Just finished reviewing the various financial reports from the 2025 annual report season, and I can’t help but feel a bit emotional. The difference this time from previous years is that entrepreneurs are starting to explain their understanding of this new world in clear terms, rather than still figuring out how to “get through” the days ahead.

The most straightforward manifestation is what’s called K-shaped divergence. In its annual report, Midea put it very plainly: “K-shaped divergence is accelerating the reshaping of the fate of nations, industries, companies, and individuals.” This isn’t a new concept, but seeing it expressed so extremely at the industrial level really is a new feature this year.

On one side, real estate has basically collapsed. Last year, Vanke lost nearly 100 billion. Not only did its core business fail to generate profits, even its assets shrank. The company itself was candid: land prices were taken at too high a cost, sales gross profit margins were too low, and its operating businesses were also losing money. Even more painful, even large-scale assets could only be sold at prices below their book values. The situation at Maike Long, a home renovation and building materials mall, isn’t much better either. Its loss of more than 20 billion mainly came from a sharp decline in the fair value of investment properties. Even MTR’s flagship commercial project in Shenzhen was reportedly unable to attract bids. A mall with a residential complex on top that had already sold out—now nobody wants to take it over.

But on the other side, upstream resource companies and the AI industry chain are running hot. Zijin Mining issued a $1.5 billion zero-coupon convertible bond, with a conversion premium as high as 40%, and investors are chasing to pour money in. Xinyisheng discussed the demand for optical modules driven by ultra-large-scale AI computing clusters. Even Haomai Technology, which makes tire molds and gas turbine components, is lamenting that the gas turbine capacity of GE, Siemens, and Mitsubishi Heavy Industries has already been booked through 2028 and 2029—all of it driven by the explosive demand from AI data centers.

So what does this reflect? I think the keyword has changed—from an efficiency logic to a security logic.

Zijin Mining wrote it very clearly in its annual report: “Key minerals have risen from being purely economic factors to becoming the focus of national security and great-power competition.” The old model of globalized division of labor has already started to fall apart. In its place are regional closed loops, localization, and green compliance. Safety first, resilience second, efficiency third.

For companies going overseas, a new trade paradigm is also right in front of them. Haier Smart Home put it plainly: “The old trade paradigm has already left; tariff games make it impossible to develop supply chain layouts in a way that is not fragmented or inefficient.” Unicom Electronics pointed to a key term—friend shoring. Globalization may not be dead yet, but everyone is starting to pick their “friends.”

Then what should companies do that aren’t on the AI computing infrastructure track? I think Li Lu’s concept about BYD is quite enlightening: becoming a “global company with Chinese nationality.”

This requires localization on three levels. The first level is manufacturing localization—this is something everyone is already doing: building factories abroad and expanding capacity. But Shenzhou International has warned later entrants that overseas labor costs will rise, efficiency may be lower, and tariffs still can’t be avoided. So exporting capacity alone is not enough.

The second level is localization of the supply chain and industrial clusters. SF Express talked in its annual report about how it helps tea beverage brands go global—not only handling customs clearance and compliance, but also integrating export needs across multiple customers to improve load factors, and providing end-to-end supply chain services. Yutong Bus’s overseas revenue already accounts for 60%, and it relies on the entire new energy industry chain—starting from Ganfeng Lithium’s lithium mines, to CATL’s batteries, and then to component manufacturers—all realizing localized layout. Only when industrial clusters themselves are localized can companies truly become part of local economic development.

The third level is R&D localization, which is the newest development. BeiGene has built clinical teams of 3,800 people across six continents. Innovent Biologics uses the “Eastern and Western dual-engine” R&D advantages. Electronic manufacturing service providers are also beginning to get deeply involved in brand customers’ product planning and design. This is a deeper level of cooperation and integration.

In its annual report, Mindray Medical said something very interesting: “What de-globalization brings is not closure, but deeper localization.” Indeed, under the new trade paradigm, the survival path for overseas-going companies is to become more localized and more deeply rooted.

But I also see another phenomenon. In its annual report, CATL emphasized not AI enablement or full-stack ecosystems, but that “subtle defects will be amplified infinitely over time and space,” stressing that quality is the lifeline. This shows that one of the winning logics of China’s manufacturing is to push physical details to the extreme. At the same time, CATL said that batteries should become “the basic unit supporting energy system buffering, stability, and dispatch.” Weichai discussed that data center backup power is new infrastructure. Haomai discussed that gas turbines are new infrastructure. In China, AI is being physicalized, and manufacturing is repositioning itself as the foundational unit of the AI era.

This kind of reshaping has its benefits, but it also comes with costs. If China’s AI story always stays at “how much power we can supply to data centers,” then it risks repeating the mistake of the smartphone era—where the most generous profits were taken by Apple and global software platforms, leaving us to just drink the scraps from the manufacturing link.

So what is truly needed is more DeepSeek, more BeiGene, and more companies that can lead this wave of technological revolution right from the R&D stage. But that requires breaking through structural constraints such as capital, talent, and market access—and these problems can’t be solved in just three or five years.

The reshaping reflected in the annual report season is, in fact, still not deep enough. What we need is source-driven innovation that brings improvements in profit margins, not merely growth on the revenue side. The most poignant words in Midea’s closing remarks, in my view, are: “Business failure is common; the vast majority of companies will ultimately become ordinary or mediocre. The ultimate ability of an enterprise that endures is not growth, but continuous restarting or regeneration in place.” The cost of prosperity is perpetual unease; destruction fuels growth, and stability is only an illusion.

That is what this year’s annual report season feels like to me.