INTRODUCTION TO AMD

AMD is one of the two designers of the chips that sit at the heart of modern computing. It sells its products to the largest technology customers on the planet: the cloud operators that build artificial intelligence, the companies that make personal computers, and the console makers that ship PlayStations and Xboxes. Understanding AMD requires first understanding what a chip company actually does, and why the ones that succeed become some of the most valuable businesses in the world.

The 7-Year-Old Explanation

AMD designs the brains of computers. The most valuable brains today are not the ones that run a laptop. They are the ones that run artificial intelligence inside the giant warehouses owned by Microsoft, Meta, and Amazon. These warehouses, called data centres, use tens of thousands of chips at once to train the models that answer ChatGPT questions.

AMD makes two main families of chips. The first is CPUs, which are the general-purpose processors that run any computer or server. AMD sells its CPUs as Ryzen for personal computers and EPYC for servers. The second family is GPUs, originally designed to draw the pictures on a video-game screen but now the preferred chip for artificial intelligence. AMD sells its GPUs as Radeon for gaming and Instinct for the data centre.

AMD also owns Xilinx, a company it bought in 2022 for $49 billion. Xilinx makes FPGAs, a specialised type of chip that can be reprogrammed after it leaves the factory. These FPGA chips end up inside satellites, factory robots, cars, and the equipment that runs the telephone network.

AMD does not own a single chip factory. This is unusual, because chip factories are among the most expensive facilities on Earth (a leading-edge factory costs more than $20 billion to build). Instead of building its own, AMD designs its chips and pays a company in Taiwan called TSMC to manufacture them. This design-only approach is called the fabless model, and it has important consequences for the whole shape of the business.

The reason AMD matters right now is that the world is spending an unprecedented amount of money on artificial intelligence. Building a large language model like GPT or Llama requires the physical thing that actually runs the calculations: a chip. Nvidia makes the best chip for that job today. AMD makes the second-best. Being second in a race that is currently spending trillions of dollars on infrastructure can still be very valuable, provided the gap between first and second stays small and provided customers want more than one supplier. That is the whole thesis: AMD is a credible second source in an enormous, structurally growing market.

The Investor Explanation

AMD is a fabless semiconductor designer headquartered in Santa Clara, California. It reports three business segments, and the way those three segments contribute to revenue and profit tells the whole story of the current phase of the company. Data Center is the fastest-growing segment and now sits at just under half of total revenue. Client and Gaming, which covers the personal-computer and console lines, is roughly the same size but growing slowly. Embedded, the Xilinx franchise acquired in 2022, is the smallest of the three but carries the highest-quality margins because it serves customers with very long product cycles who cannot easily switch suppliers.

Manufacturing is outsourced to TSMC in Taiwan on the most advanced production nodes available. The memory chips that sit next to AMD’s AI accelerators (a very specialised type of memory called HBM3e) come from SK Hynix and Micron. The company was founded in 1969 by Jerry Sanders as a second-source alternative to Intel, and it has operated as an Intel competitor for its entire history. Lisa Su has been Chief Executive Officer since October 2014 and is the primary architect of the current strategy.

PILLAR 1: CIRCLE OF COMPETENCE

Valuing a company first requires understanding where its revenue comes from, who its competitors are, and what forces shape the industry it operates in. This pillar builds that understanding for AMD.

Business Segments

Revenue mix is the share of a company’s total revenue that comes from each of its business lines. Understanding it matters because a shift in the mix changes the whole character of a business. A company that sells commodity products at low margins looks very different from a company that sells premium products at high margins. Watching a company’s mix shift over time is one of the earliest signals an investor gets about which direction the business is heading.

In AMD’s case the revenue mix is doing something interesting. Two years ago the Data Center segment was under 40% of total revenue. In Q1 FY26 it reached 56%, or $5.8 billion of the $10.3 billion of total quarterly revenue. It is now on a trajectory to be a clear majority of the company within the next two years. This matters because Data Center is by far the highest-margin part of AMD’s business. As the mix shifts toward Data Center, the corporate gross margin rises even if none of the individual product prices change. Growth in the right segment does double duty: it adds revenue and it lifts the average profitability of every dollar of revenue.

For context, Nvidia is already about 90% Data Center. Intel is the opposite: roughly 55% personal-computer chips and only 30% Data Center. AMD sits between the two today and is moving toward the Nvidia shape. Whether that transition completes is the central question every investor in AMD has to answer for themselves.

