--- title: "Understanding the \"five-layer cake\" of AI reveals the secrets to wealth over the next decade." type: "News" locale: "en" url: "https://longbridge.com/en/news/278825926.md" description: "The article by Chen Xiaomeng discusses the evolving landscape of AI, emphasizing its significance as a fundamental social infrastructure rather than just a tool. It introduces the \"five-layer cake\" model of AI, highlighting the importance of energy, chips, and the transformation of computational logic. The first layer focuses on energy, stressing the need for stable, cheap electricity for AI development, with investment targets including Vistra and Constellation Energy. The second layer addresses chips, which are crucial for efficient power conversion, with ASML identified as a key player in chip technology. The analysis aims to uncover investment opportunities in the AI sector over the next decade." datetime: "2026-03-12T05:45:35.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/278825926.md) - [en](https://longbridge.com/en/news/278825926.md) - [zh-HK](https://longbridge.com/zh-HK/news/278825926.md) --- # Understanding the "five-layer cake" of AI reveals the secrets to wealth over the next decade. Author: Chen Xiaomeng; Source: X, @MengLayer The market's understanding of AI is still too superficial. Many people simply see it as a useful little tool or a smart chat software, which is a huge mistake. In Lao Huang's vision, AI has long transcended the realm of simple software. It has completely become a fundamental social infrastructure like electricity and the internet. He has extremely precisely divided this ongoing computing power revolution into "five layers of a cake." By understanding these five layers, you can basically keep a close eye on the flow of global capital, computing power, and talent over the next decade. Below, I will provide a detailed analysis (approximately 10,000 words) combining the investment logic of the capital market. To understand future investment directions, we must first clarify what fundamental changes have occurred in computational logic. For the past few decades, the software world was entirely pre-defined. Programmers wrote code, stored data in tables, users input commands, and computers dutifully "retrieved" and output results. SQL databases, with this access logic, dominated the previous era. Now, the rules of the game have completely changed. For the first time in human history, computers possess the ability to directly understand images, sounds, text, and even extremely ambiguous contexts. When you ask it a question, it completely skips the outdated step of searching for ready-made answers in a database. It directly "reasons" in real time and generates a completely new result on the spot. Since intelligence has become a new type of commodity that can be produced by machines in real time, the entire underlying physical architecture supporting it must be completely overhauled. The grand feast of capital lies hidden in this process of overhaul and reconstruction. Let's slice open Lao Huang's five-layer cake layer by layer and analyze the core wealth secrets of each layer. The first layer: Energy – the end of computing power is electricity, and the power grid determines the upper limit of intelligence. This is the most fundamental physical foundation of everything. Here, no software code can be cleverly manipulated; everyone must obediently bow to the laws of physics. Every time you press Enter in the dialog box, every token spit out by the large model is essentially a surge of electric current and a sharp rise in chip temperature. The process of machines generating intelligence is an extremely violent energy conversion process. Energy is the most fundamental first principle for AI development. Whether there is a sufficient scale of cheap electricity, whether the local power grid can handle it, and whether the huge heat dissipation problem can be solved—these purely physical conditions severely limit how much "intelligence" humanity can produce. Currently, AI data centers cannot be built casually; they must be extremely close to large, cheap, and extremely stable energy stations. This is why top tech giants are scrambling to acquire nuclear power plant capacity worldwide. The computing power race, in its later stages, is essentially a national-level electricity race. Investment Target Reference: Vistra (VST) / Constellation Energy (CEG): The absolute leaders in North American nuclear power and independent power producers. Large companies are building extremely power-intensive AI data centers, which traditional power grids simply cannot handle. Nuclear power, with its stable and clean baseload electricity, has become an irreplaceable commodity. Tech giants are lining up to buy electricity with multi-billion dollar long-term contracts to secure their computing power advantage. GE Vernova (GEV): A global super-giant in power generation equipment and grid infrastructure. Whether you use natural gas, wind power, or nuclear power, to deliver massive amounts of electricity to AI data centers