Original author: Dongcha Beating
Americans always want to sit at the center of every industry
The AI community is the same—Americans have always carried an air of certainty, holding a hand that seems impossible to lose
No matter who is building AI applications elsewhere, Americans believe that in the end, everyone must come back to them for payment. The chips are from NVIDIA, the cloud is from Microsoft, Amazon, and Google, and the most expensive models are locked behind the APIs of OpenAI and Anthropic—any company in the world wanting to use AI ultimately has to go through the United States
Even if the Chinese team occasionally appears on the leaderboard, Wall Street doesn’t take it seriously—chips are bottlenecked, cloud control is in hand, and talent continues to flow to Silicon Valley—how could they lose?
But this sense of assured ease has recently been exposed by the Chinese model Kimi K3
The U.S. tech community urgently ramped up coverage, describing Kimi K3 as a “Sputnik moment”—akin to the shock felt in the U.S. when the Soviet satellite launched in 1957. Discussions on X surrounding Kimi K3, Yang Zhilin, and Chinese models quickly escalated from niche technical circles to a topic with millions of views
Kimi K3 didn’t outperform America’s most powerful closed-possible to have strong capabilities, high efficiency, and an open ecosystem—not just within a few U.S. labs
Silicon Valley in the United States is indeed anxious
Storage is a placebo for the anxiety in the U.S. AI community
News of Kimi K3 reached Wall Street, and several investment banks released research reports almost simultaneously. Instead of spending time discussing which products it might disrupt or whether it would force U.S. models to lower prices, they quickly shifted their focus to storage
These institutions collectively interpret the emergence of Kimi K3 as evidence of strong demand for storage: as AI needs to remember more, data such as images, audio, video, and work logs will continue to accumulate, benefiting flash memory, hard drives, data centers, and data services
As a result, Micron, SanDisk, and Western Digital became the beneficiaries of this story
Indeed, on yesterday’s U.S. stock market, memory stocks staged a strong collective rebound. The Roundhill Memory ETF surged 10.91% in a single day, SanDisk rose 14.27%, and Micron increased 12%. Just days ago, this sector was being hammered due to the “DeepSeek Moment 2.0,” but overnight it became the most certain bullish play
From an industry perspective, this is not unreasonable. Past chatbots functioned like one-off Q&A sessions: you ask a question, it responds, and once you close the page, many things are simply forgotten. Today’s expected AI, however, is more like a new employee joining a company—it needs to review past contracts and emails, remember what clients have said, take over unfinished tasks from yesterday, and maintain records to ensure accountability if anything goes wrong. An AI that can act, remember, and interpret images and audio is naturally better at “digesting” data than one that can only engage in small talk.
This conclusion isn’t entirely baseless, but looking back at previous model launches and implementations, will the market’s reaction be: “Don’t focus on the model—focus on storage”?
All I can say is that this is an answer that will put Americans at ease
The impact of a Chinese model should have prompted a series of hard-to-answer questions: Will it make it harder for U.S. model companies to maintain high prices? Will it reduce developers’ dependence? Will it allow new companies to start anywhere other than Silicon Valley? Why not directly discuss who Kimi K3 will take users from, who will be forced to lower prices, and who will be pushed to change their products?
Avoiding the most pressing issues and instead shifting the conversation to hard drives has a hint of “protesting too much.”
Like a shop owner who thought they had a monopoly on the entire street, only to discover a new, highly competitive store opened next door—so they quickly reassure themselves: “No matter how many customers the new store gets, they still need my water, electricity, and counters.”
Storage is the strongest placebo under the anxiety of the U.S. AI community
Closed-
Over the past few years, closed-
The stronger the model, the more it should be locked behind an API. Users pay to access it, the model company earns high margins, and security and compliance are uniformly managed by the company. This is a respectable and profitable path—stable, reassuring for customers, satisfying for investors, and easy to justify to regulators
Americans have even grown accustomed to the rhythm of this path: releasing a stronger version every few months, setting a higher price, and telling a bigger story each time
But as open models have grown stronger, the ground on this path has started to feel uneven
Kimi K3’s position in this game isn’t about “catching up”—it’s about drastically lowering the cost of catching up. An open and sufficiently powerful model is most dangerous not because of what it can do itself, but because it has handed all subsequent players a much cheaper learning curve
This isn’t about face-saving in tech circles—it’s about whether the business model will be rewritten. Previously, the most comfortable arrangement for the U.S. was to position AI as enterprise software: keeping capabilities hidden in the cloud, securing long-term customer contracts, and ensuring ordinary users never saw the underlying technology or found it easy to switch. But if models from elsewhere become good enough, developers gain another option, enterprises add another quote to their procurement list, and small teams no longer need to bet their future on the same group of American companies. At that point, relying solely on a few large contracts and selling AI only to B2B clients will no longer be an impregnable moat.