The 3 Business Segments

Data Center

The Data Center segment is AMD’s earnings engine. It sells two main products: EPYC server CPUs, which run the general-purpose computing workloads inside cloud data centres, and the Instinct family of AI accelerators, which train and run artificial-intelligence models. In Q1 FY26 this segment generated $5.8 billion of revenue (56% of AMD’s total quarterly revenue of $10.3 billion) and it is growing at roughly two times the rate of the rest of the business.

Client and Gaming

The Client and Gaming segment combines the Ryzen chips that sit inside consumer laptops and desktops, the Radeon graphics cards used by PC gamers, and the semi-custom chips sold under long-term contracts to Sony and Microsoft for the current generation of gaming consoles. In Q1 FY26 this segment represented roughly 30% of total revenue. Growth here runs in the mid single digits, because the personal-computer market and the console market both grow slowly and cyclically.

Embedded

The Embedded segment is the Xilinx franchise that AMD acquired in February 2022. It sells FPGAs into aerospace, defence, automotive, industrial automation, and communications infrastructure. In Q1 FY26 it represented roughly 10% of total revenue. This is the smallest of the three segments, but it carries the highest-quality margins because its customers embed Xilinx chips inside products with 20-year lifecycles that cannot easily be redesigned around a competing chip. The segment went through a multi-year inventory correction after the post-pandemic boom, and only in the most recent quarter did it return to positive year-on-year growth. That inflection is the anchor of the Bull case in Pillar 6.

How AMD Earns a Dollar of Profit

Not every dollar of AMD’s revenue is equally profitable. The highest-margin dollar comes from an Instinct AI accelerator sold to a hyperscaler under a multi-year commitment. The lowest-margin dollar comes from the consumer and console lines inside Client and Gaming, where volumes are large but margins are thin by design.

This is why the mix shift toward Data Center matters so much. Every percentage point of revenue that moves from Client and Gaming into Data Center adds several times as much profit as it does revenue. The corporate gross margin near 51% in the last full year, expanding from 47% the year before, is almost entirely explained by this mix shift. If consensus is right, the corporate gross margin continues climbing toward the mid-fifties over the next few years as the mix shift continues.

Market Share

Market share is the percentage of a defined market that a company captures, either by revenue or by unit volume. It matters because it shows how customers are voting with real orders, not with survey responses. A company that is gaining market share against a stronger incumbent is usually improving its underlying economics faster than the headline revenue number suggests. A company that is losing market share tends to see profit compression well before the revenue actually starts to fall.

AMD’s market-share position has two parts that must be read together. In server CPUs, AMD has taken share from Intel every year since 2018. The share number moved from below 5% to above 35% over that stretch. That is a durable tailwind that continues to compound. In AI accelerators, AMD holds only 5% to 10% share against Nvidia’s 90%. That is a very small challenger against a very entrenched leader, and it explains why AMD’s AI franchise is worth debating rather than taking on faith.

The blended picture is a company that is a durable share-gainer in one important market (server CPUs, where Intel is structurally weaker) and a small credible challenger in another (AI accelerators, where Nvidia is structurally stronger). Both trends are positive for AMD, but the AI story is the one that matters most for the current valuation because that is where the market is spending most of its money and most of its attention.

The Competitive Landscape

AMD competes against three specific companies in three specific ways.

Nvidia (NVDA)

AMD is the second source in AI accelerators. Nvidia’s gross margin runs near 75% versus AMD’s 53% (Q1 FY26 actual), and that gap is the clearest single measure of the software advantage Nvidia holds through its CUDA platform. CUDA is more than a decade old, deeply embedded in the workflows of every hyperscaler, and represents the single biggest reason any AI-accelerator buyer chooses Nvidia by default. AMD’s differentiation is not about beating CUDA. It is about offering more memory per chip package, lower cost per training run, and the strategic value of being a second source for customers who do not want to depend on a single supplier.

Intel (INTC)

AMD is the challenger in x86 server CPUs, and that challenge has been winning for eight consecutive years. Intel’s decision to expand into contract manufacturing while its product roadmap slipped left AMD an opening that AMD has systematically exploited. There is no obvious sign that this tailwind is close to reversing.

Broadcom (AVGO)

AMD faces a newer and less-modelled threat here. Broadcom designs the custom chips that hyperscalers use for their own in-house AI workloads: Google’s TPU, Meta’s MTIA, Microsoft’s Maia. As the biggest AI customers vertically integrate their inference silicon, the general-purpose GPU market at the very top end shrinks. This is a risk to both AMD and Nvidia, but it is a risk AMD is less well positioned to absorb because AMD has less pricing power to lose in the first place.

