AMD: A Deep Dive Through the Lens of a Long-Term Business Owner

On cost structure, AMD runs a research-heavy, manufacturing-light fabless model. It outsources wafer fabrication primarily to TSMC, uses GlobalFoundries for certain older-node HPC products, and uses TSMC, UMC, and Samsung for certain programmable devices; packaging and testing rely on external ATMP partners and a joint venture affiliated with Tongfu Microelectronics. In 2025, AMD's purchases from its ATMP joint venture totaled about $2 billion. As a result, AMD's fixed-asset base is relatively light, but its dependence on advanced process nodes, advanced packaging, supply-chain support, and capacity coordination is very high.

This business does not depend on a single customer to a dangerous degree. In 2025 and 2024, no customer accounted for more than 10% of company revenue, but the company also explicitly acknowledges that it will continue to depend on a small number of large customers for a meaningful share of revenue and receivables; and in 2023, one Client and Gaming customer accounted for 18% of total company revenue. This means AMD's explicit concentration has declined, but its implicit reliance on a few large customers has not disappeared.

If "can I understand it" means "can I explain this business to a non-expert," the answer is I can explain about 70%: it is a company that earns money by designing computing chips and platforms that are stronger, more power-efficient, and better suited to AI/cloud/edge workloads. But if "can I truly understand it deeply" means "can I confidently forecast process nodes, packaging, software stacks, product cadence, customer procurement roadmaps, and technology substitution over the long run," then it is clearly far more complex than consumer goods, insurance, railroads, or exchanges. If the stock market closed for 5 years, I would be willing to own AMD as a business—but only at a reasonable or even somewhat low price, not chasing it at the current level.

Business understandability score: 3/5.

Industry and Moat

The semiconductor industry overall remains in a state of long-term growth and short-term strong cyclicality. WSTS projects that the global semiconductor market will grow more than 25% in 2026 to nearly $975 billion; TSMC, in its 2025 annual report, likewise said the Foundry 2.0 industry grew 16% in 2025 and expects AI-related demand to stay strong in 2026. At the same time, IDC projects AI infrastructure spending of $487 billion in 2026, up about 53% year over year. The PC side is more divided: in April 2026, IDC noted on one hand that Q1 PC shipments grew 2.5% year over year to 65.6 million units, while on the other hand sharply cutting its full-year PC shipment forecast because of memory shortages, supply-chain problems, and rising prices. In other words, AMD sits in an industry puzzle where "AI is hot, PCs are cool, and Embedded is in a mixed recovery."

On the competitive landscape, AMD faces different rivals in different sub-markets. Per company disclosures, in Data Center its main rivals are Intel, NVIDIA, and Altera, along with some fabless makers of dedicated accelerators and Arm CPU vendors; in client CPUs the main rival is Intel, and in discrete graphics the main rival is NVIDIA; in Embedded and FPGAs it faces Altera, Lattice, and Microsemi as well as ASIC/ASSP vendors such as Broadcom, Marvell, ADI, TI, NXP, Qualcomm, and NVIDIA. More troubling, some of the largest customers are developing their own CPUs or accelerators, which directly compresses the serviceable market.

AMD's current position in the industry can be summarized as: in x86 server CPUs, it has gone from a fringe player to a strong challenger; in AI GPU/accelerator platforms, it is a strong number two and a catch-up player; in Embedded and FPGAs, it holds a high-quality but non-monopolistic combination of advantages. According to Mercury Research, in Q1 2026 AMD had captured roughly 33.2% of server x86 shipment share and 46.2% of server x86 revenue share, indicating that EPYC's competitiveness in high-end SKUs and ASPs has indeed improved markedly.

But note: winning in server CPUs is not the same as winning in AI platforms. NVIDIA's moat extends well beyond the chips into a platform-level capability that spans CUDA, networking, systems, the software stack, and the developer ecosystem. AMD is indeed advancing quickly in AI, but the company itself admits competition is extraordinarily fierce, technology changes fast, and standards and customer preferences keep evolving. That is why I am unwilling to simply classify AMD as a "durable, Coca-Cola-like business"; it looks more like "an excellent technology company in a good industry," not a "naturally stable machine for capital returns."

