FOCUS: THE FIVE MAJOR NASDAI DRAGONS

Summary of investments

My conclusion is straightforward: these five stocks are not the same “AI deal” but the five different nodes of the AI infrastructure chain; if the market continues to retreat because of inflation, interest rates, or bubble concerns, I will place them on a stratification watch list, rather than interpret them as “low buy-in” as a one-off chase. This report addresses U.S.A., MXL Max Linear, AMD Superpower, LITE Lumentum and VICR Vicor. Together, they benefit from capital expenditure for the AI data centre, but the sources of risk, the flexibility of performance and the manner in which valuations are absorbed vary. [1] [2] [3]

I believe that when AI entered this stage, what really mattered was not “AI has no story” but three questions: first, whether capital spending could continue to fall into real orders; secondly, whether business profits could prove valuation; and thirdly, whether the portfolio could withstand high volatility. McKinsey estimated that the global data centre could require about $6.7 trillion in capital expenditure by 2030 to meet its capacity needs, of which AI carries approximately $5.2 trillion in associated data centres; this suggests that AI infrastructure is a long investment cycle, but Fidelity also warns that profit growth, valuation, sustainability of capital expenditure and interest rate cycles will determine whether or not AI transactions will shift from a long-term theme to a short-term bubble. [1] [2]

One conclusion:I'M NOT SUREInfrastructure is still the direction I would like to take, but the point of purchase must be subject to silo discipline; at the same time, at high returns, high retreats and high volatility, there is a layering process。

FIRST, LOOK AT THE BIG PICTURE: AI INFRASTRUCTURE IS NOT THE STORY OF A GPU STOCK

The easiest mistake in the market is to simply equate AI with “buy GPU champions”. In my view, the real structure of the AI infrastructure is a capital expenditure chain: the front end requires calculus chips, the middle requires high bandwidth storage, network connectivity and light communications, and the back end requires power, heat, data centres and software movements. It is easy to follow the wrong rhythm when the valuation is extremely high by looking at a single link, and it is only when the chain is removed that it becomes clear whether each recall is a valuation, an order or a normal dishwashing of high Beta assets。

MacKenzie ' s measurement of the capital expenditure of the data centre gives this framework an important background. It does not mean that all companies will benefit simultaneously, nor that all AI-related stocks should rise, but rather that, if the demand for computing continues to grow, investment opportunities will spread along the “calculative-storage-connected-optic-power” line. [1] Morningstar ' s discussion of the AI equity framework also reminds me that the AIS stock selection should not be based on only conceptual heat, but on industry location, moats, valuation and uncertainty. [3]

MY JUDGMENT IS THAT THE OPPORTUNITY FOR AI INFRASTRUCTURE IS NOT “ONE LINE”, BUT “ONE NETWORK”. ONCE THE MARKET HAS RECEDED, IT IS NOT THE STANDARD THAT IS MOST WORTH STUDYING, BUT THE BASIC FACE OF WHICH NODE IS NOT PERJURED AND THE VALUATION IS DESTROYED BY RISK PREFERENCES。

OVER THE PAST YEAR, OPEN PRICE DATA SHOWED THAT ALL FIVE AI INFRASTRUCTURE TARGETS WERE SIGNIFICANTLY OUTRUNS FOR NASDAQ 100 AND SMH SEMICONDUCTOR ETF. LITE, MU, MXL, VICR AND AMD ALL REGISTERED HIGH INCREASES, WITH LITE AND MU PERFORMING THE MOST; BUT THE SAME SET OF DATA ALSO SHOWS THAT MOST OF THE FIVE STOCK RETREATS OVER THE PAST YEAR WERE BETWEEN ABOUT -28 AND -32 PER CENT, WHICH IS SIGNIFICANTLY HIGHER THAN NASDAQ ' S MAXIMUM RETREAT OF ABOUT -12.1 PER CENT. [9]

THIS SET OF DATA IS VERY CLEAR TO ME: STRONG TRENDS DO NOT AMOUNT TO LOW RISK, ELASTICITIES DO NOT AMOUNT TO READILY AVAILABLE. IF A TARGET INCREASES SEVERAL TIMES A YEAR, BUT IN THE PROCESS IT CAN BE REVERSED BY 30 PER CENT, THEN THE BUY-IN LOGIC CANNOT SIMPLY WRITE “A LONG-TERM LOOK AT AI”, BUT ALSO “HOW TO WITHSTAND FLUCTUATIONS”. IN OTHER WORDS, LOW BUY-IN IS NOT AN EMOTIONAL SLOGAN, BUT A FINANCIAL MANAGEMENT SYSTEM。

