Last Friday, the U.S. stock market's storage chip sector fell collectively once again.
On the surface, this is merely a continuation of the ongoing pullback in chip stocks over the past two weeks. However, beneath the surface, a new variable from China is shaking the entire market, namely the launch of Kimi K3 by Moonshot AI on July 17.
This open-source large model, boasting 2.8 trillion parameters, has topped the global code evaluation leaderboard CodeArena, surpassing Anthropic's Claude Fable 5. What makes the market even more restless is its cost: the inference cost for a single task is merely $0.94, less than half of Claude Opus 4.8, essentially on par with OpenAI's GPT-5.6 Sol.
Larger in scale, stronger in coding capability, and cheaper in inference costs—does this combination sound familiar?
In response, JPMorgan directly labeled it as: "DeepSeek 2.0 moment."
1. The Traumatic Memory of DeepSeek 1.0: Chip Stocks and Bitcoin Bleed
To understand why the phrase "DeepSeek 2.0" can cause sleepless nights on Wall Street, we need to look back at what happened in March and April of 2025.
Last year, DeepSeek achieved performance close to GPT-4 level at an extremely low training cost. Once the news broke, the market plunged into a collective panic over the high valuations of AI chips:
Nvidia's stock price plunged from a peak of about $135 to around $85, a nearly 40% drop.
Bitcoin fell from nearly $110,000 historic high to around $75,000.
The logic is straightforward: If Chinese companies can train top models using fewer chips and at lower costs, do U.S. cloud providers and tech giants still need to continue their mad procurement of GPUs and HBMs? Is the arms race for AI infrastructure actually just an overvalued bubble from the start?
The market at that time gave a clear answer—plummet.
Now, with Kimi K3 emerging with the tag of "largest open-source model + lower inference cost," its parameter scale is even more exaggerated than DeepSeek's at that time. Once fear is activated, the selling pressure in the chip sector spreads rapidly.
2. SemiAnalysis's Counter Interpretation: K3 is not Eliminating GPU Demand, But Amplifying It
But there is another side to the story.
The semiconductor research institution SemiAnalysis recently provided a completely opposite framework of reasoning. Their core argument is: K3's massive parameter scale and inference architecture will not weaken the demand for high-end AI hardware; instead, it may become another long-term demand engine for Nvidia and its supply chain.
Breaking it down, SemiAnalysis's judgment is based on several key facts:
First, the model is "too large," which in fact requires more hardware.
K3's parameter scale exceeds 2.8 trillion, with model weights alone requiring over 1.5TB of HBM (High Bandwidth Memory) space. Currently, the HBM capacity of a single top Nvidia GPU is far from sufficient to load the entire model. Even in relatively limited user concurrency scenarios, KV caches still require a significant amount to be offloaded to CPU DDR5 memory and NVMe storage devices—HBM space will not only be insufficient but actually constrained.
Second, inference deployment places extremely high demands on hardware scale.
Moonshot AI has previously revealed that K3 requires at least a large-scale expansion domain architecture composed of 64 high-end chips for efficient inference deployment. This level of cluster scale is highly consistent with the design concepts of rack-level AI systems such as Nvidia's GB200/GB300 NVL72. In other words, K3 is not "replacing" high-end hardware, but "defining" the usage scenarios for the next generation of high-end hardware.
Third, the misconception of "linear attention reducing GPU demand."
There has been a popular viewpoint in the market: more efficient attention mechanisms (such as linear attention) imply less computational load, which in turn implies a decrease in GPU demand. SemiAnalysis believes this logic has fundamental flaws. The real impact may be precisely the opposite—more efficient model architectures lower the cost of individual inference, significantly reducing the thresholds for AI application deployment, thus driving more enterprises, more scenarios, and larger scale AI deployment demand.
From a macro perspective: improved model efficiency → decreased unit costs → AI application explosion → total demand for computing power increases instead of decreases. This is a recurring script throughout the history of the semiconductor industry—every "efficiency revolution" has ultimately brought about a larger wave of hardware investment.