Customer Concentration

Customer concentration is the share of a company’s total revenue that comes from a small number of large customers. Investors watch this metric because losing a single large customer can knock the whole revenue path lower at once. A business that leans on a few buyers usually trades at a lower valuation to make up for that all-or-nothing risk.

AMD’s customer concentration has been rising as the AI franchise scales. The top two AI customers, Microsoft and Meta, together are already approaching a quarter of total company revenue. That is meaningful. Losing either one would rebase the whole forecast lower and force the model to shift toward the Bear case. For context, Nvidia is even more concentrated (its top five customers, all hyperscalers, represent well over 40% of revenue). Intel is more diversified because it sells into tens of thousands of enterprise customers rather than a handful of cloud giants.

The rising concentration is not a reason to reject the AMD thesis. It is a reason to insist on a larger cushion between price and intrinsic value when buying, because the downside case (a lost customer) is more damaging than it would be for a more diversified business. This is why Pillar 7 sets the required margin of safety at 35% to 40% rather than the standard 30%.

Top 3 Customers

Microsoft Azure

Microsoft Azure buys EPYC chips for general-purpose cloud computing and Instinct GPUs for the artificial-intelligence workloads that run on Azure. Microsoft’s motivation for buying from AMD rather than Nvidia is supply diversification: no cloud provider wants to depend on a single chip supplier for its most strategically important product line.

Meta

Meta has committed to 6 gigawatts of MI450 GPUs for its Helios rack-scale training platform. This is one of the largest AI-accelerator commitments any customer has ever announced. Meta chose AMD because the MI450’s larger memory capacity per package is a genuine advantage for training the Llama family of models at the scale Meta is trying to reach.

OpenAI

OpenAI has a six-gigawatt strategic partnership with AMD, with first deployment expected in the second half of the current fiscal year. The motivation is capacity: OpenAI is growing so fast that Nvidia alone cannot ship into it fast enough. AMD is the release valve that lets OpenAI keep expanding without waiting for Nvidia to catch up.

The Economics of This Industry

The semiconductor industry has a distinctive economic profile that matters for how the numbers in Pillar 4, Pillar 5, and Pillar 6 shake out. Five features are worth understanding before moving on.

Cyclicality

Cyclicality is moderate. AI demand is secular and growing through the whole forecast window. The personal-computer, console, and industrial parts of AMD’s business are cyclical and can move roughly 20% from peak to trough. The blended business sits somewhere in between.

Capital Intensity

Capital intensity is unusually low for a semiconductor company. Because AMD is fabless, it does not have to spend billions of dollars per year on chip factories. Its own capital expenditure runs at less than 3% of revenue, mostly on design tools and computing infrastructure. Intel, which owns its own factories, spends more than 25% of revenue on capital. That gap is the single biggest structural advantage of the fabless model, and it flows straight through to free cash flow in Pillar 5.

Customer Concentration

Customer concentration is rising. The top two customers (Microsoft and Meta) together are approaching a quarter of total revenue as AI commitments concentrate.

Supplier Concentration

Supplier concentration is very high. TSMC in Taiwan supplies more than 90% of AMD’s leading-edge wafer capacity. There is no meaningful second source. The Taiwan geopolitical exposure is carried in full by anyone who owns AMD shares.

Regulatory and Geopolitical

The regulatory environment matters. US export controls administered by the Bureau of Industry and Security already block AMD from selling its most powerful AI chips (the MI308) into China. Restrictions on the MI355 are partial. The MI400 is the next test case. Extension of restrictions to the MI400 would remove a meaningful chunk of the forecast revenue for the second half of the decade.

Verdict

AMD’s circle of competence is hyperscaler GPU accelerators and server CPUs, with Xilinx FPGAs as the wide-moat anchor. The position against Nvidia is structurally narrow, but the business is understandable and its revenue drivers are visible. That is enough to proceed to the deeper pillars. The two things to carry forward are the mix shift toward Data Center (positive for margin trajectory) and the rising customer concentration (negative for the margin of safety requirement).

PILLAR 2: BUSINESS QUALITY AND DURABLE MOAT

Every serious equity investor is trying to answer the same question. Does this business have a durable advantage, one that lets it earn returns above an ordinary company? And will that advantage still be there in ten years? That advantage, if it exists, is called an economic moat. This pillar walks through the concept of the moat, the specific measurements that reveal whether it exists, and how AMD scores on each of them.

Return on Invested Capital (ROIC)

The cleanest test of business quality compares two numbers. The first is return on invested capital (ROIC): the return a company earns on every dollar it puts to work. The second is its weighted average cost of capital (WACC): what that money costs to raise.

