This marks the 1277th original article from 'New Energy Frontier.' Simply click on 'New Energy Frontier' at the top to follow and 'star' this account for updates. Please note that the article reflects the personal views of 'New Energy Frontier' and should not be considered investment advice. The author does not manage investment groups, recommend stocks for a fee, or handle client funds.

--------

The global capital market over the past two years has undoubtedly been fueled by AI, with NVIDIA playing a pivotal role. An illustrative image captures this trend, showcasing the AI industry chain in the midst of a powerful super bull market.

01 The Hard Core Remains Strong, the Software Fluctuates

However, a closer examination reveals that, in this recent market trend, AI hardware has significantly outperformed AI applications. This is particularly evident in the Chinese capital market, where AI is predominantly software-centric—hardware stands firm, while software falters.

AI applications have been metaphorically labeled as 'playboys,' occasionally making a splash before fading into obscurity.

Logically, AI applications represent the ultimate goal of AI hardware, with hardware, exemplified by computing power, serving merely as the foundation. Yet, judging by capital market performance, it appears the market has lost sight of its original purpose, with notable disparities in secondary market performance between hardware and software. Why is this happening?

In light of this stark contrast, many investors bullish on AI applications firmly believe the market has made a mistake. Their reasoning is straightforward: 'Don't lose sight of our initial goals.' If there are no opportunities in the application segment, theoretically, there should be none in the hardware segment either!

Leo often remarks that while the market may occasionally experience temporary inefficiencies, it is generally effective. If your holdings consistently underperform the market, don't always blame the market; instead, consider what the market might be thinking at that moment.

The stronger performance of the hardware segment in the AI industry chain compared to the software segment may seem irrational at first glance, but it actually mirrors the capital market's pricing logic based on future performance certainty. Over the past two years, the hardware segment has demonstrated significantly stronger performance certainty than applications.

As for the grievances of many application investors, remember our initial goals. Without the flourishing of applications, hardware like computing power would be like water without a source or a tree without roots.

However, this implies a prerequisite: the market assumes AI is another historic revolutionary wave promoting productivity growth, following the industrial and information revolutions.

In other words, both software and hardware in the industry chain will eventually have opportunities. But in terms of investment, pricing starts with the most certain assets. In this regard, the certainty of the hardware segment far exceeds that of the application segment.

Computing power is the cornerstone of the AI revolution. If AI truly represents a historic wave, then we are just at the beginning. As applications become increasingly prosperous, demand for computing power will only surge. Additionally, the competitive landscape in the hardware segment, especially in the chip industry chain, is relatively clear, with some areas even described as 'a bright moon amidst sparse stars.' The Chinese chip industry also presents opportunities for domestic substitution. Why wouldn't the market focus on it?

02 The Uncertain Historic Opportunity of Large Models

Conversely, the application segment is fraught with uncertainties.

Theoretically, large model companies stand to benefit the most. However, the race is still in its early stages. While some large models have made rapid progress, significant uncertainty remains regarding who will ultimately prevail due to rapid technological iterations.

For instance, OpenAI initially dominated the large model market, with a market share exceeding 80%. However, with Gemini quickly catching up technologically, its market share has gradually declined to below 60%. Considering Google's early involvement in AI, its strong technological pedigree, and its super entry points like the Android system, Google's ecosystem, and being the default search engine for Apple, Gemini is almost certain to surpass OpenAI in the long run, provided it remains technologically competitive.

Another example is Baidu, which was an early leader in China's AI sector. A few years ago, Lu Qi, upon taking office, declared an 'all-in AI' strategy. Unfortunately, he left after a year. Despite Baidu's continued high-intensity investment in AI, including large models and autonomous driving technology, which once led domestically, Huawei's rapid rise in autonomous driving and DeepSeek's overnight success in large models have turned Baidu's early advantages into a joke. The replacement of traditional search engines by large models is almost certain. Having lost its competitive edge in large models and its internet entry point, Baidu's future seems clear.