with extreme stability, an epic upgrade of the existing aging power grid is necessary. GEV is the ultimate "shovel seller" providing transformers, gas turbines, and modernized power grid equipment. The second layer: Chips—Super Energy Converters and the Crown Jewel of Human Industry. With sufficient electricity, the next core issue is how to convert that electricity into computing power most efficiently. AI's computational demands are extremely brutal; it requires incredibly high parallel processing capabilities. At this level, simple computing power is absolutely insufficient; high-bandwidth memory (HBM) and ultra-fast interconnect channels between chips are the real deciding factor. The capabilities of a single graphics card are extremely limited; the real magic lies in connecting hundreds of thousands of graphics cards into a giant cyber brain. This requires transmitting massive amounts of data every second. Once the data transmission channel is blocked, even the most powerful computing power can only spin in vain. Every iteration of chip architecture and manufacturing process directly determines the life-or-death speed of the entire industry's expansion. The technological and financial barriers here are the highest in human history. Investment Target Reference: ASML: The crown jewel of the entire silicon-based industry. It monopolizes the global high-end EUV lithography machine market. Without ASML, this behemoth costing hundreds of millions of dollars, no matter how strong the foundry's technology, it simply cannot produce the most advanced AI chips. It is the ultimate gatekeeper that holds all advanced manufacturing processes in check. NVIDIA (NVDA): The undisputed global computing power leader. It offers far more than just chip hardware; it includes a complete set of system-level standards, including the CUDA ecosystem and NVLink interconnect technology. No matter how the underlying model changes, if you want to develop large-scale models, you must obediently pay Nvidia's "computing power tax." TSMC (Taiwan Semiconductor Manufacturing Company): The world's most powerful wafer foundry. No matter how fiercely Nvidia and AMD compete, or even with Google and Microsoft developing their own chips, they all ultimately have to queue up with their blueprints and have TSMC manufacture them. It monopolizes the most advanced 3nm process and CoWoS advanced packaging capacity, firmly controlling the silicon-based foundation of the entire AI era. The third layer: Infrastructure – Super factories manufacturing "intelligence," a new type of commodity. Please completely erase the traditional concept of a data center used for "data storage" from your mind. What's rising from the ground now is a true "AI factory." This involves unimaginable amounts of land permits, dedicated power supply connections, extremely complex direct chip liquid cooling systems, and tens of thousands of fiber optic cables spanning the data center. Seamlessly connecting hundreds of thousands of top-tier graphics cards and ensuring their stable, error-free operation around the clock is an epic civil engineering feat combined with an extremely complex systems engineering challenge. These super factories are born with only one pure mission: to consume electricity day and night to manufacture "intelligence," a new type of industrial product that can be infinitely replicated. This unprecedented infrastructure boom is driving a frenzy of hardware upgrades across the board. Investment Target Reference: Super Micro (SMCI): A major assembler of core AI servers. After NVIDIA manufactures its chips, it needs an extremely strong hardware engineering team to perfectly assemble the chips, motherboards, and power supplies into deliverable, high-density AI racks, which are then directly installed in the data centers of major manufacturers. Arista Networks (ANET): The irreplaceable network nervous system for AI clusters. Previously, people were accustomed to using NVIDIA's own InfiniBand network, but now Ethernet technology is becoming increasingly popular in AI clusters. Arista is the king of high-speed Ethernet switches; both Meta and Microsoft relied heavily on its network equipment to ensure data flow when building supercomputers. ## The fourth layer: The model—the digital brain that is devouring and understanding the entire physical world After the infrastructure is built, it's time for the large model we are most familiar with to take the stage. ## Don't mistakenly believe that Large Language Models (LLMs) are the end goal; language is merely a small entry point for machines to understand the world. The real disruption is spreading rapidly throughout the entire physical world. The most advanced models are now frantically absorbing and learning from biology, chemistry, meteorology, physics, and embodied robotics. Protein folding and the discovery of new materials are being compressed to the second level. Furthermore, the disruptive power of open source cannot be ignored. In the past year, open-source models have completely crossed the performance threshold. Huang