Compiled by: Big Pliers, PANews

Recently, global AI/power stocks have experienced a round of intense fluctuations, with Korean storage stocks leading the decline, and US stock AI hardware sector collectively adjusting. Morgan Stanley's Chief Economist for China, Xie Ziqiang, put forward a judgment: AI investment is entering a "halftime break," and the main theme in the second half will shift from merely "selling shovels" to true "AI adopters" that can rely on AI to reduce costs, increase efficiency, and enhance revenue.

PANews held a Twitter Space on the evening of August 11 with the same topic, inviting four guests: Investment TALK, Elio Cui, head of BIT brokerage business, Harry, head of the Virtuals ecosystem, and Sun, co-founder of The Mu, to discuss the causes of this round of adjustment, the shift in AI's second half main theme, whether the semiconductor hardware narrative has ended, HALO asset revaluation, opportunities in China's digital infrastructure, capital expenditure, and interest rates. Below is the transcript compilation.

1. Is this round of sharp decline a fundamental issue or a leverage cleanup?

Investment TALK first provided a complete review: In March, the market was worried about geopolitical conflicts; in April, retail investors were generally left out (a stark contrast to the previous year's frenzied bottom fishing during tariff declines); in May, funds rushed in, and the chip sector rose 100% within two months, forming obvious crowded trades and leverage buildup; in June, it consolidated, and the Federal Reserve meeting became the starting point for the decline; July marked the beginning of the leverage cleanup phase. His conclusion is: this round of adjustment is not largely related to fundamentals, but is essentially a cleanup of emotional premiums and leverage — Meanwhile, capital expenditures from Microsoft, Amazon, and Google, the three major cloud providers, have not decreased, but rather increased, and SpaceX's commitment to data center investment is also significant.

Elio Cui agreed with this judgment and provided three quantifiable signals to judge whether a "fundamental turning point" has truly arrived:

"A real fundamental turning point requires at least three signals to appear: First, guidance for capital expenditures begins to downgrade; second, cloud service and AI orders cannot keep up with investments; third, credit spreads clearly start to widen."

He added that from BIT's perspective as a direct connection to US exchange brokers, funds are still in a net inflow state, with no large-scale outflows, and trading is still quite active, "it's definitely not at that stage yet."

Harry provided supporting evidence from the cryptocurrency industry perspective: Although retail investor sentiment has been somewhat shaken, larger capital users are still "sweeping goods" against the trend, and it can be observed that capital is continuing to flow out of crypto and into AI-related US stock assets.

Sun offered a different interpretation from the perspective of the entrepreneurial ecosystem, his expression was very straightforward:

"AI adoption is actually growing significantly now, many entrepreneurs are starting to have AI involvement in any project. I think it's just a simple issue now, which is that the numbers just don't add up."

He believes that after the previous overheating and excessive speculation in investments, the market needs to calm down and enter a phase of "seeing how much monetization capability AI companies can really bring."

2. From "selling shovels" to "AI adopter": Who is truly turning AI into money?

In response to Morgan Stanley's judgment on the main theme switch,Investment TALK provided specific targeted directions from a holding perspective: large tech (Amazon, Microsoft) and certain software companies (such as Adobe). He specifically mentioned a change that has been long underestimated by the market — over the past year and a half, the market has generally believed that software companies would be "overturned" by AI, but the financial reports of companies like Microsoft show that if software companies can effectively integrate AI with existing businesses, their revenue growth is actually accelerating, and profit margins have not been significantly impacted. He predicts that the implementation of AI will more frequently occur in the B-end, monetizing through the existing user base of software companies, evolving from a "buyout system" to a "pay as you go" business model.

He also reminded against taking a blanket view on hardware: "The targets adjusted in July need to be viewed separately; some are optical modules, some are chips; as long as there is fundamental support, I remain optimistic about long-term growth."

Elio Cui supplemented from the B2C2C ecological perspective: The bitcoin mining company "Little Deer" under his group has completely transformed into an AI data center (AIDC), symbolizing the shift from bitcoin computing power to AI computing power, while also providing an analysis of the transmission order in the semiconductor supply chain — the storage price increase cycle has not yet completed (capacity has been booked up to after 2027), the upgrade path from 800G to 1.6T to 3.2T for optical modules is still progressing, and CPO/silicon optics are longer-term increments. He provided a standard for judging when the "story is over":

"When will the entire AI infrastructure narrative be completed? It will definitely be when excessive construction leads to a sharp decline in supply-side prices. I believe the construction period is not yet done."