DeepSeek further illustrates the brutal competition in the large model sector. Despite its initial explosive popularity, often hailed as a product of national significance for China, DeepSeek has significantly lagged behind Doubao within a year.

Are there no opportunities for other products? Not necessarily. For example, Kimi's K2.5 recently topped the global open-source rankings for both code and visual capabilities on the LMArena leaderboard. It was also the first main model recommended by OpenClaw official and has consistently ranked first in OpenClaw's usage volume since February 4, surpassing Gemini 3 Flash and Claude Sonnet 4.5.

Given the immense resource demands of large models, the competition, while not as brutal as the 'Hundred-Group Battle,' remains fierce as giants view it as the next revolutionary opportunity. Currently, Google, Meta, Alibaba, ByteDance, and startups like OpenAI, Moonshot AI, and DeepSeek are locked in competition. However, Apple, Microsoft, Amazon, and Tencent have yet to fully unleash their potential, indicating that the competition in the large model sector is far from over.

So, as an investor, how would you choose in the large model sector?

Of course, vertical large models that have truly achieved closed-loop operations are also highly sought after by the market, such as Anthropic, which gained significant traction in the primary market due to Claude, and domestic companies like Zhipu and Minimax that recently entered the capital market.

If large models are already this torturous, the downstream applications are even more so.

03 The Misconception of AI Application Benefits

Consider the two waves of AI applications pursued by the market this year: advertising agencies and GESEO.

The market believes they directly benefit from technological advancements in large models, but is this really the case? Or is the extent of their benefit as high as the market pursues?

GESEO is even more laughable, as it was originally a primary target for large models to reduce hallucination rates, yet the market treats them like treasures.

After the recent viral success of Seedance 2.0, the market has sent film and IP sectors soaring. However, do these sectors truly benefit as much as the market believes from the improved film and television creation capabilities of large models? This requires careful consideration.

While Seedance 2.0's advancements in video creation capabilities seemingly offer film companies opportunities to reduce production costs and improve profit margins, a further analysis reveals that lower production costs also lower the barriers to film production, potentially turning the industry into a highly competitive one. Where then lies the advantage of traditional film production companies?

Similarly, while IP may seem to benefit from large models' enhanced video creation capabilities, as many IPs now have the potential for film and television adaptation, a further step back reveals that improved video creation capabilities lower the barriers, allowing anyone with a creative idea to produce excellent video content and potentially become a new IP. This raises questions about the value of previously established IPs. However, with an infinite supply of content, the scarcity value of well-known IPs will become prominent again, leading to a new valuation basis for IPs.

Consider software companies. Initially, the market believed many SaaS companies would benefit from technological advancements in large models, as they could reduce development costs, with savings translating into profits. However, as large model technologies continue to advance, the market sees another possibility: with infinite progress in large model technologies, software may become unnecessary, as a single general-purpose model could handle everything. This has led to the notion that general-purpose large models will devour everything, causing software stocks to be abandoned by the market.

But is this correct? Theoretically, considering the endgame, it may be right, but it's challenging in the short term, and many niche areas may require further deliberation.

What about companies like Tencent and Apple, which control entry points? Are they truly that secure? Looking at the endgame, it's hard to say. Who knows if AI advancements will lead to better hardware and entry points?

So, as you can see, AI applications are fraught with complications, leading to endless reversals and layers of complexity upon closer inspection.

Given the uncertainty of AI applications, how can the capital market price them? Their certainty is far lower than that of the hardware segment, so why not focus on the more certain hardware segment?

After all, the market is ultimately effective.

Of course, this does not mean AI applications have no opportunities. On the contrary, the opportunities in AI applications are far greater than those in hardware, and they are gradually starting to manifest. In the future, there will be batches of opportunities tenfold that of Tencent, but it's still unclear which specific assets will seize them. Now is the time for gradual exploration and discovery!

If you are also interested in the enterprises I follow and the summaries of industry expert exchanges, you can join 'Value Treasury' to view them.