Advanced Micro Devices (AMD)
Tecnología / Semiconductores
The #2 in semiconductors: gaining server CPU share from Intel (EPYC 46%) and NVDA's #2 challenger in AI GPUs (Instinct MI400, OpenAI/Meta deals for 12 GW). But it rallied +237% off the low to an extreme valuation (~105× net operating profit, a premium over NVDA) → Overvalued: the AI story is fully priced in and then some; the OpenAI/Meta warrant (~20% dilution) is an overhang.
- Price
- $476.88
- Intrinsic value (5y, base)
- $378
- Total annual return (5y)
- -4.5%
- Status (nominal)
- Overvalued
- Margin of safety
- No margin
The essentials
- The #2 semiconductor company: gaining server CPU share from Intel (EPYC 46% of x86 revenue) and NVDA's #2 challenger in AI GPUs (Instinct, OpenAI/Meta deals for 12 GW). Revenue $34.6bn (+34%), Data Center the engine (+32%). Net cash $9bn, no dividend.
- GAAP is distorted by Xilinx amortization (~$2.3bn/year); non-GAAP (EPS $4.17) adds it back but also adds back SBC ($1.6bn, a real cost — 20% of FCF, §3) → it is valued on economic EV/NOPAT, which expenses SBC. It is a net diluter (SBC + warrants), not a repurchaser.
- Rallied +237% off the low ($149→~$477, high $585) to an extreme valuation: ~105× net operating profit (or ~51× forward non-GAAP, a premium over NVDA's ~21-24×) → Overvalued. Even a spectacular AI ramp barely justifies the price; the OpenAI/Meta warrant (~20% dilution) subtracts. Stacked optimism.
Intrinsic value — two valuation methods
By both methods, the value today (DCF $287 · Multiples $303) is below the market price ($477).
Pillars of the analysis
The verdict — today vs 5 years
Today — expensive, no margin of safety: at $477 trades ~57.2% above its value discounted to today (~$303); the expected return does not even reach the risk-free rate (4.5%).
At 5 years — Sobrevalorado: the expected total return is negative — the price already discounts a demanding scenario that, if not met, results in a loss.
The bridge: the return at 5 years falls below the risk-free rate (4.5%) — which is why there is not even a discount to today's value. To require a 15% annual return, it would need to be bought at ~$188.
Thesis
The business
AMD is a rising-quality semiconductor company: it gains CPU share from Intel (a solid, structural engine) and is NVDA's #2 challenger in AI GPUs (the explosive engine, validated by OpenAI/Meta). It is a very high-value business (intellectual property, chiplet architecture) with a net-cash balance sheet. But the moat in AI — the market that justifies the valuation — is a challenger's moat against NVDA's CUDA, and SBC plus the OpenAI/Meta warrants dilute. The quality of the business is real; the question is the price.
The valuation
Valued on EV/economic NOPAT (net operating profit, which adds back Xilinx's non-cash amortization but expenses SBC — unlike the non-GAAP figure the market uses, which adds SBC back, a real cost, §3). The base scenario projects a value of ~$378 per share in five years, an annual return of ~-5% from the current ~$477.
The tension: AMD rallied +237% off the low ($149 → high $585 → ~$477) and trades at ~105× today's net operating profit (or ~51× forward non-GAAP earnings, a premium over NVDA's ~21-24×). Even projecting an aggressive AI ramp — revenue tripling to ~$126bn by 2031 with margin expanding to ~24% — net operating profit reaches ~$25.6bn, which at a disciplined multiple of ~24× (top of the semis band) gives a value below the current price. The market is paying for flawless execution AND more; it's the case of stacked optimism (aggressive growth × premium multiple).
The margin of safety
The verdict is Overvalued: No margin of safety: the price already discounts a demanding scenario.. At ~$477 AMD trades at an extreme valuation (~105× net operating profit, a premium over NVDA) that already prices in a spectacular AI ramp. There is no margin of safety: even the favorable scenario — the AI ramp fully materializes, margin expands, the multiple holds — barely reaches today's price, and the OpenAI/Meta warrant (~20% dilution) subtracts from it. The expected return (~-5%) is negative in the base case: the growth, however real, is already paid for. For a value investor, this is a great business at a price that leaves no room to run — a case for selling the euphoria, not buying it.
What to watch
The central disconfirmation test: does ROCm close the gap with CUDA and does AMD capture a large, profitable share of the AI GPU market, or does it remain the second-choice supplier at lower prices? The signals: the ramp of Instinct GPUs (MI400/Helios in 2026), Data Center gross margin (is AI accretive or dilutive to margin?), server CPU share versus Intel, and the vesting of the OpenAI/Meta warrants. And the macro risk: semiconductor cyclicality (AMD had losses in prior cycle downturns) and export controls to China. At this valuation, any execution stumble is punished hard.
Educational / informational. Does not constitute investment advice.