3. What is the Market Trading? A Tug of War Between Fear and Rationality
The current market is in a state of extreme division.
The bears have a logical intuition: Kimi K3 has once again proven the breakthrough capabilities of Chinese AI companies on the path of low cost and high efficiency. If this trend continues, AI firms will inevitably re-evaluate their capital expenditure (capex) plans. The narrative of AI chips being "in short supply" may be being replaced by the narrative of "good enough."
The bulls have equally compelling counterarguments: K3 is not a smaller version of DeepSeek, but an enlarged version—it is so large that high-end hardware clusters cannot operate without it. Research by SemiAnalysis indicates that HBM is not in surplus; instead, it may actually become more scarce. Moreover, if AI applications become widely adopted because costs are reduced, the total demand for computing power will show exponential growth.
Both sides have valid points. And this is precisely the most painful moment for the market—when two completely opposing logics can find solid data support, prices are not reflecting fundamentals, but pricing emotions.
4. In a Split Market, You Don't Have to Bet on a Single Direction
In this highly uncertain environment, the hardest part is not judging direction but rather controlling the costs of exposure to the wrong direction.
Bulls claim that the pullback in chip stocks is a "miscalculation," while bears argue that the super cycle of AI chips has peaked. Both viewpoints are supported by data and endorsed by reputable institutions. For ordinary investors, rather than betting on one direction in this information haze, it might be better to adopt a different mindset—ensuring that regardless of which way things go, they won’t be completely knocked down.
The BIT platform's options feature is about to launch, conveniently providing several core tools for such "uncertainty trading":
① Holding chip stocks + buying protective put options
If you are heavily invested in Nvidia, Micron, or SK Hynix but are concerned that the K3 event could trigger further pullbacks, you can buy corresponding stock put options. If the stock price continues to fall, the option's profit can cover the stock's loss; if it rebounds, you abandon exercising the option, with the maximum loss being just the option’s premium.
② Buying single-direction call options
If you agree with SemiAnalysis's "miscalculation" logic, believing that chip stocks will return to fundamentals after emotional selling, you can directly buy call options. Betting on the rebound direction with funds far below the stock price, with maximum loss capped at the premium.
③ Buying single-direction put options
If you think the "DeepSeek 2.0" storyline will repeat itself and that chip stocks still have room to fall, you can buy put options to short directly. No need for margin calls or borrowing securities, with maximum loss being just the option fee.
④ Buying both directions simultaneously
If you are certain chip stocks will move but are unsure whether it will be up or down, you can buy both call and put options simultaneously. As long as the stock price has enough volatility, any direction's profit may cover the costs of both options.
5. In Conclusion
The release of Kimi K3 has pushed AI chip investment to a critical crossroads. Is it the beginning of "DeepSeek 2.0," or a prelude to "demand being redefined"? The answer to this question may require time to fully manifest.
But before the answer is revealed, the most valuable thing investors can do is not to guess who is right in the end, but to prepare an exit strategy for their judgment errors.
The BIT options feature covers core chip targets like Nvidia, Micron, and SK Hynix. Whether you are bullish or bearish, there are corresponding tools to express your views—while also locking in your maximum potential losses within a range you can bear.
In an unclear market, those who have options retain control.
Risk Warning: The market trends, valuation calculations, and product descriptions mentioned in this article are for reference only and do not constitute investment advice. Trading in U.S. stocks and their derivatives carries market volatility, leverage, and liquidity risks; short selling may face unlimited loss risks; options trading may involve the full loss of premiums; past performance does not represent future returns. Investors should make cautious decisions based on their own risk tolerance and consult professional investment advisors if necessary.
Disclaimer: This article represents only the personal views of the author and does not represent the position and views of this platform. This article is for information sharing only and does not constitute any investment advice to anyone. Any disputes between users and authors are unrelated to this platform. If the articles or images on the webpage involve infringement, please provide relevant proof of rights and identity documents and send an email to support@aicoin.com. The relevant staff of this platform will conduct an investigation.