<\/div><\/div>AI PC Era: AI Phase 2 Explosion – Trillion-Dollar Wealth Flow in Next Decade<\/p>\n
On June 1, 2026, COMPUTEX Taipei opened. That same day, NVIDIA GTC Taipei was held simultaneously. Jensen Huang took the stage again. This time, in the eyes of Wall Street institutions, he is no longer just a tech company CEO but the 'pace controller' of the entire AI cycle.<\/p>\n
Over the past three years, the market has repeatedly verified a rule: whenever Huang releases key signals at important occasions, capital flows in the AI sector shift rapidly, institutional positions restructure, and the valuation logic of the entire supply chain gets recalibrated.<\/p>\n
This time, his key signal is the new PC era. NVIDIA is no longer satisfied with just selling GPUs; it is trying to control the AI computing entry point for the next generation of Windows. From high-performance training and inference computing in data centers, extending all the way to personal computers, laptops, and even every end smart device.<\/p>\n
If AI shifts from pure 'cloud computing' to 'end-side entry' dominance, will the pricing logic of the entire capital market be completely rewritten? Who will be the real long-term winners? Those well-known tech giants, or infrastructure companies that have not yet been fully priced in?<\/p>\n
Today, I'll give you a comprehensive breakdown from the perspectives of capital flows, supply chain restructuring, and practical actions for ordinary investors. Combining the latest public reports from IDC, IEA, TrendForce, and others, I ensure every step is clear and easy to understand, helping you see this structural opportunity.<\/p>\n
First, let's talk about the deeper meaning of this signal.<\/p>\n
Over the past three years, every major statement by NVIDIA has triggered systematic institutional position adjustments. This is not short-term sentiment, but rational judgment based on long-term trends, from upstream chip design, to midstream supply chain, to downstream applications, re-laying out the entire chain.<\/p>\n
At the core of this 'new PC era' are NVIDIA's N1X and other ARM architecture AI PC chips. The goal is clear: bring AI computing power directly from the cloud to users' local devices, allowing ordinary people and enterprises to efficiently run local large model inference, real-time AI assistants, image generation, and more on their own computers without relying on the cloud every time.<\/p>\n
In the past, the competitive focus was on GPU performance parameters: whoever had stronger computing power and higher energy efficiency won. Now, the competition dimension has upgraded to 'who can control the complete computing entry point.' Whoever can build an end-to-end closed loop from cloud to client will master software ecosystem integration, data privacy control, and future capital expenditure direction.<\/p>\n
This is no longer a hardware specification competition, but a platform-level and ecosystem-level game. Controlling the entry point means controlling how AI is perceived, used, and creates value by end users. Imagine, before, AI was like a powerful central kitchen; now it wants to walk into every household's dining table, allowing everyone to 'cook' anytime, anywhere. The impact of this transformation will far exceed our current imagination.<\/p>\n
NVIDIA remains the absolute core. After the large-scale deployment of the Blackwell platform, the next-generation Vera Rubin platform will continue to improve data center computing efficiency and energy efficiency. But the institutional perspective has shifted; they no longer focus on a single GPU leader, but evaluate the complete computing delivery capability from 'cloud to client.'<\/p>\n
AMD is pushing the MI series as an alternative, Intel is guarding the traditional PC entry point. But the logic has changed: institutions no longer pick a single winner, but buy systematic solutions. Whoever masters the complete path from large-scale cloud training, edge inference to client-side local acceleration will take a larger share in the next round of AI capital spending.<\/p>\n
According to the latest forecast from TrendForce, global AI server shipments in 2026 are expected to grow over 28% year-on-year, far exceeding the overall server market growth of about 12.8%. Behind this is the continuous increase in investment by cloud service providers, with the rise of AI PCs further driving client-side demand.<\/p>\n
The Nasdaq index acts as an amplifier here. The index weight is concentrated on a few leaders, forming a positive feedback loop: leaders rise → index rises → more capital inflows. This mechanism allows full-stack computing companies to get valuation premiums beyond fundamentals.<\/p>\n
For us ordinary investors, what is the insight from this main line? The opportunity does not lie in betting all on a single hardware, but in finding companies that can integrate upstream and downstream and deliver complete solutions. Their moats are deeper and their cycle adaptability is stronger. For example, if you only buy a single GPU company, you may miss the entire ecosystem dividend from training to inference; investing in full-stack players is like buying a long-term ticket covering the entire AI lifecycle. At this node of 2026, this full-stack thinking is increasingly becoming an institutional consensus.<\/p>\n