Harry provided two sub-sectors that he believes funds will focus on in the second half of the year: optics/photonics and robotics, emphasizing the need to think along with capital flow.

Sun added two details from front-line entrepreneurial observations: firstly, many software engineers are starting to shift to hardware, and aerospace companies are also using AI to accelerate research and development; secondly, concerns about "whether AI applications will be completely consumed by large models" are being disproven — larger models cannot satisfy these more segmented and specific needs. But he also raised a sharp question: Apart from geeks and technology circle users, the willingness of consumers and enterprises to pay for AI is actually relatively limited. "With all this power generated, everyone’s funding situation is still the same, and the numbers just don’t add up, which is a bit strange."

3. Is the main narrative of semiconductor/power hardware over?

Investment TALK clearly stated that he does not believe the logical line has ended, taking Intel as an example, believing that the company's internal transformation in areas such as optical interconnect can support it through this cycle. However, he also reminded investors to have reasonable expectations:

"This vehicle is too heavy; it will definitely require some people to get off before it can be pulled up more easily. Any targets that experienced overheating will go through this process."

He also pointed out an incremental direction worth noting: The competitive pressure from open-source models in mainland China on the two closed-source models in the US may gradually transfer AI's "value creation" from the closed-source model level to the infrastructure (hyperscaler) and application levels. He particularly mentioned that the market size of the coding sector is about $2 trillion, while the entire software market in the US is $6 trillion, and the real landing space in the application layer is far larger than that of coding.

Elio Cui explained the reasons behind the divergence of "strong fundamentals, weak stock prices" from a more macro perspective, pointing out that the core of the problem is not in the industrial outlook, but in liquidity and credit:

"This round of investment has been too large, relying heavily on debt issuance. If interest rates or credit spreads widen, financing costs will rise, and the market's patience for returns will diminish. Ultimately, it may not be individual investors leveraging but rather the listed companies themselves taking on significant leverage — this leverage is more easily pierced than that of individual investors."

Harry added a data comparison from a valuation cyclical perspective: SK Hynix and Samsung's PE has long been below 10 times, while companies like Nvidia with more software attributes have a PE above 20 times — the market is still assessing whether the storage sector in this cycle is a "super cycle," and he believes this assessment will be difficult to change in the next three to four months.

4. "HALO assets": Why should electricity and energy be revalued?

This is the segment with the highest information density in this discussion.Investment TALK started from the capital expenditure rhythm of large tech companies Meta and SpaceX, pointing out that the computing power bottleneck has shifted from GPU shipments to electricity supply and provided a three-layer power supply path from cheap to expensive: direct supply from the grid → self-generated electricity (BTM, behind the meter, such as Siemens, Caterpillar, GE turbine companies) → more expensive fuel cell solutions (like Bloom Energy). He predicts that the logic line of electricity demand will "only get stronger in the future year to year and a half."

Elio Cui provided the most actionable analysis of the entire discussion — a complete tiered framework for HALO asset investment:

"Electricity is very difficult to become outdated, very hard to be eliminated as infrastructure, which is its biggest difference from GPUs — GPUs become obsolete in three to five years, but power plants, power grids, and transformers can be used for many years."

He provided a priority ranking: The first tier is power infrastructure and electrical equipment; the second tier is nuclear power and natural gas power companies; the third tier is cooling and data center related assets; and the fourth tier is materials like copper and aluminum. He repeatedly emphasized a "first principle":

"Only by being closer to the center and sticking to first principles can you achieve higher investment returns. Those traditional industrial engineering constructions that are more peripheral are distant relatives and not that core."

He also reminded that not all AI infrastructure stocks are still worth buying — the key is to see whether the speed of profit upgrades can outpace the speed of valuation expansion; if the valuation of a stock has already risen too quickly, but forward PE does not have corresponding upward adjustment space, it may instead be an overvalued asset.

