On July 19, 2026, what should have been just an ordinary trading day in AI narratives was marked on-chain as a timeline that needed re-pricing. The Alibaba Cloud Qwen team announced the open-source release of the 24 trillion parameter model Qwen3.8, self-assessing its overall performance as second only to Fable 5, bringing Chinese large models further in line with global cutting-edge in computing power. Almost simultaneously, Academician Zheng Weimin reminded the market: what is truly scarce in the era of intelligent agents is not larger computing power, but the system's ability to stably, cost-effectively, and at a high level produce Tokens. Optimistic figures from the industry quickly leveraged the computing power narrative—SK Group’s Chairman Choi Tae-won predicted that global semiconductor demand will grow by at least 50% next year, with AI-related semiconductor demand expected to rise by 60%-100%, while new supply is nearly zero. The capital expectations at the application level are equally high, with the AI company Moons of Darkness planning to go public in Hong Kong in as little as six months, with a post-financing valuation exceeding 30 billion USD. Against a unified expectation of “exploding demand and scarce supply,” an on-chain address 0xf29…41244 chose to bet against the trend: opening a short position of 2.155 million USD with a total of 300,000 CXMT tokens, investing about 15 million USD as margin, shorting storage-related assets at a planned average price of about 7.1825 USD, with its true identity and institutional affiliation still unknown, leaving only the transaction structure itself as a clue for interpretation. When the arms race of computing power, supply-demand gap, and high valuation stories formed an optimistic chorus at the primary and industry level, yet on-chain there were concentrated signals of shorting storage assets, the core conflict became clear: expectations for AI and semiconductors are shifting from “always scarce” to “profits and asset prices might not align,” and this reversal began to alter the risk appetite and pricing framework of high beta crypto assets like BTC and ETH.
Shorting CXMT on-chain, AI hardware valuations cool
The address 0xf29…41244's position on CXMT is not an emotional impulsive short, but a clearly structured mid-term bet: a scale of 2.155 million USD, corresponding to 300,000 CXMT tokens, slowly building up at a planned average price of about 7.1825 USD as TWAP, supported by around 15 million USD as margin. Such a low-leverage, timed trading design means this capital does not expect a “flash crash,” but rather bets that the computing power and storage hardware chain associated with Changxin Storage will experience a valuation correction in the near future. CXMT, as a token related to Changxin Storage, is already seen as a microcosm of the overall AI hardware valuation, and the appearance of such a large-scale systemic short on-chain essentially carves a clear question mark on this valuation anchor.
What’s more indicative is that this short appeared when the industry side was almost entirely bullish on semiconductors. Chairman Choi Tae-won projected a global semiconductor demand increase of at least 50% next year, with AI-related demand growth between 60%-100%, emphasizing that new supply is nearly zero. In the traditional asset world, this is typically translated into higher valuations and lower risk premiums. However, on-chain funds chose to short the storage sector at this time, interpreting the “demand explosion” as “overly pre-placed capacity and capital expenditures, with profits unlikely to cover current prices.” From the perspective of macro variables, this short essentially raises the risk premium on AI hardware and all tech cycle assets, demanding higher returns to compensate for potential corrections. This upward premium will transmit along the channel of “high beta tech risk assets” to BTC and ETH. For funds holding AI concept tokens, computing power-related projects, and mainstream crypto assets, the on-chain shorting of CXMT serves as a clear re-pricing signal: the tech cycle is no longer viewed unconditionally as a tailwind, the leverage and positions of high volatility assets need to be adjusted, and the pricing of BTC and ETH starts to embed more discounts of “AI hardware bubbles” and “profit realization capability,” rather than merely following the upward trajectory of the computing power expansion narrative.
Qwen3.8 open source and Token production scarcity
At the same time that on-chain investors began hitting the brakes on semiconductor valuation with the CXMT short, the Alibaba Cloud Qwen team announced the open source release of the 24 trillion parameter model Qwen3.8. Officially regarded as comparable to the world’s cutting-edge models, with overall performance self-rated only second to Fable 5, this indicates that China’s speed and scale in computing power arms are approaching the global ceiling: the sheer volume of parameters, training input, and inference computing power are no longer exclusive labels of overseas giants but are openly displayed by domestic cloud providers and released to society through open-source. This open-source move transforms the previously closed computing power arms race into a more open, replicable technology competition, reinforcing market expectations of “AI computing power is always scarce.”
However, appearing almost simultaneously was a splash of cold water from Academician Zheng Weimin. He pointed out clearly that the rapid expansion of computing power scale does not equate to efficient Token production capabilities; what is truly scarce in the era of intelligent agents is the system's ability to stably, cost-effectively, and at a high level produce Tokens. This statement directly challenges the mainstream AI narrative currently dominating the crypto market: the valuations of on-chain AI concept tokens and computing power projects are largely still built on the story of “how large is the parameter scale and how many GPUs have been invested,” while there is a noticeable lack of focus on how many effective Tokens can be produced per unit of computing power in real scenarios, and whether these Tokens can form a sustainable economic loop. With top models like Qwen3.8 accelerating open sourcing, the barriers of computing power and model technology are relatively decreasing, making the simple “computing power stacking” premium logically begin to erode, and funds are more likely to shift toward protocols and applications that genuinely improve Token production efficiency and possess clear economic viability. During this re-pricing process, more speculative AI concept coins that only talk about “computing power arms race” will face valuation compression, while BTC and ETH, as underlying assets, stand a chance to be re-narrated as basic chips that carry “real production capacity and security,” with some risk appetite flowing back from the AI themes to these two assets, becoming a core direction of current AI computing power bubble warnings on an on-chain level.