On the specific types of moat, here is AMD's situation. Brand advantage exists but is moderate in strength: Ryzen, EPYC, Instinct, and Xilinx/Versal are already strong brands among professional and technical customers. Cost advantage is weak: it does not have the structural low-cost advantage of a TSMC or certain software companies. Scale advantage is moderate-to-strong: it has R&D scale spanning CPU/GPU/FPGA/DPU/AI NIC, but it is still smaller than NVIDIA's combined scale in AI platforms. Network effects are weak-to-moderate: far weaker than CUDA on the AI software ecosystem, but stronger in embedded toolchains and customer relationships. Switching costs vary widely by business: embedded, FPGA, and server-platform validation carry real switching costs, while consumer PCs are far lower. Channel advantage is moderate: its relationships with OEMs, ODMs, cloud, and distribution are strengthening but not irreplaceable. Patent / licensing barriers exist—as of year-end 2025 the company held roughly 12,600 granted patents and 6,300 pending patents worldwide, along with extensive cross-licenses—but the company also explicitly states that it does not consider any single patent to be material to the business. Data advantage is relatively weak. Corporate culture and operating capability are strong, especially in product cadence and cross-category integration.

My overall judgment on the moat is: at the company level the moat is "stable, with a slight widening," but the moat around the core AI platform is not as deep as market sentiment is pricing in. For a new entrant to replicate AMD's full-stack design capability would likely take many years and a cumulative R&D investment in the billions of dollars or more; but for the giants and key customers that already exist, AMD's moat is not deep enough to bring competition to a halt.

In an inflationary environment, AMD has some pricing power, but it is not stable. For example, 2025 client revenue growth came partly from higher average processor selling prices, and the higher gross margin in Q1 2026 was also tied to a rising share of Data Center revenue; this shows that when product strength leads, supply is tight, and the mix improves, AMD can raise ASPs. The problem is that this looks more like a "performance- and supply/demand-driven mix/ASP improvement" than an unconditional, durable pricing power.

In a downturn, AMD has the ability to survive and may not lose money too badly; but it is by no means a company whose profits stay unmoved through a recession. In 2023, total company revenue fell from $23.601 billion to $22.680 billion, operating income dropped from $1.264 billion to $401 million, and the Client segment even swung to an operating loss; the company overall stayed profitable, however, and maintained its cash flow. This shows it has resilience, but not the "counter-cyclical stability" of a consumer monopoly.

Industry attractiveness score: 4/5. Moat strength score: 3/5.

Management and Capital Allocation

Looking only at operational execution, AMD under Lisa Su is one of the best turnaround and product-revival cases in U.S. tech hardware over the past decade. You can see this in the change in revenue mix: 2023 Data Center revenue was $6.496 billion, and by 2025 it had risen to $16.635 billion; total company revenue grew to $34.639 billion in 2025, with gross margin improving to 50%. This was not engineered through financial tricks but achieved through product roadmaps, customer penetration, and a mix upgrade.

On incentive design, AMD's governance is generally well-structured: executives face stock-ownership requirements, long-term equity incentives are a multi-year mix of time-vesting and performance-vesting awards, there is at least a one-year vesting period, and a clawback mechanism exists. In the 2025 annual bonus, about 70% was based on company financial performance and 30% on strategic milestones, with financial metrics including adjusted non-GAAP net income and adjusted free cash flow.

But there are places here where I want to reserve judgment. First, the alignment between management and ordinary shareholders comes mainly from equity incentives, not from founder-style large ownership stakes; this means they are motivated to lift market value but not necessarily to "buy back only when undervalued and answer only to per-share intrinsic value." Second, AMD's bonus metrics use a fair amount of adjusted non-GAAP figures that explicitly exclude items such as acquisition-related intangible amortization and stock-based compensation. For operating management, that is not necessarily unreasonable; but for long-term shareholders, these items cannot be treated as entirely "costless."

