Which stocks do you firmly看好 following 2026?

We are relatively bullish on SDEK. “SDEK has achieved substitution of Japanese giants—including Toray, Fujifilm, Nittodenko, and Mitsubishi—across multiple domains thanks to its vertically integrated industrial chain and continuous technological breakthroughs. It stands among the rare domestic material platforms capable of competing head-to-head with Japanese firms simultaneously in three strategic sectors: optical display, electronic components, and new energy. With China’s countermeasures against Japan taking effect and domestic substitution accelerating, the company is poised for dual growth—in both earnings and valuation—earning it the moniker ‘the Huawei of materials.’” “On January 23, 2026: The company officially announced that its subsidiary had introduced a strategic shareholder (Huawei), signaling capital support for accelerated commercialization of its magneto-electric storage business.” A market capitalization of ¥50 billion appears quite certain at present; further upside would require additional catalysts or earnings releases. For the tech sector, it is genuinely difficult to discern whether a given opportunity is merely hype—or truly executable. Even if execution proves feasible, the magnitude of resulting earnings remains uncertain. Thus, buying tech stocks at low valuations inevitably involves an element of speculation. If one wishes to avoid speculation, waiting until full certainty emerges before entering positions makes achieving outsized returns exceedingly difficult.

Momenta found Fairy Island.

On January 29, 1886, German engineer Carl Benz patented his ‘Benz Patent Motor Car,’ marking the birth of the first modern automobile possessing practical utility and commercial viability. Over the past 140 years, countless engineers have relentlessly refined and transformed the automobile. Every physical component—from engines to brakes—has undergone revolutionary evolution, propelled by advances in engineering and Moore’s Law. Yet the ‘core component’ seated behind the steering wheel—the human driver—has remained virtually unchanged since the automobile’s inception, constrained by innate physiological and cognitive limitations. This harsh reality is reflected in the approximately 1.3 million annual global road traffic deaths. Cao Xudong’s father was involved in a serious traffic accident—an experience that became one of Cao’s key motivations for entering the field of autonomous driving. At Momenta’s founding, he and his team established three decade-long visions; the top priority was: ‘Within ten years, save one million lives.’ It is precisely this profound reverence for life that has embedded safety into Momenta’s corporate DNA. Yet enabling true vehicle safety remains extraordinarily challenging—especially teaching the car to understand the physical world. In recent years, digital AI has advanced rapidly: large language models (LLMs) can now absorb ‘common sense’ about the digital world from vast internet text corpora. The physical world, however, offers no pre-existing, neatly annotated corpus. The trajectory of a rolling apple, subtle shifts in lighting along a highway, or the minute sway of a cyclist’s shoulder—none of these phenomena appear in any textbook. To tackle this challenge, Momenta’s solution is: reinforcement learning + world models. Jeff Hawkins posits that intelligence fundamentally arises from prediction. This logic is straightforward: just as LLMs compress knowledge of the digital world by predicting the next word—and thereby acquire text and natural language understanding—breakthroughs in physical AI must similarly emerge from predicting ‘what will happen next.’ By compressing the laws governing the physical world into models, AI can learn to forecast future states and interaction dynamics, gaining intuitive grasp of object properties, causal motion relationships, and potential interactions. Lift a cup, then release it—the cup falls. Humans intuitively grasp this through millions of years of evolution, encoded directly into our cerebral cortex; even without knowing Newton or gravity, we understand it instinctively. Machines lack such innate intuition and must instead acquire analogous common sense via world models trained on massive real-world data. Xia Yan, Momenta Partner and Senior Vice President of R&D, explains that during world model pre-training, the system first ‘internalizes’ physical laws, common sense, and causal relationships embedded in vast volumes of real-world driving data—forming the model’s foundational understanding of the physical world. Based on this foundation, the world model is deployed within closed-loop autonomous driving simulations, enabling the system to project how the world will evolve in response to changes in its own behavior. Leveraging highly efficient scenario simulation capabilities, the system evaluates performance on long-tail edge cases—and ultimately conducts reinforcement learning within this high-fidelity virtual environment, allowing the system to evolve through repeated trial-and-error under near-realistic conditions. Sun Gang, Momenta Partner and Senior Vice President of R&D—and Cao Xudong’s colleague—conducted a small-scale experiment a decade ago.