SK bullish on semiconductors, connection of costs and risk assets
Choi Tae-won’s expectations on the industry side can be viewed as the “optimistic script” for this round of computing power arms. He predicts that next year global semiconductor demand will increase by at least 50%, with AI-related semiconductor demand growth residing in the 60%-100% range, while new supply is nearly zero. The explosive demand and fixed supply indicate that it is not a “mild boom,” but rather closer to a comprehensive squeeze on computing power and storage, forcing the price center up, and at the same time raising the cost lines of AI companies and tech enterprises. Under such structural conditions, even if revenue continues to grow, profit margins and free cash flow will be hard-pressed, naturally pushing up the risk premiums in cash flow discount models, leading to a rapid decrease in the tolerance for high-valuation growth stocks, while the overall risk appetite for tech indices begins to shift from “only pricing forward stories” back to “revisiting current cash flows.”
When the prices of semiconductors and computing power rise, compounded with the environment where interest rates and inflation are still being closely monitored by global funds, the result is often that some capital pulls out of the most crowded high-growth tech sectors, returning to cash-like and highly liquid assets, and redoing beta selections in the tech cycle. From this perspective, BTC and ETH are still viewed as high beta risk assets and will be the first to face selling pressure when the tech risk appetite collectively cools, but they also possess high liquidity and decentralization characteristics, enabling them to bear trades betting against the “cooling of the tech bubble” itself: as part of the capital withdraws from the high-valuation targets in the AI and semiconductor chains, it does not completely exit, but rather shifts towards on-chain assets represented by BTC and ETH, hedging and reallocating from the AI hardware bubble, with this capital rotation being a key interactive variable worth ongoing tracking under the current reassessment of computing costs.
Moons of Darkness IPO and capital migration directional signals
In contrast to on-chain capital collectively hitting the “valuation brakes” on CXMT, the primary market is still continuing to leverage the AI application layer. As an AI company, Moons of Darkness has already initiated the process of going public in Hong Kong in as little as six months, with a post-financing valuation expectation exceeding 30 billion USD. This figure serves as a clear signal for the Hong Kong market and global tech stock funds: at the application layer, capital remains willing to pay high premiums for “storytelling, quickly attracting new users” AI products. At the same time, within the same macro window, on-chain address 0xf29…41244 established a TWAP short position of 2.155 million USD and 300,000 CXMT tokens at a target average price of about 7.1825 USD, investing about 15 million USD in margin, presenting a starkly different pricing attitude for storage assets, forming a structural differentiation of “software hot, hardware cold.”
This differentiation directly reflects the migration path of on-chain capital. On one side is an emotional amplification effect brought by IPO expectations for Moons of Darkness, prompting traditional capital to seek “the next AI application target,” naturally extending this impulse to on-chain AI tools, Agent protocols, and data service tokens. On the other hand, caution regarding hardware and storage tracks leads to the double bubble pressure of valuation and leverage on computing power and storage-related tokens. In this hot structure at the application layer and cold at the hardware layer, some crypto traders have started to view BTC and ETH as “track-neutral hedges”: both accepting the risk capital that has withdrawn from AI stocks and the computing chain, as well as acting as underlying positions for long-short hedges between AI application tokens and traditional tech stocks. The combination of Moons of Darkness IPO expectations and the CXMT on-chain shorts paints a clear outline for the current migration of capital—capital is reshaping the configuration and risk balance between high-valuation software longs and hardware storage shorts, via BTC and ETH.
AI and semiconductors interweaving long and short
Bringing together the signals of July 19, 2026, it is no longer just a “day of concentrated positive news,” but a clear cycle watershed: on one end are the open-sourcing of large models like Qwen3.8, SK Group's optimistic expectations for a 50%-100% increase in semiconductor and AI chip demand next year, and Moons of Darkness's IPO plan aiming for a 30 billion USD valuation, pushing the computing power arms race and application narrative toward higher leverage; on the other end are Academician Zheng Weimin’s cold observations that “computing power ≠ efficient Token production” and the on-chain address 0xf29…41244 using 2.155 million USD, accompanied by 15 million USD margin to short CXMT, betting on about 7.1825 USD TWAP trading structure, which directly questions the true profitability and valuation of the storage and hardware tracks. Technological breakthroughs, supply-demand gaps, and valuation divergences are sharply overlapping within the same time window, indicating that the AI and semiconductor industries have entered a phase of “high prosperity + high divergence” coexistence of long and short, where efficiency and costs begin to outweigh simple parameter scale, and the risk of hardware bubbles is systematically hedged on-chain for the first time. In such a macro environment, the pricing of BTC, ETH, and other high beta crypto assets is no longer just a function of internal liquidity and macro interest rates, but is being driven by the tech cycle and AI industry expectations: funds are simultaneously increasing shorts and hedges on AI concept tokens and computing power-related projects, while adjusting leverage ratios and overlaying structured positions, placing BTC and ETH in a neutral hub position that “bears tech corrections while absorbing risk aversion and rotation.” The pace of valuation corrections for AI hardware and storage, whether the on-chain shorts and hedge positions for AI hardware and computing tokens continue to expand, and the speed at which BTC and ETH migrate from simple high beta targets to core cyclical hedge positions will jointly determine how capital reallocates risks and returns between AI and crypto in the next stage.
Join our community to discuss and become stronger together!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