On capital allocation, over the past few years AMD has mainly used cash for R&D, M&A, modest buybacks, and maintaining a strong balance sheet. The Xilinx deal clearly changed the company's business quality, moving it into the deeper-moat territory of FPGA/adaptive SoC/embedded; but it also brought a very large burden of goodwill and intangibles. By Q1 2026, AMD carried $25.344 billion in goodwill and $16.154 billion in net acquisition-related intangibles. This shows that M&A is not "without cost."

In 2025, management continued to optimize the asset portfolio. After the ZT Systems-related transaction, the company retained the ZT Design Business, which is more critical to its data-center design capability, while the manufacturing business was divested, completed on October 27, 2025—indicating that management prefers to keep high-value-added design and systems capabilities rather than hold low-return manufacturing assets long term. I consider this move a rational, capital-allocation positive.

What genuinely warrants caution is the quality of buybacks. AMD's annual report discloses that, as of year-end 2025, its total buyback authorization was $14 billion, of which $9.4 billion remained unused; in 2025 the company repurchased about $1.3 billion of stock under the program, plus about $607 million for buybacks tied to tax withholding on equity awards. The issue is that shares outstanding at year-end did not shrink; they actually rose from about 1.622 billion at year-end 2024 to about 1.630 billion at year-end 2025. This shows that AMD's buybacks, at least so far, are more about offsetting equity-compensation dilution than about substantively shrinking the share count through large buybacks when undervalued, as a truly excellent value-oriented capital allocator would do.

So my conclusion on management is: operationally credible, strategically forward-looking, and reasonably candid in disclosure; but its capital allocation still looks more like that of an "excellent tech-company management team" than that of a "Buffett-style capital allocator whose sole North Star is per-share intrinsic value." This is not a knock on Lisa Su, but a distinction between an "extremely strong CEO" and an "extremely strong capital allocator."

Management and capital-allocation score: 3.5/5.

Financial Quality

First, the key financial table. The table below is compiled mainly from AMD's official 10-K/10-Q filings: 2021–2023 data comes mainly from the 2022/2023 annual reports, 2024–2025 data from the 2025 annual report, and Q1 2026 data from the Q1 2026 10-Q. Some ratios are approximations I calculated on a consistent basis, as noted in the table.

• The 2025 annual-report basis includes the ZT Manufacturing Business classified as discontinued operations; on a continuing-operations basis only, 2025 operating cash flow was about $6.493 billion. ** On a continuing-operations basis.

This table reveals several key facts. First, AMD's revenue growth has reaccelerated. 2023 was a trough year, 2024 recovered, and 2025 jumped noticeably, driven mainly by Data Center and Client/Gaming; Q1 2026 again posted high revenue, with Data Center the most important growth engine.

Second, the margin improvement is real, but it is not an unconditional, linear climb. In 2023 the company's gross margin was 46% and operating income only $401 million; by 2025, gross margin was 50% and operating income $3.694 billion; in Q1 2026 gross margin rose further to 53%. This shows that the mix tilting toward Data Center, the recovery in Client, and continued profit contribution from Embedded are indeed improving earnings quality.

Third, cash flow looks prettier than profit on the surface, but you must separate "continuing operations" from "discontinued operations," and guard against being dazzled by large amortization. The 2025 annual-report basis shows operating cash flow of $7.709 billion, far above net income of $4.335 billion; but within that, continuing-operations operating cash flow was about $6.493 billion, with discontinued operations contributing about $1.216 billion. Even looking only at continuing operations, cash conversion is decent, but not as dramatic as the headline number looks.