I'LL USE THIS TABLE AS THE STARTING POINT FOR POSITION MANAGEMENT. I WOULD BE WILLING TO OBSERVE IN BATCHES DURING RETREATS FOR MORE ROBUST BENCHMARKS SUCH AS MU AND AMD; FOR HIGH ELASTIC NODES SUCH AS MXL, LITE, VICR, I WOULD WRITE DOWN THE CEILING OF THE WAREHOUSE BEFORE CONSIDERING THE PRICE POSITION. THE REASON FOR THIS IS SIMPLE: VOLATILITY IS ITSELF A COST, NEGLECTING THE “LOW BUY-IN” OF COSTS AND, IN THE END, IT IS EASY TO TURN INTO PASSIVE CARRYING。

II. The difference between the five stocks: not who goes up much more and who buys who, but who's the more complete chain of evidence

I DON'T APPROVE OF PUTTING THESE FIVE COMPANIES IN THE SAME BASKET. THE CORE OF MU IS THE STORAGE CYCLE AND AI HBM REQUIREMENTS, THE CORE OF AMD IS THE DATA CENTRE COMPUTING PLATFORM, THE CORE OF LITE IS CLOUD-AI-RAY COMMUNICATION, THE CORE OF VICR IS HIGH-POWER SERVER POWER DELIVERY, AND MXL IS MORE IN FAVOUR OF AID DATA CENTRE CONTROL PLANE AND HIGH-SPEED CONNECTIVITY. THEY ALL BENEFIT FROM AI, BUT FINANCIAL FLEXIBILITY, CLIENT STRUCTURE AND VALUATION DIGESTION PATHS VARY。

According to the company's public information, Micron disclosed quarterly revenues of $11.315 billion in FY2025 Q4 press releases, $37.378 million in FY2025 revenue throughout the year and linked robust performance to the needs of AI data centres; AMD Q3 2025 press releases disclosed quarterly revenues of $9.246 million, compared with 36 per cent, and $4.3 billion, compared with 22 per cent; Lumentum FY2026 Q3 press releases disclosed revenues of $808.4 million, representing an increase of 90.1 per cent, and emphasized light technology related to AI, cloud computing and next-generation communications; Max Linear public press releases presented its Coronado and Laguna USB UART programmes aimed at the control horizontal interface of AI data centres; and Victor highlighted in the open information the demand for 48V modular power systems for the growth of AI, HPC and data centres. [4] [5] [6] [7] [8] [8] [8]

I'M NOT SORTED SIMPLY BY "SCALING UP." LITE AND MU ARE THE BRIGHTEST IF LOOKING ONLY AT THE INCREASE OF THE PAST YEAR; MU AND AMD ARE MORE EASILY TRACKED BY INSTITUTIONAL FUNDS IF LOOKING AT THE BASIC EVIDENCE CHAIN; AND MXL, LITE, VICR PROVIDE A STEEPER YIELD CURVE IF LOOKING AT THE HIGH ELASTIC SATELLITE WAREHOUSE, BUT AT THE SAME TIME REQUIRE STRICTER CUT-OFF AND POSITION CAPS。

III. ROLE OF RISK: The top right corner is not paradise, but discipline

Many investors prefer to see high yield maps, but do not like to look back. The opposite is true of my view: the rate of return for the high-level Beta target is only the result, and the maximum withdrawal is the provision that must be accepted before entering. Figure 3 shows the previous year's rate of return and the largest withdrawal in the same picture, and it can be seen that five equities are located in high-yield areas, but the vertical retreat is also deep. That means..

