AI PC WAR: NO CAMP, NO TOLL STATION

Roger Lee |BIT, American Stock Exchange Analyst
TWENTY-ONE YEARS OF EXPERIENCE WITH INVESTMENT, MANAGEMENT AND FINANCIAL INSTITUTIONS, WITH A LONG-TERM FOCUS ON THE AI INDUSTRIAL CHAIN, MACRO-MARINESS AND OPTIONS STRATEGY。

 

YIN WEIDA AND UDEC WENT INTO AI PC, APPARENTLY CONSUMING A NEW SET OF CHIPS FOR THE PCWindows end-side AI Ecology from single-point testing to multiplayer competitionI don't know. My judgment is that this war should not be reduced to a “x86 against Arm” religious station; what really deserves to be studied is who can cross the switch cycle and continue to access Maori, cash flow and chain pricing。

I'LL SEE AI PC AS A THREE-TIER OPPORTUNITY:

• First floorAdvanced processing toll stationsWHOEVER WINS, IT'S EASIER FOR TSMC TO PAY FOR THE ROAD
• Second floorCalculator and Platform SpillAMD and NVDA, respectively, represent x86 attacks and GPU infrastructure extension
• Third floorStructure proliferation and dilemma reversalBOTH ARM AND INTC ARE FLEXIBLE, BUT WE HAVE TO BE MORE DISCIPLINED

I. The industry ' s judgement:AI PCFrom concept to delivery certification period

Gartner, which in 2024 was expected to produce 1142.2 million AI PCs in 2025, accounting for 43 per cent of the PC market; after being updated in 2025 and affected by tariff and procurement tempo disturbances, was projected to repair to 77.792 million, or 31 per cent, but was still expected to reach 143.13 million in 2026, or 54.7 per cent penetration. This set of data has inspired me not to "AI PC needs perjury" but..The short-term rhythm will fluctuate and long-term calibration will remain unchangedI don't know。

FROM THE INVESTMENT POINT OF VIEW, THE REAL DIFFICULTY FOR AI PC IS NOT "NO NPU," BUT..WHETHER USERS ARE WILLING TO SWITCH FOR LOCAL AI EXPERIENCEI DON'T KNOW. IF THE APPLICATION LEVEL IS CONFINED TO MINUTES OF MEETINGS, PHOTO GENERATION AND SIMPLE ASSISTANTS, THE FLEXIBILITY OF THE SWITCH WILL BE LOWER THAN THE MARKET’S MOST OPTIMISTIC EXPECTATIONS; BUT IF THE ENTERPRISE BEGINS TO USE PRIVACY CALCULATIONS, LOW-DELAYED REASONING AND LOCAL KNOWLEDGE BASE DEPLOYMENT AS STANDARD CONFIGURATIONS, AI PC WILL MOVE FROM CONSUMER ELECTRONIC STORIES TO BUSINESS IT UPDATES。

II. COMPETITION PLACE: CHECK AT THE CHECK AND CHECKING Fees

The surface narrative of AI PC is Arm's challenge x86, but I'm more concerned about where the profit pools move. NVIDIA is strong in GPU and AI, AMD is strong in x86 CPU and GPU, high-intensity in low-capacity and communications, and Intel is strong in stock ecology and business channels. They each have advantages, but common ground is clear:High-end chips can't get past advanced programmingI don't know。

TrendForce disclosed that in the second quarter of 2025, global round-generation earnings were approximately $41.7 billion, with a TSMC share of 70.2 per cent; in the fourth quarter of 2025, global round-generation earnings were about $46.3 billion, with a TSMC share of about 70.4 per cent. This means that as long as AI PC, AI servers, cell phones AP and marginal AI chips continue to compete for advanced programming, TSMC is not simply a periodic unit, but more like the toll entrance of the entire AI hardware age。

I don't think every new product release is worth pursuing, but I think every time there is increased competition in the industrial chain, one question should be asked:If the winner is not sure, who can charge all the winnersIN THE AI PC LINE, MY ANSWER IS STILL..ADVANCED PROGRAMMING, CONTAINMENT, CRITICAL IP AND PLATFORM SOFTWAREinstead of simply placing a structured slogan。

III. Sorting of the target: look at the bottomTSMLet's goAMDLet's seeIntel/ARM

About a year and a half of the calibrations have been traded in advance for AI PC, end-end AI and algorithm spills. The Yahoo Finance daily prices show greater elasticity in the sample ranges AMD, Intel, ARM and TSM, but they represent different risk returns. My approach is not to buy all AI PC-related objects together, but to rank them according to certainty, valuation discipline and industry chain location。

 

My core conclusion is simple: it is not a war that buys only winners, but a war that buys a toll station, a platform, a definite cash flow. If the market fills up the mood on the day of the press release, I prefer to wait; if I return to pull back the company’s risk gains back to a reasonable range, I will give priority to TSM and AMD, and then to ARM and Intel。

Risk tips

The risks of this main line cannot be ignored:

FIRST, AI PC APPLICATIONS MAY FALL SHORT OF EXPECTATIONS, RESULTING IN AN UNIMAGINABLY STRONG REPLACEMENT CYCLE。

Secondly, the narratives of high-profile and new entrants will be suppressed if the improved compatibility of Windows on Arm is too slow。

THIRDLY, TARIFFS, SUSPENSION OF ENTERPRISE PROCUREMENT AND MACRO-UNCERTAINTY COULD AFFECT PC DEMAND。

FOURTHLY, TSMC MAY ALSO HAVE A VALUATION RETREAT IF THERE IS A PHASED MISMATCH BETWEEN ADVANCED SUPPLY AND DEMAND。

FIFTH, THE ENTIRE AI CHAIN IS OVERVALUED, AND ONCE THE US EQUITY RISK PREFERENCE DECREASES, THE MOST RESILIENT TARGET TENDS TO BE THE FASTEST。

THEREFORE, I PREFER TO CONSIDER AI PC ASLong-term industrial migrationNot short-term news deals. The real professional method of buying is not to buy slogans on the day of the conference, but to buy ecology, to buy a toll station, and to buy a company that can cash the cash flow on a sustainable basis。

Description of data sources