If ROIC is above WACC, the business is creating value on every dollar it reinvests. Its owners are getting more than the market would give them elsewhere. If ROIC is below WACC, the business is destroying value on every dollar it reinvests, because the same money could have earned more sitting in another asset. The gap between ROIC and WACC, measured in percentage points, is the single most important measurable quality metric of a business.

AMD’s blended ROIC in the last full year was roughly 6%. Its WACC is 11.88%. That is a shortfall of about 6 percentage points per year on invested capital, which is a genuine weakness in headline terms. But the blended number hides an important fact: incremental returns on Data Center investments sit well above 40%, comfortably clearing the WACC hurdle. It is the older segments (Client, Gaming, and the still-recovering Embedded) that earn below the hurdle and drag the blend down.

For context, Nvidia’s blended ROIC sits near 55%, which is a 42-percentage-point premium to its WACC and the clearest single measure of how valuable the CUDA software franchise is. Intel’s blended ROIC is close to 3%, below its WACC, meaning Intel is currently destroying value. TSMC’s blended ROIC is roughly 24%, well above its own WACC.

The AMD thesis rests on the blended ROIC crossing above WACC by roughly 2027. If the Data Center mix keeps expanding and the incremental Data Center ROIC stays where it is, the blend arithmetically has to lift. If either of those two things breaks, the intrinsic value collapses toward the Bear case.

Gross Margin

Gross margin is the share of each sales dollar that remains after paying the direct cost of the product. Gross margin is the single cleanest read on pricing power. Businesses with strong competitive advantages sustain high gross margins because their customers pay a premium for what only they can provide. Commodity businesses see their gross margins compress to the marginal cost of production, because customers can substitute one supplier’s product for another with no consequence.

AMD’s gross margin was 53% in Q1 FY26, up from 51% in FY25 and 47% in FY24. That expansion is not the result of any single price rise. This is almost entirely a mix-shift story. Data Center carries gross margins well above the company average. So as Data Center grows as a share of revenue, the company-wide average margin rises on its own, even if no single product gets more expensive.

For peer context, Nvidia sits near 75% (the CUDA software moat plus AI-cycle scale). Intel sits near 40% (structural decline as it loses share to AMD and to Arm). TSMC sits near 60% (foundry pricing power at the leading edge, 59.9% FY25 actual). AMD sits between Intel and TSMC, well below Nvidia. Consensus points toward AMD gross margin expanding another two or three percentage points over the next couple of years as the MI355X ramps.

The 4 Types of Moat

The financial analyst Pat Dorsey codified competitive advantage into four categories, and this framework is a useful lens through which to view any business. The four categories are intangible assets, switching costs, network effects, and cost advantage. A business with none of these is a commodity. A business with one strong one has a moat. A business with two or more strong ones is a wide-moat business of the type Buffett prefers to own.

Intangible assets: Wide in Embedded. Xilinx FPGAs embed inside satellites, factory equipment, and defence systems with product lifecycles of 20 years or more. Once a customer designs a Xilinx chip into a satellite, that chip is not swapped out for the life of the satellite. This is the strongest moat AMD has anywhere in the business, and it is the reason the Embedded segment supports the wide-moat piece of the whole thesis. In Data Center, intangible advantages are much weaker because hyperscalers can and do run mixed fleets of chips.

Switching costs: Moderate to high in Embedded, because the FPGA design tools (Vivado) and the requalification cycle for a new chip in a safety-critical product take 12 to 18 months. In Data Center, switching costs are low: hyperscalers deliberately keep their software portable across chip suppliers to preserve negotiating leverage.

Network effects: None in any AMD segment. Chips are inputs to a customer’s product, not platforms with user-driven network value. This is one of the categories where AMD has no advantage at all.

Cost advantage: Narrow. TSMC manufactures for both AMD and Nvidia, so there is no captive-fab cost moat. AMD’s chiplet architecture gives it a modest yield and cost advantage in server CPUs relative to Intel’s monolithic dies, but that lead can be replicated by any competitor willing to redesign around the same approach. It is a lead, not a permanent moat.

Moat Trajectory

The direction of a moat matters as much as its current width. AMD’s moat is widening in Embedded as Xilinx FPGAs find new applications at the artificial-intelligence inference edge, where power efficiency matters more than raw throughput. AMD’s moat is narrowing in Data Center as Nvidia’s next architecture and Broadcom’s custom-chip franchise both erode the value of being a second source. On net, the blended moat is stable, with the Embedded widening compensating for the Data Center erosion.