specifically mentioned DeepSeek-R1 in his article. The complete open-sourcing of this inference model has directly delivered cutting-edge complex logical reasoning capabilities to developers worldwide for free. The moat of large models has been ruthlessly breached, forcing the entire industry to shift from focusing on model parameters to real-world applications. Investment Target Reference: Alphabet (GOOGL): As an absolute pioneer in the AI ​​field, Google possesses terrifying full-stack capabilities. It holds the most powerful multimodal models in the Gemini series and is aggressively expanding its developer ecosystem through the Gemma open-source model. More critically, Google has seamlessly integrated these top-tier AI capabilities into search, Android, and YouTube—super-cash cows with billions of users—building an unparalleled data and distribution moat. Microsoft (MSFT): The king of commercial applications of closed-source models. It has deeply integrated the strongest capabilities of OpenAI and cleverly embedded them seamlessly into Office 365 and Azure cloud services. In the enterprise market, where data security is paramount, Microsoft's moat is unfathomably deep. The fifth layer: Applications—the ultimate monetization pool of trillions of dollars in capital expenditure. This top layer is the true commercial arena where monetization can take place. The trillions of dollars burned in the lower four layers are all expected to be recouped here. Autonomous driving, AI drug development, humanoid robots taking over production lines, fully automated super programmer assistants… the same underlying architecture has evolved into countless disruptive business models at this layer. This layer is also the true ultimate engine of the entire AI ecosystem. Here, Jensen Huang proposes a crucial principle of business dynamics: "pull-down force." Once a killer application emerges at Layer 5 (such as an AI that can perfectly replace human lawyers in reviewing contracts), it will instantly create a massive demand black hole. This black hole will frantically devour the model reasoning capabilities of Layer 4, forcing Layer 3 to frantically expand its AI factories, ultimately passing the extreme pressure of computing power down layer by layer to the underlying chip foundries and nuclear power plants. Investment Target Reference: Palantir (PLTR): A top-tier hardcore player in B2B AI applications. While others are still developing toy-like chatbots, it has already directly integrated its AI operating system into the US military's combat command system and the top-secret production lines of Fortune 500 companies. Its AI platform helps companies directly transform the intelligence of large models into extremely real production efficiency and huge profits. Tesla (TSLA): If most tech companies dominate AI in the digital world, then Tesla is firmly targeting AI in the real physical world. Relying on the high-quality video data generated daily by millions of cars on the road, Tesla possesses an incredibly formidable real-world data barrier and edge computing power advantage in the fields of end-to-end autonomous driving (FSD) and Optimus humanoid robots. It is essentially a hardcore AI giant disguised as a car company. Summary: This five-layered pie far surpasses ordinary technological cycles. It is a complete industrial revolution reshaping energy distribution, redefining factories, and disrupting human organizational structures. Each layer is tightly interlocked, pulling each other in an extremely frantic manner. To support this colossal, rapidly spinning flywheel, we are witnessing the largest and most costly infrastructure construction frenzy in human history. Hundreds of billions of dollars have already been poured into it globally, yet the shortfall remains trillions. And this is truly just the beginning. Most of the infrastructure is still on paper, and most of the era's benefits have yet to be realized. Don't resist this epic trend; follow these five layers of logic to find your own niche. ### Related Stocks - [SOXX.US](https://longbridge.com/en/quote/SOXX.US.md) - [CEGX.US](https://longbridge.com/en/quote/CEGX.US.md) - [SMH.US](https://longbridge.com/en/quote/SMH.US.md) - [ASMH.US](https://longbridge.com/en/quote/ASMH.US.md) - [ASML.US](https://longbridge.com/en/quote/ASML.US.md) - [CEG.US](https://longbridge.com/en/quote/CEG.US.md) - [ASMG.US](https://longbridge.com/en/quote/ASMG.US.md) - [VST.US](https://longbridge.com/en/quote/VST.US.md) ## Related News & Research - [32% TSMC Spending Surge: Massive AI Wave Heading Straight For ASML](https://longbridge.com/en/news/282875686.md) - [Ritholtz Wealth Management Acquires 1,146 Shares of ASML Holding N.V. $ASML](https://longbridge.com/en/news/282414571.md) - [Venturi Wealth Management LLC Purchases 767 Shares of ASML Holding N.V. $ASML](https://longbridge.com/en/news/282416258.md) - [Why ASML stock popped today](https://longbridge.com/en/news/282086302.md) - [This Magnificent Energy Stock Is Down 20%. Buy It Before It Sets a New All-Time High.](https://longbridge.com/en/news/282853242.md)