Deepseek 2.0 Moment Didn't Arrive: Chip Stocks Stabilize, U.S. Market Awaits Earnings Season
- Core Viewpoint: Kimi K3 did not trigger a chip selloff like DeepSeek 1.0, as its narrative shifted from "computing power is sufficient" to "computing power is in short supply"; Google plans to launch an AI chip with embedded model architecture, the Frozen v2, further reinforcing hardware demand. However, with the earnings season approaching, whether capital expenditure growth can continue to exceed expectations remains a key test for the market.
- Key Elements:
- The K3 model has gone viral for its 2.8 trillion parameters and extremely low inference cost, but this has strained Moonshot AI's inference infrastructure, triggering demand for computing power expansion, shifting the narrative to "efficiency unlocks demand."
- DeepSeek 1.0's narrative of "low cost, high computing power efficiency" once disrupted the chip investment thesis; whereas K3 proves that improvements in model efficiency have instead stimulated an expansion in chip demand.
- Google is developing an AI server chip codenamed Frozen v2, which writes part of the Gemini model's architecture directly onto the silicon. Its energy efficiency ratio is expected to be 6 to 10 times that of the seventh-generation TPU Ironwood.
- The Frozen v2 is planned for deployment before 2028, signaling a shift: AI giants are turning their competition towards customized, specialized chips, creating more segmented semiconductor demand, which is a long-term positive for the industry chain.
- The earnings season in the coming week is a critical test. The market is watching whether capital expenditure at Google, Meta, and Microsoft can be sustained, and whether SK Hynix can prove the profitability of memory chips.
- The market's threshold for exceeding capital expenditure expectations is already high. Any marginal slowdown (such as downward guidance revisions) could trigger a new round of selloffs, putting the chip sector in a window of high uncertainty.
On July 21, after the U.S. stock market closed, the chip sector, which had been rattled by Kimi K3, finally got a breather.
The "DeepSeek 2.0 moment" that the market once feared did not materialize further. Although SK Hynix still closed down 1.86%, previously star stocks like SanDisk and Micron turned upward, showing signs of stabilization and recovery. After more than a week of intense selling, the entire semiconductor sector seems to have found a temporary balance point.
Why, this time, did Kimi K3 not replicate the panic script of DeepSeek 1.0?
1. K3's Narrative Has Changed: Not "Computing Power is Sufficient," but "Computing Power is Scarce"
The reason DeepSeek 1.0 caused a huge stir in March and April last year was its core narrative of "low cost, high efficiency, no computing power shortage"—it trained a model close to top-tier levels using very few GPU resources, directly challenging the investment logic that "AI performance must rely on疯狂堆砌 chips."
But Kimi K3's story is completely different.
Although K3 shocked the industry with its 2.8 trillion parameters and extremely low single-inference cost, another side effect post-launch was equally striking: a computing power shortage. The explosive popularity of K3 exceeded expectations, putting immediate pressure on Moonshot AI's inference infrastructure, making computing power expansion an urgent priority.
What does this mean? It means K3 isn't proving that "we don't need that many chips," but rather that "even if model efficiency improves, the speed of demand growth will always outpace supply." This narrative of "computing power scarcity" is precisely the most favorable support for the chip sector—it leads investors to believe again that demand for AI chips won't disappear because of improved model efficiency, but might instead continue to expand due to the explosion of application scenarios.
The logic of DeepSeek 1.0 was "efficiency replaces scale," while the logic of K3 is "efficiency unleashes demand." These two narratives have diametrically opposite impacts on chip stocks. This is the fundamental reason why "DeepSeek 2.0" did not unfold.
2. Google's Bombshell: Etching Gemini "Into" the Chip
At a critical moment when the chip sector was searching for direction, Google threw out a trump card that could change the rules of the game.
Alphabet is developing a brand new AI server chip, codenamed Frozen v2. The design concept for this chip is extremely aggressive: directly etching part of the Gemini model's architecture into the silicon itself.
This is not the traditional sense of "optimizing software to fit hardware," but rather "baking the model blueprint into the chip"—by reducing data movement between compute units and memory, it significantly cuts the power consumption and latency of each inference.