Verdict

AMD is a narrow-blended-moat business. Wide only in Embedded, which is 10% of revenue today but the source of the Bull thesis. This verdict carries straight into Pillar 7. Because the moat is only narrow and the AI-chip race is uncertain, the margin of safety we demand rises to 35% to 40%, above the usual 30% benchmark.

PILLAR 3: MANAGEMENT QUALITY AND CAPITAL ALLOCATION

Two things separate a great business from a merely good one over decades: the durability of its competitive advantages, and the quality of the people running it. Pillar 2 covered the first. This pillar covers the second.

Operational Track Record

When investors talk about management quality, they usually mean track record: the decisions a CEO has made over her tenure, and how they played out against the promises she made at the time. Track record is the closest thing to hard evidence. It matters enormously. A great operator can compound value over decades; a mediocre one in the same seat can destroy it in a few years.

Lisa Su became Chief Executive Officer of AMD in October 2014. The company she inherited was in serious trouble: shares were trading near $2, revenue was about $5 billion and shrinking, gross margin was under 30%, and free cash flow was negative. Eleven years later the shares trade at $548.13, FY25 revenue was above $46 billion (and Q1 FY26 alone printed $10.3 billion, up 38% year on year), gross margin is 53%, and free cash flow runs above $5 billion per year. That is not an incremental improvement. It is a full transformation.

For contrast, Intel has cycled through three CEOs over the same eleven-year window (Brian Krzanich, Bob Swan, Pat Gelsinger) while losing market share to AMD every year. Had Lisa Su gone to Intel instead of staying at AMD, the two companies would have swapped fortunes.

4 Operational Wins in 11 Years

The track record breaks down into four specific wins, each of which would be a career achievement in its own right.

Roadmap execution: The Zen microarchitecture roadmap on schedule for eight straight generations. No other semiconductor CEO can match that track record over that window.

Server share: x86 server CPU market share from below 5% to more than 35%, taking share from Intel every year since 2018.

Xilinx integration: The 2022 Xilinx deal (worth $49 billion in stock) closed and integrated cleanly. Xilinx’s segment margins survived the integration, which is unusual for a semiconductor deal of that size. This is what powers the wide-moat piece of the thesis identified in Pillar 2.

AI pivot: From a standing start in 2020 to a $10-billion-plus Instinct business by 2026, with named commitments from Microsoft, Meta, Oracle, and OpenAI (the same customers named in Pillar 1).

Capital Allocation

Capital allocation is how management deploys the cash the business generates. Every dollar of cash a business produces has to be spent somewhere: reinvested in the business, used to buy another business, paid out as a dividend, used to buy back shares, or held on the balance sheet. Capital allocation is the clearest read on whether management is serving owners or diluting them.

AMD reinvests about 17% of its revenue back into research and development each year. That is a heavy reinvestment rate by any standard, and it is the right choice for a business whose product roadmap has to stay competitive with Nvidia and Intel. Any less would risk falling behind on the process node or on the chip architecture.

The 2022 Xilinx acquisition, at $49 billion in stock, was the most expensive semiconductor deal in history at the time it was announced. Strategically, it was the right call, because it added a genuine wide-moat franchise to a business that was otherwise a narrow-moat challenger. Financially, the price paid was full, and the post-acquisition ROIC contribution from Xilinx is still recovering from the industrial-inventory correction. This is a case where the strategy was clearly right even though the price was not obviously a bargain.

The buyback record is disciplined. In the last full year AMD repurchased about $1.5 billion of its own shares at an estimated average price of $140. Today’s share price is $548.13, so those buybacks are already worth about four times what AMD paid for them. Management has publicly signalled that the remaining buyback authorisation will be deployed more selectively at current price levels, which is exactly the right instinct.

Insider Ownership and Compensation

Lisa Su holds roughly 1% of AMD’s shares outstanding. That is standard for a professional large-cap semiconductor CEO who was not the founder. It is not the founder-operator ownership structure Buffett prefers, where the CEO holds 10% or more of the business, but it is enough that her personal fortune moves meaningfully with the share price.

Executive compensation is primarily performance share units tied to relative total shareholder return and adjusted operating income. This is a well-designed structure: it aligns her personal outcomes with multi-year shareholder outcomes rather than with quarterly earnings-per-share beats.

Verdict

Excellent operator with grade-A capital allocation. The four operational wins comprise an eleven-year record that few semiconductor CEOs anywhere can match. The principal unmodelled risk is CEO succession: Lisa Su is 56 years old and there is no publicly identified successor.