Fourth, AMD's profit is broadly real cash profit, but it is not a pure, "maintenance-free" cash machine. 2025 continuing-operations operating cash flow of $6.493 billion does indicate that profit has cash support; but working capital consumed a lot in 2025, especially an inventory increase of about $2.2 billion. Management explicitly explained this was to support the ramp of advanced-node Data Center products. For bulls, this is strong demand; for conservative investors, it also means growth must first tie up cash.

Fifth, capital expenditure is low, but growth is not entirely "capital-light." AMD's direct capex is far below that of an in-house-fab company like Intel—a natural advantage of the fabless model—but it "indirectly capitalizes" through prepayments, inventory, supply-chain commitments, packaging capacity, the software stack, and M&A. So it is correct to call AMD an "asset-light company," but wrong to call it a software business that "can grow almost infinitely with virtually no capital."

Sixth, the balance sheet is a clear strength for AMD. At year-end 2025, cash, cash equivalents, and short-term investments were about $10.6 billion, and total debt principal was about $3.3 billion; by the end of Q1 2026, cash and short-term investments together were about $12.347 billion. This means AMD is in a clear net-cash position. At the same time, 2025 interest expense was about $131 million, very low relative to operating income of $3.694 billion, so interest coverage is very high. Even in an industry downturn, survival risk is small.

Seventh, I see no clear evidence of aggressive accounting or financial fraud; but accounting complexity is indeed high. The complexity comes mainly from the large intangible amortization brought by Xilinx and subsequent acquisitions, the discontinued-operations treatment of ZT, and the inventory write-downs and reversals caused by AI export controls. In 2025, MI308 incurred about $800 million of inventory-related charges in Q2 due to U.S. export restrictions, with about $360 million reversed in Q4, for a net impact of about $440 million. This looks more like an external regulatory shock than profit manipulation; but it reminds us that AMD's GAAP profit can be materially affected by one-off regulatory and acquisition-amortization swings.

To close on the "Buffett-style question": AMD is not a company that needs more and more cash as it grows, but it is also not a company that needs less and less cash as it grows. It can both grow and generate cash flow; it is just that the certainty, distributability, and sustainability of that cash flow are not yet strong enough to support the current 100x-level free-cash-flow valuation.

Owner Earnings and Intrinsic Value

First, my conservative estimate of Owner Earnings. Here I do not use the loose method of "adding back stock-based compensation in full," because that would overstate the cash truly available to ordinary shareholders; nor do I treat all capital expenditure as maintenance capex, because part of AMD's spending clearly serves growth. My framework is:

• Starting point: continuing-operations net income, using 2025 continuing-operations net income of about $4.269 billion;

• Add back: depreciation and amortization, plus acquisition-related intangible amortization, about $750 million and $2.254 billion respectively in 2025;

• Do not add back: stock-based compensation, because although non-cash, it is a real cost to shareholders;

• Subtract: maintenance capital expenditure, which I conservatively estimate at about $350 million to $450 million;

• Subtract: normalized working-capital needs, which, given the significant cash tied up by inventory and receivables in 2025, I deduct over a range of $1.5 billion to $1.9 billion.

Along these lines, AMD's conservative Owner Earnings for 2025 are roughly $5.0 billion to $5.5 billion, and I take the midpoint of $5.2 billion as the valuation starting point. This figure is close to 2025 continuing-operations free cash flow of about $5.5 billion, so it is not pulled out of thin air but a deliberately conservative approximation of "distributable cash-generating capacity."

Tying this figure to the current price makes things clear: at the current market cap of about $683.2 billion, AMD's stock price is equivalent to roughly 124x to 137x the Owner Earnings I estimate; even on a reported 2025 FCF basis, it is still above 100x. Such a valuation is very aggressive for a semiconductor company that—however high-quality—is still clearly exposed to technological competition and industry cycles.