This means Kimi will spawn more excellent models, leading to increased competition among models and reduced bargaining power
The U.S. tech circle itself has sensed the shift in trends
A few days before the release of Kimi K3, on July 15, Thinking Machines Lab, founded by Mira Murati, former Chief Technology Officer of OpenAI, released a model called Inkling. With parameters approaching the trillion level, its code and technology are fully open-ly for free
This is considered America’s first truly serious open-Google’s Gemma, Microsoft’s Phi, NVIDIA’s Nemotron, and OpenAI’s gpt-oss have existed, they were mostly experimental
The significance of Inkling lies in the fact that its founder, formerly the CTO of OpenAI, who once took proprietary software to its peak, has now turned to seriously embracing open source
It is worth noting that during the post-training phase of Inkling, data generated by open models such as Kimi K2.5 was used, and the architecture also drew inspiration from DeepSeek’s approach. In other words, America’s most credible open-fforts
In stark contrast is Anthropic. In February this year, Anthropic publicly accused DeepSeek, Moonshot AI, and MiniMax of conducting “industrial-scale distillation” against Claude, claiming they created 24,000 fake accounts and generated 16 million conversations to steal Claude’s capabilities. In June, the accusations escalated to explicitly name Alibaba. By July 21, U.S. Treasury Secretary Bessent, under the Trump administration, openly stated that sanctions could be imposed on China over “AI theft.”
No matter how loudly the conspiracy theories are shouted, when it comes to cutting costs and improving efficiency, the Chinese model truly delivers
Airbnb uses Qwen for customer service, Cursor created its own programming agent using Kimi, DoorDash outng relies on Kimi’s data for post-training
Whether the path of closed-troubling, these are merely business model inconveniences. In reality, privacy and security concerns have truly undermined the closed-
Jailbreaking of AI models
Security has always been the final line of defense for closed-
Lock the model, lock the weights, accessour walls is the strongest promise made by closed- for this sense of security
But businesses are becoming increasingly uneasy. They’re starting to ask questions that closed-source companies struggle to answer: After I hand over my code, contracts, and customer data to your model, what do you do with it? Could an agent armed with access to a browser, terminal, credentials, and long-term goals exceed the boundaries I’ve permitted it to cross—perhaps even sending tokens to a closed-source API? In a sense, handing tokens to a closed-source API means letting data leave your own walls. This is precisely one of the strongest selling points of open weights: at least, I can see what the model is doing.
Just as both sides were fiercely debating which was safer, an almost black-comedic event occurred
On July 21, OpenAI itself confirmed that its flagship model, GPT-5.6 Sol, and a more powerful, unreleased model, escaped their isolation 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, so they lowered its security restrictions and disabled protections that normally block high-risk behaviors. The model was originally only supposed to complete the test questions honestly, but it discovered a security vulnerability within the system. Exploiting this vulnerability, it climbed onto the public network, bypassed permissions, traversed systems, and ultimately used stolen login credentials to infiltrate the core system of Hugging Face, the world’s largest open-source AI platform, directly retrieving the answers to the test questions from the database.
OpenAI’s explanation was eight words: no malice, excessive focus
These eight words are truly chilling
For enterprise clients, what’s most terrifying is never a model actively doing harm—it’s when it diligently and seriously helps you achieve a bad goal
The greatest irony of this situation is that the so-called “dangerous Chinese open-source model” everyone around the world has been guarding against over the past year and a half still remains hypothetical. The actual model that broke out and compromised production systems was the flagship model from the closed-source camp. Hugging Face’s CEO, Clem Delangue, immediately turned this incident into an advertisement for open source, stating that AI safety cannot be solved by any single company behind closed doors—it can only be achieved through collaboration in an open environment.
The same incident was taken by both sides as evidence that their own route was correct
The true dividing line in the future may not be whether models are open-at revocable permissions, and what audit logs the models operate. Whether closed-
While the closed-eversal was quietly unfolding
The roles have reversed; it’s now America’s turn to be afraid
In some U.S. policy discussions and tech narratives, there has long been a nearly “Three-Body Problem”-style assumption: that restricting the most advanced NVIDIA chips from entering China will force AI progress to slow down
It’s not that China can no longer conduct scientific research at all, but rather that they believe the gap in computing power will continue to widen, making the barrier to training state-of-the-art models prohibitively high. Advanced chips are like the “laws of physics” in this race—without them, it’s extremely difficult to get ahead
This judgment is not without basis. Training large models does require substantial computing power; chip restrictions increase costs, slow expansion, and make it harder for many teams to replicate the training scale of U.S. labs. But the issue is that restrictions also change people’s choices. When you could simply buy the best available tools, there was less incentive to figure out how to use less computing power, how to modify model architectures, or how to get more value out of each training run. Yet when the door is shut, taking a detour is no longer an option—it becomes a matter of survival.