PILLAR 4: FINANCIAL HEALTH AND BALANCE SHEET (Q1 FY26)

The balance sheet reveals whether a business can survive the downside cases and whether it can fund the upside cases. Weak balance sheets force companies to issue equity at the worst possible moments, which permanently dilutes existing shareholders. Strong balance sheets do the opposite: they let a company buy back shares in downturns, when everyone else is a forced seller. AMD is firmly in the second camp.

How Much Debt versus How Much Cash

Any balance-sheet analysis begins with two figures: how much the company owes, and how much cash it holds against those obligations. At the end of the most recent quarter, AMD held $12.3 billion of cash on its balance sheet against $3.2 billion of total debt. The difference, $9.1 billion of net cash, is money the company owes nobody and can deploy at its own discretion.

The professional shorthand for this comparison is the ratio of net debt to EBITDA. AMD’s is negative 1.7 times, which is the same as saying the company holds about 1.7 years’ worth of operating cash flow more than it owes. That is a strong position. Nvidia’s ratio is even stronger at around negative 2.5 times. Intel’s ratio is positive 2.5 times, meaning Intel carries about 2.5 years of operating cash flow of net debt. That is manageable but genuinely burdensome. TSMC’s ratio is around negative 1.0 times.

AMD has the second-strongest balance sheet in the semiconductor peer group, behind only Nvidia. This is a support to the thesis, not a constraint on it.

Interest Coverage and Debt Maturities

A related question is whether the company earns enough operating profit to comfortably cover the interest it pays on its debt. AMD earns operating profit of about $3.7 billion per year and pays interest of roughly $150 million per year, so its interest coverage is above 24 times. Anything above 10 times is considered very safe. AMD is not stretched.

The debt maturity schedule shows when a company has to repay or refinance the debt it currently owes. AMD’s debt is laddered across a series of maturities running from 2027 out to 2052. Nothing is due before 2027, which means AMD does not have to enter the bond market in any near-term year to refinance. That protects AMD from a spike in interest rates.

Goodwill

When a company buys another company, it usually pays more than the target’s tangible assets are worth on paper. The excess price paid over tangible book value shows up as an intangible line called goodwill. Goodwill sits on the balance sheet indefinitely and is written down (impaired) only when the acquired business demonstrably underperforms the price paid.

AMD’s balance sheet carries $25.3 billion of goodwill from the Xilinx acquisition, on total assets of $79.6 billion. That is roughly 32% of total assets, which is high by semiconductor standards. Nvidia’s goodwill is less than 5% of assets. Intel’s is close to 10%. TSMC’s is near 1%.

The high goodwill share is not itself a red flag today, because AMD’s auditors have not indicated any impairment. But it is a latent risk. If the Embedded segment fails to recover to its historical 36% to 42% operating margin range, AMD may eventually have to write down some of that goodwill. A write-down would not affect cash generation, but it would reduce reported equity in the quarter it happens and could rattle the share price. This risk is monitored explicitly in Pillar 10.

Verdict

The balance sheet is a fortress. $9.1 billion of net cash lets AMD fund every reasonable growth scenario without diluting shareholders. The Xilinx goodwill is the one line to keep an eye on, but no imminent impairment risk exists today. Balance sheet supports the thesis, not constrains it.

PILLAR 5: EARNINGS QUALITY

Not every dollar of reported profit is a real dollar of cash. Companies with weak earnings quality use accounting choices (accruals, reserve reversals, non-recurring gain reclassifications) to make their reported profits look better than the cash they actually generate. Companies with strong earnings quality do the opposite: their cash generation runs at or above their reported profit, because their accounting is conservative. This pillar tests whether AMD’s earnings are the real thing.

The Gap Between Accounting Profit and Real Cash

In the last full fiscal year AMD reported net income of about $4.3 billion. Over that same period the business actually generated free cash flow of about $5.5 billion. The cash produced was roughly 27% higher than the profit reported.

That gap exists for a specific and mostly benign reason. When AMD acquired Xilinx in 2022, the deal created a large stack of intangible assets: patents, customer relationships, brand value, and other items the accountants require to be written down over 7 to 10 years. Every year that write-down (called amortisation) reduces reported net income but no cash leaves the business. So the accounting profit understates the real cash economics for as long as the Xilinx amortisation schedule keeps running.

By roughly 2029, that Xilinx amortisation will be substantially complete. At that point the two numbers, net income and free cash flow, will converge. Reported earnings will look significantly better than they do today, purely because a paper charge will have finished working its way through the income statement. This is a structural tailwind for reported earnings that is already visible in the disclosures on close reading.