Discounted Owner Earnings

I use a 10-year DCF with a uniform 10% discount rate—not because AMD's risk is extreme, but because the technological substitution, competition, and valuation volatility in semiconductors are all higher than in a typical consumer-monopoly industry. The three scenarios are as follows:

These figures are not "precise answers"; they tell you that even with quite favorable growth assumptions, AMD's current price is already far beyond what a DCF can explain. If I push the assumptions to something very aggressive—say more than 20% growth in years one through five, still double-digit growth in years six through ten, and terminal growth near or above 4%—the valuation can rise noticeably; but that is no longer "conservative value investing," it is a growth bet that "AMD will become a quasi-platform king in AI infrastructure over the long run." That bet may turn out right, but it does not fit a "margin-of-safety-first" method.

Relative Valuation

Even placed inside the "expensive AI/semiconductor basket," AMD's current valuation is not cheap. Based on current market quotes and each company's most recent financial disclosures, AMD's current P/E is about 135.8x, NVIDIA's about 54.1x, Broadcom's about 102.3x, and Intel's is negative; on a rough basis using current market cap and most recently disclosed free cash flow, AMD's P/FCF is roughly 100x to 124x, NVIDIA's about 52.6x, Broadcom's about 74.3x, and Intel's free cash flow is negative; on a rough basis using current market cap and book equity, AMD's P/B is about 10.8x, NVIDIA's about 34.3x, and Intel's about 5.7x. You cannot simply say "because NVIDIA is more or less expensive, AMD is reasonable"; what truly matters is that AMD's current price demands cash flow and returns that are already very close to, or even above, some stronger platform-type peers.

Likewise, roughly estimating AMD's enterprise value using 2025 data, current market cap, Q1 2026 cash and short-term investments of about $12.347 billion, and year-end 2025 total debt of about $3.3 billion, its EV/Revenue is about 19.5x, and using 2025 operating income plus depreciation/amortization and intangible amortization as an EBITDA proxy, EV/EBITDA is roughly 100x. This is not a "cheap growth stock," nor even a "reasonably priced growth stock"; it looks more like an asset for which "the market is willing to pay a very high premium for a long-dated AI option."

Asset Value and Liquidation Value

An asset-value approach cannot support AMD's current stock price. At the end of Q1 2026, AMD had about $12.347 billion in cash and short-term investments, $8.045 billion in inventory, and $6.035 billion in receivables, but it also had $25.344 billion in goodwill and $16.154 billion in acquisition-related intangibles. In other words, a large part of book assets comes from intangibles formed by past acquisitions. For a platform company that keeps creating high returns in operations, that is not necessarily bad; but viewed from a "liquidation value" angle, it is clearly far below the current market cap of more than $680 billion. So the asset approach is more of a "downside survival floor" for AMD than a "valuation support."

Price-Range Judgment

Combining DCF, relative valuation, and the asset approach, I offer the following ranges:

At the current price of around $414, AMD trades at a significant premium to every one of my conservative, reasonable, and optimistic intrinsic-value ranges; even valued at the top of the most optimistic range, $190, the current price is still roughly more than double that. This is why I define it as "Watch," not "Buy."

Margin-of-safety conclusion: none, and not even close.

Risks, the Bear Case, and Comparisons

The most important risk, first, is competitive risk. In server CPUs, AMD is indeed taking share; but in AI accelerator platforms, it faces NVIDIA, which is stronger across chips, networking, systems, software, and developer ecosystem. NVIDIA's fiscal-2026 revenue was $215.938 billion, net income $120.067 billion, operating cash flow $102.718 billion, and Data Center revenue $193.737 billion; this shows AMD is chasing a platform-type rival whose size, profit, and cash flow are all extremely strong.

Second is technological substitution and customer in-house-design risk. AMD itself explicitly writes in its annual report that some customers are internally developing their own data-center microprocessors or accelerator products; this is a threat to AMD's TAM itself, not merely a fringe variable affecting the competitive landscape. For a high-valuation company, this kind of risk is especially lethal, because it hits both the growth expectation and the valuation multiple at the same time.