They are not low-volatility growth units, but rather high-flexible assets that need to be absorbed with silo discipline. [9]

I'LL DEAL WITH THIS KIND OF STOCK AT THREE LEVELS. THE FIRST TIER IS “CORE TRACEABILITY”, I.E. MORE COMPLETE BASIC EVIDENCE AND BETTER INSTITUTIONAL COVERAGE, SUCH AS MU AND AMD. THE SECOND LAYER IS “HIGH ELASTIC SATELLITES”, I.E., THE TARGET FOR INDUSTRIAL LOGIC WITH HIGH VOLATILITY, SUCH AS LITE AND VICR. THE THIRD LAYER IS “OBSERVATIVE ELASTICITY”, I.E., THE TARGET WHERE PRODUCT ORIENTATION IS IMAGINATIVE BUT WHERE FINANCIAL PERFORMANCE REQUIRES MORE QUARTERLY VALIDATION, SUCH AS MXL。

THEREFORE, MY DEFINITION OF “LOW BUY-IN” IS NOT A DROP IN PRICES, BUT A GRADUAL ABSORPTION OF FLUCTUATIONS IN ACCORDANCE WITH THE PRE-ESTABLISHED SILO RULE WHEN PRICES ARE REVERSED, FUNDAMENTALS ARE NOT DETERIORATING AND THE CAPITAL EXPENDITURE CHAIN IS STILL ON TRACK. IN PARTICULAR, HIGH-VOLATILE MARKERS SUCH AS MXL, LITE, VICR, THE SIZE OF THE WAREHOUSE IS MORE IMPORTANT THAN THE PURCHASE PRICE。

Industrial chain rating: five shares are not the same transaction, but the same nodes

IN ORDER TO AVOID MIXING ALL OF AI STOCKS INTO ONE CONCEPT, I GRADED FIVE EQUITIES IN FIVE DIMENSIONS: DIRECTNESS OF CALCULATION, SENSITIVITY OF AI CAPITAL EXPENDITURE, CYCLICAL FLUCTUATIONS, VALUATION REALIZATION PRESSURE AND PORTFOLIO FRAGMENTATION. THIS RATING IS NOT A YIELD FORECAST, NOR AN INVESTMENT RATING, BUT RATHER HELPS ME TO DETERMINE WHAT ROLE EACH STOCK IS IN IF AN AI INFRASTRUCTURE OBSERVATION BASKET IS TO PLAY。

THIS CHART HAS INSPIRED ME TO THINK THAT MU AND AMD ARE MORE LIKE CORE EVIDENCE ASSETS ON THE AI INFRASTRUCTURE MAIN LINE; THAT LITE AND VICR ARE MORE LIKE HIGH ELASTIC NODES IN THE CHAIN THAT CAN BE MAGNIFIED BY MONEY; AND THAT MXL IS MORE ORIENTED TOWARDS THE OBSERVATIONAL MODEL THAT “THE PRODUCT MAY BE REVALUED AFTER IMPORT”. FIVE EQUITIES HAVE RESEARCH VALUE, BUT THE LOGIC OF BUYING MUST NOT BE THE SAME。

MY IDEA OF CONFIGURATION IS TO GIVE PRIORITY TO MORE COMPLETE MUS AND AMDS IF I WANT ONLY THE CORE OF AI; TO USE LITE, VICR AS SATELLITE OBSERVATION IF I AM WILLING TO TAKE ON HIGHER FLUCTUATIONS; AND TO RECOGNIZE THE UNCERTAINTY OF ITS SMALL DISK PROPERTIES AND REVENUE REALIZATION, THE POSITION IS MORE RESTRAINED THAN A FEW OTHERS。

The operational framework: the real point of purchase is the simultaneous occurrence of the three things: "Roving back, confirming, splitting up."

I DON'T TAKE ANY RETREAT AS A BUY-IN BECAUSE I'M STRONG. THERE ARE AT LEAST THREE CONDITIONS TO BE MET FOR A TRULY WORTHWHILE RETREAT: FIRST, PRICES HAVE RELEASED SHORT-TERM EMOTIONS; SECONDLY, THE BASICS OF THE ENTERPRISE HAVE NOT DETERIORATED SIMULTANEOUSLY; AND THIRDLY, THERE ARE CASH AND RISK BUDGETS IN THE PORTFOLIO. ANY MISSING ITEM WOULD TURN INTO A EMOTIONAL TRANSACTION。

Fidelity's framework for AI foam risk is worth putting here. It reminds us that while the theme of AI may still be a multi-year cycle, investors must track profit growth, the quality of profits, valuation, sustainability of capital expenditure and interest rate cycles. I totally agree with that. Rather than being unbuyable, AI cannot mask short-term risks with “long-termism” at the time of the most expensive, hottest and full-time valuation。