How Operating Cash Converts into Free Cash

The other clean read on earnings quality is the ratio of operating cash flow to net income. AMD’s operating cash flow last year was about $6.5 billion, which is roughly 1.5 times its net income of $4.3 billion. A ratio above 1.0 is healthy. It says the working capital of the business (receivables, inventory, payables) is scaling sensibly with growth rather than absorbing cash the way it does in businesses with hidden accrual problems.

In AMD’s case, the working capital is genuinely scaling with growth. HBM3e memory (the specialised memory that sits next to each Instinct AI accelerator) has to be pre-purchased months before the accelerators ship, and that pre-purchase temporarily ties up cash in inventory. This is a sensible cost of running the Data Center growth story, not a warning sign about earnings quality.

Stock-Based Compensation

One accounting item deserves special attention because it is increasingly the largest hidden cost in technology-company earnings: stock-based compensation, or SBC. SBC is the portion of employee pay issued as shares in the company rather than as cash. Because it does not leave the balance sheet as cash, many companies present a version of their earnings that adds SBC back to make cash generation look higher.

This is a mistake, and a serious one. Every share issued to an employee dilutes the ownership stake of every existing shareholder by a small amount. Over time these small dilutions add up to material ownership transfer from existing owners to employees. Any honest valuation charges SBC as a real cost, because it is a real cost, even though it is not a cash cost.

AMD spends about 4.7% of revenue on SBC each year. That is in line with Nvidia (also 4% to 5%) and higher than Intel (3% to 4%). The higher rate reflects that AMD is competing for the same engineering talent as Nvidia, and that talent is expensive to attract in the current AI hiring market. For scale, Meta and Google both run SBC above 10% of revenue, so AMD’s 4.7% is not extreme by the standards of the broader software-and-hardware peer set.

The Compoundex methodology charges SBC in full in the DCF and never adds it back. This is more conservative than the sell-side norm and it reduces AMD’s intrinsic value estimate by roughly 12% compared with an add-back approach. It is the right treatment because it represents the true cost of running the business.

Verdict

AMD’s earnings are clean. Real cash generation runs above reported profit, cash conversion is healthy, and SBC is charged as a real cost. Nothing about the earnings pattern relies on accruals or one-off gains. The cash the valuation discounts is the cash the business actually produces.

PILLAR 6: VALUATION

This pillar values AMD three different ways: a full discounted cash flow model, a reverse-engineered version of the same model, and a shortcut method based on cash yield. All three agree that today’s price does not offer a discount.

The Headline

My Case intrinsic value stands at $492.42 per share against a market price of $548.13 on 28 June 2026. That is a fair-value margin of safety of negative 11.3%. The market charges 11% more than the conviction case says the shares are worth.

The 4 Scenarios

Each row is the DCF-implied share price under its own revenue, margin, and discount-rate switches. Margin of safety is measured against intrinsic value at a spot price of $548.13. Only the Bull case clears.

The Discount Rate (WACC)

The DCF discounts future cash back to today using a rate that represents what AMD’s investors demand at similar risk elsewhere. That rate is called the Weighted Average Cost of Capital, or WACC. For AMD it is 11.88% as of 28 June 2026. This is above the semiconductor median (Intel is closer to 9%, TSMC 10%) because AMD’s capital stack is almost all equity and its beta is high.

Reverse DCF

The reverse DCF takes today’s price and solves for the growth and margins the market must already be assuming. For AMD at $548.13, the implied path is roughly $130 billion of Data Center revenue by 2030 with sustained 30% Data Center operating margins. That is close to the conviction case. The market is not offering a discount.

Owner-Earnings Yield

A shortcut check: divide free cash flow by market capitalisation. AMD generates about $5.5 billion of free cash flow against a market cap near $904 billion, which is a yield of under 1%. Ben Graham’s rule is that this yield should at minimum clear the risk-free rate plus the equity risk premium (roughly 8.7% today). AMD falls short by about 8 percentage points. On this test the shares do not offer investment-grade returns unless future growth closes the gap.

Triangulation

The three methods agree. The DCF says the price sits at a premium to My Case intrinsic value of $492.42, roughly 11% above fair value at the current $548.13. The reverse DCF says the price already embeds the conviction case. The owner-earnings yield clears nothing. No framework offers a cheap entry point.

THE DCF WALKTHROUGH FOR AMD

The context behind each step. The story behind the specific numbers. The mechanics that drive the company toward its final intrinsic value.