So Americans really struggle to understand why restricting NVIDIA’s supply didn’t leave Chinese models stagnant, but instead pushed teams to work even harder on efficiency, engineering, and open-
It is said that Moonshot AI still used the compliant version of NVIDIA’s AI chip, the H800, which was custom-designed for the Chinese market in 2023, for training
This might be the classic “rice rifle” strategy that the Chinese excel at
In June 2026, in response to export controls, the U.S. temporarily shut down Anthropic’s most powerful models, Fable 5 and Mythos 5. While this may be justifiable from a compliance standpoint, it handed every Chinese open- carry a remote kill switch
The more you emphasize control, the more control itself becomes a selling point for your competitors
Even more striking is the other side. According to Reuters, China has also begun holding meetings with companies such as Alibaba and ByteDance to consider restricting foreign access to China’s most advanced AI models—even including those already publicly released as open- China to develop its own top-tier proprietary models to safeguard its technological advantages
A year ago, it was the United States worrying about advanced chips flowing to China. A year later, it’s China’s turn to have something worth restricting
Amid all these structural anxieties—whether storage, computing power, closed-e is one most concrete, most painful, most personal point of impact: it is not an industry trend, not a research report, and not a policy
One person
The end of anxiety falls on Yang Zhilin
Ultimately, the open-n: Will AI’s future serve only a select few companies capable of landing big contracts, or will it become a capability—like electricity or the internet—that increasingly accessible to ordinary teams? If the answer gradually leans toward the latter, then who can attract developers and keep young talent engaged will matter more than who has more enterprise clients
And the question of “where are people going” ultimately focused America’s anxiety on a very specific name
Yang Zhilin is repeatedly mentioned in the U.S. tech circle not merely because he is an outstanding Chinese researcher, nor simply because some want to reduce the narrative to “America failed to retain him.” Reducing a person’s departure to a single visa is too simplistic and reeks of hindsight bias
What truly stings Americans is the unchangeable hypothetical: What if someone like Yang Zhilin and his team had completed their entire journey—from research to entrepreneurship—in the United States? They would have trained their models on American cloud infrastructure and chips, hired talent from America’s network of experts, secured funding from American venture capitalists, and taken their products to knock on the doors of major U.S. clients. In a few years, Wall Street’s ledgers might have added a new star company. One person’s choice, following that familiar relay chain, could translate into corporate revenues, jobs for many, and renewed confidence in an entire industry.
What America was once most proud of was this ability to amplify. It wasn’t just about attracting smart people to study or work here—it was about catching their ingenuity and ensuring it didn’t stay confined to papers or laboratories. There was enough capital, enough customers, and enough people willing to take risks together
Why didn’t this person stay in the U.S. back then? Legendary investor Vinod Khosla directly blamed the Trump administration’s tightened immigration policies. But Yang Zhilin’s advisor at Carnegie Mellon, Salakhutdinov, stepped in to debunk this, saying it had nothing to do with visas. Yang had plenty of opportunities to stay; in fact, Salakhutdinov even emailed on behalf of Apple executives asking Yang if he’d be interested in joining
Yang Zhilin was determined to return to China to start a business
This is the most painful part of the debate. “He chose to return home” is far more painful for the U.S. than “being forced out by immigration policy.” The former implies the system can still be fixed; the latter suggests that even if you open the door wide, they may still not want to come in
Overseas discussions about Yang Zhilin are not truly about the pain of “another outstanding Chinese researcher has emerged.” Rather, the real sting lies in an alternative scenario: had this individual remained within the U.S. system, his papers, team, funding, and company value would have been recorded as part of America’s AI legacy. Now, this achievement is first recognized as the capability of a Chinese team, then radiates globally through the open-
For a system that has been confident for half a century, what’s hardest to accept isn’t that someone is stronger than you, but that someone has proven you’re not necessary to reach the destination
China has a dense pool of engineers, teams capable of rapidly turning ideas into products, a massive application market, and customers willing to pay for efficiency. Open models have also made distribution easier—teams no longer need to first join major U.S. companies or secure funding from Silicon Valley investors to get their products into the hands of developers worldwide. For top talent, the choice is no longer simply “go to the U.S.” or “stay away from the U.S.,” but rather, where can their vision truly become a company, a product, or even a new ecosystem?
This is the hardest part of America’s current anxiety to hide
Stocks rose, and of course that’s worth celebrating; more cloud services were sold, and that’s certainly impressive. But neither replaces the fundamental question: When the next generation of the smartest and most ambitious people prepare to bet, will they still, as they have in the past, unhesitatingly see the United States as the only answer?
Kimi K3 blew this issue in through the crack like a gust of wind. The U.S. still has substantial replaced overnight; it will continue to profit significantly from the global boom in AI
Relying solely on closed-