IN SUMMARY, I WILL PLACE THE FIVE SHARES IN THE AI INFRASTRUCTURE OBSERVATION POOL, BUT WILL NOT CONSIDER THEM ALL AS AN EQUAL-WEIGHT BUY-IN LIST. FOR ME, THE CORRECT ORDER IS TO DEFINE THE ROLE, THEN THE SLOT, AND FINALLY THE PRICE。

CONCLUSION: Buyable when it's low, but ask if you can handle fluctuations

The final conclusion goes back to the title: when the five major NASDAI shares are bought low, they can be studied, but they cannot be lazy. If capital expenditure for the AI data centre continues to expand, the storage, computing, photocommunication, power supply and connection of the MU, AMD, LITE, VICR and MXL will all continue to benefit from the base; but if interest rates resume, cloud capital expenditure slows, AI orders fall short of expectations, or valuations are overstretched in the coming seasons, these high Beta assets will be quickly withdrawn。

MY STRATEGY IS CLEAR: THE CORE WAREHOUSE GIVES PRIORITY TO ASSETS WITH A STRONGER BASE CHAIN OF EVIDENCE, THE SATELLITE WAREHOUSE GIVES ELASTIC BUT VOLATILE NODES, AND THE OBSERVATION WAREHOUSE GIVES SMALL AND MEDIUM DISKS THAT STILL NEED TO BE VALIDATED. PURCHASES MUST BE MADE IN BATCHES, WAREHOUSE SLOTS MUST BE LIMITED AND RISKS MUST BE WRITTEN IN ADVANCE ON PAPER. A TRULY MATURE AI INVESTMENT IS NOT EXCITED TO SEE A RECALL, BUT TO KNOW WHICH ONE CAN BUY, HOW MANY, AND WHAT TO DO WRONG。

ONE SENTENCE SUMMARIZES THE LONG-TERM LOGIC OF THE AI INFRASTRUCTURE, BUT LOW BUY-IN IS NOT AN ASSAULT SIGN, BUT A DISCIPLINE SHEET; FIVE STOCKS ARE TO BE SPLIT INTO FIVE NODES, FOLLOWED BY WAREHOUSING AND TIME TO ABSORB FLUCTUATIONS。

Risk tip

THE PRESENT REPORT IS FOR DISCUSSION ONLY AND DOES NOT CONSTITUTE ANY PROMISE OF PROCEEDS OR A PROPOSAL FOR A BUY-IN. AI INFRASTRUCTURE-RELATED COMPANIES ARE GENERALLY HIGHLY VOLATILE, VALUATION-SENSITIVE AND PROCYCLICAL, AND INVESTORS NEED TO BE INDEPENDENTLY JUDGED ON THEIR OWN RISK TOLERANCE. THERE ARE FIVE CATEGORIES OF RISKS THAT FOLLOW MOST THAT NEED TO BE TRACKED: FIRST, CLOUD MANUFACTURERS' CAPITAL SPENDING MAY BE RE-PRICING THE AI HARDWARE CHAIN ORDER IF IT FALLS SHORT OF EXPECTATIONS; SECOND, IF INTEREST RATES GO UP AGAIN, THE HIGH-VALUE GROWTH UNIT WILL FACE DISCOUNT RATE PRESSURE; THIRD, THE BREAKDOWN OF STORAGE, PHOTOCOMMUNICATION, POWER SUPPLY AND CONNECTIVITY HAS INVENTORY CYCLE AND CUSTOMER CONCENTRATION RISKS; FOURTH, HIGH-FLEXING MARKERS MAY HAVE LIQUIDITY AND VALUATION FLUCTUATIONS; AND FIFTH, THE AI THEME MAY SHIFT FROM “LONG-TERM SPACE PRICING” TO “CURRENT CASH-FLOW PRICING” IF THERE IS INSUFFICIENT PROFITABILITY。

THE REPORT HAS BEEN PREPARED BY THE INVITED ANALYSTS. THE VIEWS EXPRESSED IN THE REPORT REPRESENT ONLY THE INDIVIDUAL POSITION OF THE AUTHOR AND NOT THAT OF THE BIT PLATFORM. THIS MATERIAL IS FOR REFERENCE PURPOSES ONLY AND DOES NOT CONSTITUTE AN INVESTMENT RECOMMENDATION。