Third is regulatory and export-control risk. AMD has already incurred large inventory-related charges from MI308 export restrictions to China in 2025; the Q1 2026 10-Q further discloses that although MI325 obtained some licenses, uncertainty remains over whether it will be allowed into China and over the related inspection and tariff arrangements. If the U.S. further expands restrictions in the future, the impact may extend beyond China revenue to the product roadmap and inventory management.

Fourth is supply-chain and manufacturing-dependence risk. AMD's advanced products depend heavily on TSMC, advanced packaging, and ATMP partners, and TSMC itself stresses in its annual report the importance of AI demand and capacity expansion in advanced process nodes and advanced packaging. For AMD, this means "industry strength" and "supply-chain scarcity" often coexist: when demand rises, it becomes even more dependent on whether it can secure key capacity. The company also explicitly acknowledges that an inability of the ATMP JV to meet demand could cost it orders.

Fifth is overvaluation risk. This is the risk I consider most real today and most likely to cause permanent capital loss. With earnings per share of about $3.05, even if results keep growing over the next few years, simply having the multiple the market is willing to pay fall from 135x toward the 20x to 35x range more typical of a mature, high-quality hardware company could still produce a very large drawdown. In the crudest terms, if EPS fails to expand quickly over the long run while the multiple falls to 30x, the stock would correspond to roughly $90; this implies enormous downside room from the current price. I am not predicting it will definitely fall to there—I am illustrating that a high-quality company at a high valuation can still deliver poor long-term returns.

The strongest bear case is actually not complicated: AMD may be a very good company, but the market treats it as "the next platform overlord," whereas it is more likely just a very excellent second-tier core supplier in the AI era. If this judgment holds, then even if its operations keep succeeding, that may not be enough to match the current price. What the bears see is not necessarily "the company will get worse," but "the company is not bad enough to justify being this expensive."

What facts would make me admit I was wrong, or force a re-rating? Several are key. If, over the next two to three years, AMD keeps proving it can not only ramp volume in AI accelerator platforms but also steadily lift its software ecosystem, system-level capability, and gross margin, then my low-valuation judgment today may be too conservative. Conversely, if Data Center revenue growth clearly slows, gross margin fails to stay above 50%, inventory keeps rising, and customers begin shifting more toward in-house design or NVIDIA-exclusive platforms, that would show the premium the market pays today has no fundamental support.

Comparing AMD with several kinds of alternative opportunities, the answer is also blunt. Versus NVIDIA, AMD's quality is not bad, but its platform capability and financial quality are still a notch weaker, and its valuation is not necessarily safer; versus the S&P 500 index, AMD carries higher single-stock risk, technology-path risk, and valuation-drawdown risk, yet the current price does not clearly offer a higher expected return; versus the current roughly 4.6% yield on U.S. 10-year Treasuries, AMD of course has far greater long-term growth potential, but its price has already front-loaded too much of that potential. For a balanced, long-term investor, buying AMD today is not meaningfully better than buying the index or waiting for a better price.

Open questions and limitations: First, I have not built an engineering-level model in this report down to SKU, customer, AI-accelerator BOM, and packaging supply; second, some peer-comparison metrics use the most recent public filings and current quotes for approximate calculations, so the bases are not entirely uniform; third, if AI demand over the next two to three years far exceeds the optimistic scenario in my model, the valuation ranges in this report will be too low. None of this changes a more central conclusion: at the current price, the margin of safety is insufficient.

Investment Checklist and Final Conclusion

First, the checklist.

Final Rating

Watch

One-Sentence Investment Thesis

AMD is a high-quality semiconductor design company that is getting stronger, but buying at the current price is more like overpaying for "a very good future" than buying real cash flow with a margin of safety.

Core Bull Case

• Data Center has become the core growth engine, with 2025 revenue of $16.635 billion and Q1 2026 up another 57% year over year.

• The competitive position in server CPUs has improved markedly, with Q1 2026 server x86 revenue share of about 46.2%.

• The business mix is better than before, with Xilinx/embedded/FPGA improving product breadth and adding some long-cycle characteristics.