This line is often underestimated by ordinary investors, but in institutional models, it is the 'blood circulation system' of AI expansion. According to the latest IDC data, global DRAM market revenue in 2026 is expected to reach approximately $418.6 billion, a significant increase from the previous year, with the entire memory-related market size approaching $595 billion.<\/p>\n
AI data centers consume about 70% of global high-end memory, especially HBM, which is in severe short supply. HBM production capacity of suppliers like SK Hynix, Samsung Electronics, and Micron is basically booked up through the end of 2026 or even further, with some major customers locking contracts years in advance.<\/p>\n
Why long-term tight balance? Because AI-driven memory demand has shifted from cyclical fluctuations to long-term structural growth. Data centers need massive HBM to support training and inference, and the popularity of AI PCs further amplifies the demand for high-bandwidth, high-capacity memory for local inference. Running local large models requires faster access speeds and larger capacity for low-latency experiences.<\/p>\n
This makes memory a core asset of AI infrastructure. As long as AI capital spending continues, demand will steadily drive production capacity and prices. The rise of Nasdaq has also led to significant valuation repricing for these companies, with the market viewing them as an extension of the NVIDIA ecosystem, and institutions willing to allocate continuously.<\/p>\n
This line is closely linked to the computing power main line: without sufficient high-performance memory, no matter how strong the GPU, its potential cannot be unleashed. The two form an 'engine + fuel' combination. Imagine GPU as the engine, memory as high-quality gasoline and oil, both are indispensable.<\/p>\n
This is the key variable most easily overlooked by retail investors, yet determines the length and ceiling of the AI cycle. The International Energy Agency (IEA) predicts that by 2030, global data center electricity consumption will approach 945 TWh, and US data center power demand may double around 2027.<\/p>\n
A high-density AI rack cabinet can easily reach hundreds of kilowatts per unit. This means the real bottleneck of AI is often not the theoretical computing power of chips, but the actual energy acquisition and heat dissipation capabilities. Companies specializing in liquid cooling technology, advanced power management systems, and high-efficiency transformers are therefore being repriced by institutions.<\/p>\n
This logic naturally extends to the client side: AI PCs are gradually becoming 'small local data centers,' with high-performance local inference placing higher demands on heat dissipation and power supply. Whoever can solve the energy consumption problem will gain order priority and higher bargaining power. In the next few years, order visibility for companies on this line is expected to improve significantly.<\/p>\n
Putting the three main lines together, you will see a clear picture of capital flow: GPU full-stack computing is the core engine, memory and storage are the high-speed blood, and power and heat dissipation ensure sustainable operation. The three together build the full-stack AI infrastructure. Capital is shifting from single-point products to full-chain ecological layout. The structural rise of Nasdaq is the external manifestation of this reallocation, allowing major capital to efficiently follow NVIDIA's 'new PC era' strategy.<\/p>\n
So, as ordinary investors, how should we respond to this change?<\/p>\n
First, prioritize leading companies with full-stack delivery capabilities. These companies not only have core technology but can also coordinate upstream and downstream, providing integrated solutions from data centers to AI PCs, with more stable revenue and stronger anti-cycle capability.<\/p>\n
Second, focus on companies that have already secured real large orders, have deep technical barriers, and mature supply chains. The AI cycle lasts for years, and not every company can reach profit realization. Real orders and delivery capabilities are key filtering indicators.<\/p>\n
Third, deeply realize that this is a structural long-term trend, not short-term hype. The Nasdaq rise reflects a systematic tilt of capital toward underlying infrastructure. Look long term and focus on key node assets in the full chain. Chasing short-term hotspots can easily get you harvested; long-term investment in infrastructure can capture the most stable profit pie.<\/p>\n
To give you more specific handles, let's extend with some dry data and segmented opportunities.<\/p>\n
In the full-stack computing field, besides NVIDIA's Vera Rubin iteration, the market is closely observing AMD's MI series and Intel's Gaudi as complements in vertical scenarios. The strong growth of AI servers in 2026 will drive upstream design and packaging testing demand. Companies with custom acceleration capabilities may gain additional growth in edge computing and hybrid deployment.<\/p>\n