Download

PILLAR 7: MARGIN OF SAFETY

Margin of safety is the cushion between what a share is worth and what the market charges for it. If intrinsic value is $500 and the market charges $400, the buyer gets a 20% margin of safety. If the market charges $520, the buyer pays a 4% premium and has no cushion at all. The size of the required cushion depends on the quality of the business: wide-moat businesses need less, narrow-moat and cyclical businesses need more.

For AMD, the required cushion is 35% to 40%. This is more than the standard 30% because the moat is narrow (per Pillar 2) and the AI accelerator race is unsettled. Today the price sits 11% above the conviction view rather than 35% to 40% below it. The cushion is not merely absent; the price is on the wrong side of the conviction view entirely.

A pullback into the $250 to $260 range would give the Base case a 30% cushion. A pullback into the $380 to $420 range would give My Case a meaningful positive cushion.

Munger Placement

A fair-to-quality business at a rich price. Wonderful companies at fair prices are the prize. Fair companies at wonderful prices work only with a deep discount. AMD offers none today.

RISK MITIGATION AND WATCHLIST

These are the five developments serious enough to break the reason for owning AMD. Each one is both damaging and reasonably likely over a three-to-five-year holding period. If one happens, the disciplined response is to trim or sell, not to keep buying more as the price falls (“averaging down”) hoping it bounces back.

Risk 1: Nvidia ships an MI400-killer architecture

AMD’s entire data-centre case assumes its new AI chips (the MI355X, and the MI400 due next) stay competitive with Nvidia’s. If Nvidia pulls far ahead again, making its chips meaningfully cheaper to run for the same work, the large cloud providers would simply swing their orders back to Nvidia, and AMD’s most important growth engine stalls.

Preparation: Each quarter, check Nvidia’s results, the independent AI-chip speed tests (the MLPerf benchmarks), and what the big cloud companies say about their spending plans.

Risk 2: Loss of a major AI customer

A small group of giant customers, OpenAI, Oracle, Meta and Microsoft, make up much of AMD’s expected AI revenue. Losing even one of them would force analysts to cut their 2027 forecasts, and the shares would likely follow.

Preparation: Watch for sudden regulatory filings (8-Ks) that flag a lost contract, and listen to the big cloud companies’ earnings calls for any change in their AMD plans.

Risk 3: Data Center EBIT margin guides down more than 2 percentage points

The profitability of the data-centre business, its operating margin, the profit left from each dollar of sales, is the single biggest driver of the optimistic case. It currently runs at 27.7%. If it slips durably below the mid-twenties, that bull case no longer holds together.

Preparation: Each quarter, read AMD’s filing and check the profit reported specifically by the Data Center segment.

Risk 4: Xilinx recovery delayed beyond FY27

The Embedded business (built on AMD’s Xilinx acquisition) is its most durable and highest-margin unit, the steady anchor beneath the AI story. The bull case needs it to recover to its normal profitability by 2027. A longer delay weakens the whole picture.

Preparation: Track the Embedded segment’s revenue and margin, and cross-check similar chipmakers (Lattice, Microchip) as an early read on the recovery.

Risk 5: US-China export-control escalation

China once accounted for more than a fifth of AMD’s sales. If US rules tighten further and block the new MI400 chips from being sold there, revenue takes a real and immediate hit.

Preparation: Watch for changes to US export rules (issued by the BIS agency) and the risk-factor section of AMD’s annual report.

CORE GRAPHICS →

6 key graphics for Advanced Micro Devices (AMD): visual trends for the actuals and the projected figures for UFCF, revenue growth %, EBIT margin, and EBIT.

FINAL THESIS

Current quality

A fair-to-quality business at a rich price. AMD is executing operationally on the MI355X ramp and the Xilinx integration, and Lisa Su's leadership is a genuine asset. The business is improving. The price, not the business, is the problem.

What to focus on next quarter

Q2 FY26 reports in August 2026. The one thing to watch is whether the Data Center operating margin holds at or above 27 percent. If it does, the conviction case strengthens; if it slips, the main support under the case gives way.

Holding period and sell discipline

Three to five years for the thesis to compound through Data Center scale, Embedded recovery, and the Client AI-PC tailwind. The position is sold on a broken thesis, not on a falling price.

Closing

As of 28 June 2026 (Q1 FY26), at $548.13, AMD is a fair-to-quality business at a rich price. Against My Case intrinsic value of $492.42, this offers no margin of safety: the price sits about 11 percent above the estimate of worth. Only the optimistic Bull scenario rises above today’s price. The discipline is to wait.