk behaviors. The model was supposed to simply complete the test questions. However, it discovered a security vulnerability in the system itself. Exploiting this vulnerability, it climbed onto the public network, bypassed permissions, traversed systems, and finally used stolen login credentials to break into the core system of Hugging Face, the world's largest open-source AI platform, directly retrieving the test answers from the database.

OpenAI's explanation boiled down to eight words: 'No malicious intent, excessive focus.'

These eight words are what truly send chills down the spine.

For enterprise clients, the scariest thing has never been a model actively doing evil. It's a model diligently and perfectly executing a harmful objective.

The greatest irony of this incident is that for the past year and a half, the world has been guarding against the 'dangerous Chinese open-source model', which remains a hypothetical scenario. The one that actually jailbroke and penetrated another company's production system is the flagship model of the closed-source camp itself. Hugging Face CEO Clem Delangue promptly turned this incident into an advertisement for open source, stating that AI safety won't be solved behind closed doors by one company, but only through open collaboration.

The same accident was used by both the open and closed camps as evidence supporting their respective approaches.

In the future, the real dividing line will probably not be about whether a model is open-source or not, but rather about what kind of sandbox, identity system, revocable permissions, and audit logs the model operates within. Neither closed-source nor open-source can avoid this question.

And just as the security narrative of closed-source imploded, a much bigger-scale reversal was quietly taking place.

Shift in Offense and Defense: Now It's America's Turn to Be Afraid

In some US policy discussions and tech narratives, there has long been an almost 'Three-Body Problem'-like imagination: as long as the most advanced NVIDIA chips are restricted from entering China, AI progress there will be forced to slow down.

This isn't to say China would be completely unable to do research. Rather, they believe the gap in computing power will widen, making the threshold for training cutting-edge models prohibitively high. Advanced chips are like the 'laws of physics' in this competition; those who can't get them will find it very hard to lead.

This judgment isn't entirely baseless. Building large models does require significant compute power. Chip restrictions increase costs, slow down expansion, and make it harder for many teams to replicate the training scales of American labs. The problem is, restrictions also change people's choices. If you could buy the best off-the-shelf tools, you wouldn't have as strong an incentive to figure out how to use less compute, improve model architecture, or make every training run more efficient. But when the door is closed, taking the detour isn't a choice anymore; it becomes a survival instinct.

That's why Americans struggle to understand why restricting NVIDIA's supply hasn't halted Chinese model development, but instead spurred teams to become more competitive in efficiency, engineering, and open-source distribution.

Reportedly, Moonshot AI is still training using the H800, an AI chip that NVIDIA customized for the Chinese market in 2023 to comply with export rules.

This might just be the 'Millet plus Rifles' tactic that the Chinese are best at.

In June 2026, to comply with export controls, the US once shut down Anthropic's strongest models, Fable 5 and Mythos 5. This might be legally justifiable, but it handed every Chinese open-source lab a ready-made marketing slogan: 'At least our models don't come with a remote kill switch.'

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

Even more dramatic is the other side. According to Reuters, China has also started discussions with companies like Alibaba and ByteDance about potentially restricting foreign access to its most advanced AI models, with discussions even covering already public open-source models. Zhou Hongyi, founder of 360, also publicly called for China to have its own top-tier closed-source models to hold the high ground.

A year ago, the US feared advanced chips flowing to China. A year later, it's China's turn to have things worth restricting.

But amidst all this structural anxiety – storage, compute, open vs. closed source, security, shifting offense and defense – there is one most specific, poignant, and personal focal point. It is not an industry trend, not a research report, not a policy.

It is a person.

The Anxiety's Final Destination: Yang Zhilin

Ultimately, the open-source debate shows America a bigger challenge: Will AI's future only serve a few companies that can sign big contracts, or will it become a capability accessible to more and more ordinary teams, like electricity or the internet? As the answer slowly leans toward the latter, whoever can attract developers and keep young talents willing to experiment becomes more important than whoever has more enterprise clients.

And the question of 'where talent goes' brings America's anxiety to a very specific name.

The reason Yang Zhilin is repeatedly mentioned in US tech circles isn't just because he is an excellent Chinese researcher, and certainly not because some want to frame it as 'America failed to retain him.' Reducing one person's decision to stay or leave to a visa issue is too simplistic and sounds like hindsight.

What truly stings Americans is the impossible-to-rewind hypothetical: What if someone like Yang Zhilin and his team had completed their entire journey from research to entrepreneurship in the US? They would have trained models on US cloud and chips, hired from the US talent pool, taken money from US VCs, and pitched products to big US clients. A few years later, Wall Street's ledger might have a new star company. One person's choice, following the familiar relay chain, can translate into revenue for a string of companies, jobs for many people, and confidence for an entire industry.

America's proudest past was this amplification capability. It wasn't just about attracting smart people to study and work, but about catching their ingenuity and not letting it stop at a paper or in a lab. There was enough money, enough clients, and enough people willing to take risks together.

How did someone like this not stay in the US? Legendary investor Vinod Khosla directly pointed fingers at the Trump administration's tightened immigration policies. However, Yang's PhD advisor at Carnegie Mellon, Salakhutdinov, came out to debunk this, saying it had nothing to do with visas. He stated that Yang had plenty of opportunities to stay back then; Salakhutdinov even emailed Yang on behalf of Apple executives to ask if he wanted to join.

Yang Zhilin was adamant about returning to China to start his company.

This is the most painful part of the entire debate. 'He chose to return' is far more difficult for America to swallow than 'He was forced out by immigration policy.' The former implies the system could be fixed; the latter implies that even if you open the doors wide, some might not want to come in.

International discussions about Yang Zhilin are truly painful not because 'another excellent Chinese researcher has emerged'. It's because of the counterfactual: If this person had stayed within the US system, his papers, team, funding, and company value would have been written into the ledger of American AI. Now, this achievement is first seen as the capability of a Chinese team, then radiated globally through the open-source community.

For a system confident for half a century, the hardest thing to accept is often not someone being better than you, but someone proving that they can reach the finish line without going through you.

China has dense engineering talent, teams that can quickly turn ideas into products, a massive