In memory, the introduction of the HBM4 standard will be an important catalyst for 2026-2027. Currently, the HBM supply-demand gap is significant, supporting high gross margins for suppliers. Micron's market share increase reduces single-supplier dependency risk.<\/p>\n
In power and heat dissipation, liquid cooling penetration is expected to increase rapidly from 2026 to 2028. Equipment supplier orders have already extended beyond 2027. New power demand from US data centers will become a long-term growth engine for grid upgrades and supporting infrastructure companies.<\/p>\n
In addition, there are three hidden tracks that are relatively undervalued by the market, with high profit concentration and strong barriers, worth long-term tracking: high-end integrated liquid cooling solutions, high-speed interconnect technology for optical modules, and specific-domain high-efficiency power management chips. These tracks have high technical thresholds and strong customer stickiness, and are expected to contribute stable cash flow and valuation improvement beyond expectations in the next three years.<\/p>\n
Let's delve deeper into the specific progress and impact of AI PC penetration.<\/p>\n
According to multiple research institutions, AI PC's share of global PC shipments in 2026 is expected to reach or exceed 50%, with some optimistic forecasts even higher. This means client-side AI computing is moving from early adopter stage to the eve of mainstream adoption.<\/p>\n
New platforms like N1X are expected to accelerate the process. They bring stronger local AI performance, promote the maturity of the Windows on Arm ecosystem, reduce power consumption, improve battery life, while maintaining software compatibility. This is especially important for enterprise users: local processing of sensitive data better meets privacy compliance while reducing cloud costs.<\/p>\n
But penetration improvement won't happen overnight; it requires hardware-software synergy, including development tool optimization, application adaptation, and large model compression. There may be initial compatibility challenges, but after crossing the critical point, the growth curve steepens. This is like the transition from feature phones to smartphones: initial pain gives way to explosive growth. 2026 is likely a key inflection year.<\/p>\n
Now, let's objectively analyze the current cycle position and risk factors.<\/p>\n
From capital structure and fundamentals, the current situation is more like mid-cycle acceleration rather than a top. The main supports are threefold:<\/p>\n
First, major cloud vendors' AI capital spending continues to expand without obvious slowdown, and the industry is optimistic about long-term prospects.<\/p>\n
Second, key supply chain links remain tight or even in shortage, including high-end GPUs, HBM, and power equipment.<\/p>\n
Third, AI PC penetration has only just begun and is far from saturation, with huge room for future growth.<\/p>\n
Of course, risks must be soberly considered:<\/p>\n
First, if global interest rates tighten again due to inflation, capital spending may slow, and large projects may be delayed or scaled back.<\/p>\n
Second, persistent bottlenecks in raw materials or production capacity in the supply chain could lengthen project implementation time.<\/p>\n
Third, if the PC software ecosystem compatibility and developer support lag behind, AI PC adoption speed will be limited, and user experience improvement will be compromised.<\/p>\n
What is the next most critical inflection point? In my personal opinion, it is the tipping point when AI PCs truly go from 'early adopter' to 'mainstream adoption.' Once penetration stably exceeds 50%, and full-stack infrastructure production capacity gradually matches, capital reallocation will enter a deeper acceleration phase. At this point, the long-term profit realization capability of infrastructure companies will far exceed application-layer narratives.<\/p>\n
Finally, let's wrap up and elevate the entire logic.<\/p>\n
This round of AI cycle is not essentially about a few application companies becoming stronger, but a fundamental restructuring of the entire modern computing system: from large-scale cloud computing power, to high-speed memory support, to power and heat dissipation safeguarding, and finally extending to client-side AI entry popularity. This is a full-chain systematic transformation.<\/p>\n
The Nasdaq rise is just a surface reflection. The real importance is capital redefining the boundary between the 'infrastructure layer' and the 'application layer.' This definition will profoundly affect future profit flows. Companies controlling infrastructure are more likely to obtain stable long-term returns; those merely participating in concept narratives may face greater divergence.<\/p>\n
If your current AI story is still stuck at the application layer excitement, the real profit distribution may be hidden in deeper infrastructure structures. The true winners may not only be the well-known giants, but also a group of roles silently providing underlying support, currently undervalued.<\/p>