• The balance sheet is strong, with Q1 2026 cash and short-term investments of about $12.347 billion, far above total debt.

• Operating execution is proven, with gross margin improving from 45% in 2022 to 50% in 2025, reaching 53% in Q1 2026.

Core Bear Case

• The current valuation is extremely high, with the stock at about $414, a P/E of about 135.8x, and very aggressive P/FCF and EV/EBITDA as well.

• The AI-platform moat is still clearly weaker than NVIDIA's; AMD looks more like a strong challenger than a platform overlord.

• Export controls have already caused real losses, and future policy could still broaden the impact.

• Growth comes with working-capital usage; the 2025 inventory build was a real drain on cash flow.

• Buybacks have not truly shrunk the share count, which still rose modestly, so shareholder dilution has not been fully offset.

Key Assumptions

• AMD can keep expanding its server CPU share and defend high ASPs.

• The AI accelerator business can keep scaling, rather than being a one-off ramp.

• TSMC and advanced-packaging supply will not form a persistent bottleneck.

• U.S. export controls will not permanently exclude more high-end products from key markets.

• Management will not keep using highly valued stock for low-return acquisitions.

• The valuation will eventually converge toward cash-flow capability, rather than being propped up forever by sentiment.

Fair Buy Price

$55 to $95. The basis: within my conservative-to-neutral intrinsic-value range, leaving an additional margin of safety of about 30% to 40% is more in keeping with the discipline of a long-term value investor.

Target Holding Period

If bought at a reasonable or even somewhat low price, I think it suits a holding period of 5 to 10 years or more; but at the current price, it is better to track rather than build a position.

Expected Annualized Return

The following is a rough inference assuming a purchase at around the current $414, based on different operating and exit assumptions—not a price forecast, but a reverse estimate of "what you might roughly end up with if you buy today":

The reason even the optimistic scenario is not high is not that AMD's business is poor, but that the current price has already pulled forward a lot of future growth.

Maximum Loss Risk

I think the worst permanent-capital-loss scenario is not bankruptcy, but that the company keeps doing very well, yet the market discovers it has not grown into the super-platform monopolist the current valuation implies. In that case, if the valuation reverts to 20–35x earnings while profit growth falls short of expectations, the stock could have 60% to 85% of medium-to-long-term drawdown room.

Tracking Metrics

Going forward, I will focus on the following metrics rather than the short-term stock price:

• Data Center revenue growth and its continuity.

• Data Center operating margin.

• Whether overall company gross margin stays above 50%.

• Changes in server CPU share and ASP.

• The revenue share of AI accelerators and the breadth of customer expansion.

• Whether inventory and receivables grow faster than revenue.

• Free cash flow on a continuing-operations basis.

• Whether the share count keeps being diluted by equity compensation.

• The real impact of export controls on the MI-series products.

• TSMC/advanced-packaging supply and the cadence of delivery.

Signals That Trigger Re-Evaluation

Once any of the following appears, I think the investment logic must be reviewed:

• Data Center revenue clearly slows for several consecutive quarters.

• Gross margin falls back below 50% with no visible path to recovery.

• AI accelerators ramp in volume but margins do not improve in step.

• Inventory keeps rising sharply and cash flow deteriorates.

• Important customers shift to in-house design or switch platforms.

• Export restrictions expand to more high-end products.

• Buybacks still fail to offset dilution, or another high-cost acquisition is made.

Final Recommendation

Calmly put, AMD deserves respect, but the current price does not deserve impulse. If you are a long-term business owner, you should put your attention on three things: whether this company can turn its AI and data-center advantages into a decade of high-return cash flow; whether competition and regulation will leave it as only an "excellent number two"; and whether you can buy it at a price that does not require flawless execution to earn a return. At today's price, the third point is not met. For balanced, long-term-oriented investors, my advice is not "FOMO-style chasing," but high-quality tracking, patient waiting, and acting only when a margin of safety appears.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.