Google engineers estimate that the energy efficiency of Frozen v2 will reach astonishing levels: the number of tokens processed per unit of power consumption could be 6 to 10 times that of the current state-of-the-art Ironwood TPU. For comparison, Ironwood is already Google's seventh-generation TPU, and it merely doubled performance per watt compared to its predecessor. The generational leap of Frozen v2 far exceeds any previous chip upgrade from Google.
After the news broke, Alphabet's stock price rose from around $350 to around $359 during trading. The market is clearly reassessing Google's long-term competitiveness in the AI infrastructure field. If Frozen v2 can truly achieve commercial deployment by 2028, Google will possess one of the world's most efficient large-model inference infrastructures, making its cost advantage in AI services difficult for competitors to replicate.
But for the entire chip sector, the significance of Frozen v2 goes far beyond "Google's own business."
It sends a key signal: AI giants are not slowing down their hardware investments due to improvements in model efficiency. On the contrary, they are pushing the competition to a more fundamental and customized dimension—application-specific integrated circuits (ASICs). From NVIDIA's general-purpose GPUs to Google's dedicated TPUs, and now to Frozen v2 embedding the model architecture directly into silicon, the competition in AI hardware is shifting from "who has more cards" to "whose cards are smarter, more efficient, and more specialized."
This shift means that demand for AI chips won't shrink because "models become smaller and cheaper." Quite the opposite, the trend towards specialization and customization will create more diverse and segmented chip demand—which is a long-term positive for the entire semiconductor supply chain.
3. The Earnings Season Litmus Test: The Coming Week Will Determine the Direction
Although the chip sector has temporarily stabilized, the test is far from over.
Over the next week or so, key players in the AI supply chain will successively report their quarterly results. This earnings season is extraordinarily significant—the market is not just looking at whether the numbers are good, but also at several key questions determining whether the AI narrative can continue:
For Google, Meta, and Microsoft, will capital expenditures keep flowing? Can the revenue generated from cloud businesses and AI services cover the increasingly heavy costs of depreciation, leasing, and electricity? If not, a crack will appear in the "AI monetization" narrative, and the inflection point for capital expenditure growth could arrive sooner than we expect.
For SK Hynix, can the money spent by cloud providers ultimately translate into pricing power for memory chips, market share expansion, and profit growth? Hynix needs to prove in its earnings report that it is a true profit-making link in the AI supply chain, not just a "middleman taking a cut."
What this round of earnings truly needs to assess is not just whether capital expenditure is "high," but whether it "can continue to exceed expectations." After several consecutive quarters of "surprises," the market's threshold for beating expectations has been raised extremely high. Any marginal slowdown—whether it's guidance cuts, softened language, or key data falling short—could become the trigger for a new wave of selling.
4. Final Thoughts: In the Eye of the Earnings Season Storm, Options are a Crucial Anchor
The chip sector is navigating a window of high uncertainty. Bulls argue that K3's computing power shortage and Google's Frozen v2 prove that AI hardware demand is far from peaking. Bears argue that the growth rate of capital expenditure is about to peak, valuations are already stretched, and earnings will struggle to keep exceeding expectations.
In this environment of information fragmentation, betting on a single direction carries risks far greater than potential rewards.
BIT's options functionality offers a perfectly matched tool for this "directionless, high-volatility" market environment:
- Holding the underlying chip stock + buying put options: Insure your position before earnings, locking in downside risk
- Buying call or put options in a single direction: Bet on the post-earnings direction with far less capital than the underlying stock, with maximum loss limited to the premium
- Buying options in both directions simultaneously: Unsure if earnings will be a surprise or a shock? Bet on both sides; profit as long as volatility is high enough
The storm of earnings season is approaching. In a market without clear direction, only those with options can afford to be composed.
Risk Warning: The market conditions, valuation estimates, 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 involves market volatility, leverage, and liquidity risks. Short selling may involve unlimited loss risks. Options trading carries the risk of total loss of premium. Historical performance does not guarantee future returns. Investors should make prudent decisions based on their own risk tolerance and consult professional investment advisors when necessary.