Sun provided a forward signal from the perspective of entrepreneurial trends: energy is becoming a new startup hotspot, from geothermal power to "distributed solar energy" for "every household," "electricity is always eternal."

5. China's version of digital infrastructure: Where are the opportunities in open source + low costs?

Investment TALK frankly admitted that he does not cover A-shares but provided a logic from the perspective of US stocks: if China's open-source models do not substantially impact the acceptance of American closed-source models, it is actually a good thing for American cloud providers — because the cost of APIs is the cost for software companies; the more tokens consumed (even if they are cheaper), ultimately, the ones benefiting are still the cloud providers that bear these loads.

Elio Cui proposed a dialectical analysis of the "bear market logic" and "bull market logic" of HALO assets: model efficiency improvements (cheaper inference costs) combined with excessive infrastructure construction will theoretically depress computing power prices, which is the biggest bear market risk for HALO assets; however, he leans towards an optimistic judgment because AI demand is growing exponentially; even if supply-side efficiency continues to improve, demand growth may still outpace the rate of efficiency improvement, similar to the historical path of the IT/internet industry where "performance improvements and excess infrastructure led to increasing penetration rates."

Sun contributed the most information-rich section from observations in his entrepreneur community, worth highlighting separately. He mentioned that the core advantage of Chinese AI models is not just being cheap, but more importantly, being open source:

"Different countries and even different universities are asking whether we can reach a cooperation with Chinese AI model companies to establish joint research institutes. The reason is that many countries realize that the US is not a long-term stable partner; the data security and embedded values of closed-source models raise concerns."

He mentioned a specific case: An AI safety and privacy research institution in London initially thought that Chinese AI performance was not sufficient, but after Kimi K2 gained popularity, they began to reassess and seek collaboration with Chinese open-source models to build their own country's "Sovereign AI." He also mentioned regions like Africa and Latin America, which have very low AI penetration rates but a large number of entrepreneurs and users; the subscription prices of closed-source products like OpenAI (around $100-200 per month) are too high for local users, and the combination of Chinese open-source models + local photovoltaic expansion creates a low-cost infrastructure mix that has more opportunities for penetration. He provides a judgment over a five-year dimension:

"If OpenAI and Anthropic remain closed-source — they likely must remain closed-source, as the business model won't work otherwise — then in the future, more and more countries will establish their own sovereign AI, most of which will choose Chinese open-source models because they are cheap, efficient, and can be personalized."

6. Capital expenditures and the Federal Reserve: When will the divergence of strong fundamentals and weak stock prices converge?

Investment TALK used a previously contradictory example from market logic to illustrate the current misalignment: A month and a half ago, the market was debating whether Micron had already "decycled" (should be given a 15-20 times valuation), but at that time, the market simultaneously disliked giving money to large tech companies that "issue" to Micron — this is logically not valid because whether storage manufacturers can escape cyclical pressures ultimately depends on whether the capital expenditures of large tech truly yield returns. He judges that the latest financial reports from Microsoft, Amazon, and Google have begun to verify this, and the turning point in market sentiment has already appeared in the earnings season from two weeks ago. He also provided a critical deduction:

"If capital expenditures can be maintained in the range of $1.6 trillion to $1.8 trillion in 2027, this growth rate is acceptable, provided that this money can really be earned back. Once capital expenditures decline dramatically, the logic of storage will also collapse — it fundamentally remains a cyclical stock."

Elio Cui explained why the AI sector is so sensitive to liquidity from a perspective of interest rate discounting:

"AI is a typical high-valuation, long-cycle growth stock; a significant part of its pricing is based on the discounting of future cash flows. Whenever interest rates move, the elasticity on the valuation side is much larger than the changes on the performance side."

Regarding the question of a potential 25 basis point rate hike in the fourth quarter,Investment TALK judges that rate hikes are more "market noise" rather than a mainline factor — Unless the Federal Reserve actually reopens the rate hike cycle, the long-term fundamental impact on the AI sector is limited; Elio Cui emphasized that interest rates also greatly impact corporate debt issuance costs, and thus the rhythm of capital expenditure; Sun's judgment leans more toward the application end: "The Federal Reserve's interest rates essentially affect the financing situation of AI companies and new companies; the phase of purely narrative and storytelling is probably already past, and the next contest is who can actually build cash flow business."

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。