AMD compounds AI data-center compute with ~36% YoY revenue growth and gross margin recovering to ~52.8%, but at $511.59 it trades ~168x trailing EPS and ~150x EV/EBITDA after a +332% year, ~44% above ~$355 fair value, with fresh CEO insider selling.
Researched 4 days ago
Earnings Aug 3· After close· in 6 weeks
AMD compounds AI data-center compute with ~36% YoY revenue growth and gross margin recovering to ~52.8%, but at $511.59 it trades ~168x trailing EPS and ~150x EV/EBITDA after a +332% year, ~44% above ~$355 fair value, with fresh CEO insider selling.
Estimates disagree on fair value — holding until the picture clears.
Recommendation
Conviction
35/100
solid
Upside
22/100
bull 40% · ~30% odds · +10% expected
Risk-adjusted upside
0/100
+0% after downside pressure
Thesis quality
7.9/10
Opportunity
4.4/10
Risk pressure
8.8/10
Valuation
FairAI fair value
$355.00
Fundamentals check
$456.76
12-24mo fair-value range
$250.00 / $355.00 / $500.00
width 70%
Estimates disagree — look closer.
Buy below
—
Trim above
$600.00
Implied expectations
demanding48.2% implied revenue CAGR
Today's P/S implies ~48.2% revenue CAGR for five years versus 34.4% realized growth.
Agree a market reframing from 'GPU second source' to 'full-stack player' is the most plausible source of further multiple expansion, and EPYC + Instinct support that arc. But the cited +125%/+68% post returns are price moves, not verified fundamentals, and CUDA lock-in still caps the accelerator economics the bull re-rating needs — so this is optionality, not a basis to overpay at ~168x trailing EPS.
Consistent with the X commentary: AMD named as Absolics glass-core substrate qualification customer is a credible, low-significance supply-chain detail that confirms early advanced-packaging adoption without moving the valuation case.
AMD has direct, growing AI-infrastructure exposure via Instinct MI accelerators and EPYC server CPUs, with latest-quarter revenue up ~36% YoY and gross margin recovering to ~52.8%. It is the only scaled accelerator alternative to NVIDIA, but CUDA lock-in caps pricing power and ROCm share. The x86 server franchise is a durable second leg. Forward PEG ~0.95 implies growth is fairly priced at consensus, yet ROIC of 7.68% (diluted by Xilinx goodwill) and a ~332% one-year run leave little margin of safety.
Moat
Narrow moat: x86 share gains are durable but ROCm lacks CUDA's lock-in, capping accelerator defensibility.
Bottleneck fit
Direct AI-compute exposure as the only scaled NVIDIA accelerator alternative, plus EPYC server leg.
Valuation
~168x trailing EPS, ~150x EV/EBITDA, ~44% above fair value — no margin of safety.
Catalyst
Q2 FY2026 print (early Aug) is a specific MI-accelerator revenue and margin readout.
Why not higher
Premium valuation with negligible margin of safety and a narrow moat capped by CUDA lock-in keep it below HIGH_CONVICTION.
Description
AMD designs x86 server/client CPUs (EPYC, Ryzen), Instinct MI data-center GPU accelerators, and adaptive/embedded silicon (Xilinx FPGAs), manufactured at TSMC.
Value Chain
Fabless compute-platform supplier: component/infrastructure layer selling accelerators and CPUs into hyperscaler and enterprise AI data centers.
Moat
narrowDurable x86 server share gains vs Intel and the only scaled GPU alternative to NVIDIA, but ROCm lacks CUDA's software lock-in, capping defensibility.
Pricing Power
moderateEPYC commands share-driven pricing leverage on Intel, but Instinct must price at a discount to NVIDIA to win sockets, limiting accelerator pricing power.
Customer Concentration
Not disclosed in packet; AMD's accelerator revenue is concentrated among a few hyperscalers, a structural dependency the filings flag qualitatively.
| Metric | Value |
|---|---|
| Revenue growth YoY | 35.80% |
| Gross margin | 52.80% ↑ |
| Operating margin | 14.40% |
| FCF margin | 23.10% |
| Cash position | 12350000000 |
| Net debt / EBITDA | -1.60 |
| Share count change YoY | 1.50% |
| ROIC | 7.68% |
Forward P/E
53.22
Trailing P/E
167.90
PEG
0.95
EV/EBITDA
149.58
P/S
22.20
| Peer | Metric | Value |
|---|---|---|
| NVDA | AI GPU position | Wide-moat CUDA leader; AMD is the credible but structurally discounted #2 |
| INTC | x86 server CPU share | AMD is the EPYC share-gainer amid Intel manufacturing struggles |
| AVGO | Custom AI silicon | Competes via hyperscaler XPU ASICs, not a direct GPU rival |
| MU | HBM supply | Supplies HBM to Instinct accelerators; complementary, not competing |
Bull fair value
$760.00
Probability
30%
Horizon
Long term
YTD
—
1Y
331.70%
Vs sector
—
From 52w high
-6.38%
| Risk | Severity | Explanation |
|---|---|---|
| Valuation compression after +332% one-year run | high | At ~168x trailing EPS and ~150x EV/EBITDA TTM, any growth disappointment or estimate cut de-rates the multiple sharply with negligible margin of safety. |
| CUDA software moat caps accelerator economics | high | ROCm lacks NVIDIA's ecosystem lock-in, forcing Instinct to win on price and limiting the pricing power the price-implied growth requires. |
| CEO and executive insider selling | medium | Lisa Su sold ~55k shares around $470-476 on 2026-06-10/12 per Form 4; sustained insider selling into strength is a governance/valuation caution. |
| Hyperscaler customer concentration | medium | Accelerator demand is concentrated among a few hyperscalers; a capex pause or supplier shift would hit revenue disproportionately. |
| Independent financial-statement cross-check unavailable | low | FMP income, balance-sheet, cash-flow, and ratios endpoints returned 403; metrics rely on Finnhub without a second provider to verify. |
1%–2%
CONSIDER on a narrow-but-widening-moat franchise with genuine AI exposure and forward PEG ~0.95, but band held at 1-2% given the ~44% premium to ~$355 fair value, fresh CEO insider selling, and beta ~2.52. Add size only on a meaningful pullback toward $250-$355.
The lone positive valuation anchor — forward PEG ~0.95 / forward P/E 53.2 framed as 'growth fairly priced' — is a mirage built on an unverified consensus forward EPS of ~$9.6, i.e. a ~3.1x jump from trailing EPS $3.0476 in 12-18 months. Every other multiple screams expensive: 168x trailing EPS, ~150x EV/EBITDA TTM, 22x sales, after +332% in a year (beta 2.52). The 'smoothly expanding' margin story is contradicted by the packet's own quarterly series: Q2 2025 (2025-06-28) gross margin collapsed to 39.8% with a NEGATIVE operating margin of -1.74%. Earnings are cyclical and lumpy, not a clean ramp. If forward EPS is even $6 not $9.6, forward P/E is ~85 and PEG >1.5, and the entire 'fairly priced' prop — which the report leans on for sizing and re-rating — disintegrates. You are paying a 44% premium to the model's own fair value for an unconfirmable estimate.
| Load-bearing assumption | Why it might be wrong | Severity |
|---|---|---|
| Forward PEG ~0.95 means growth is fairly priced at consensus. | PEG/forward-EPS is unverified — FMP income/cash-flow endpoints 403'd, so the implied ~$9.6 forward EPS (3.1x trailing $3.05) has no second-source cross-check. A modest EPS haircut flips PEG above 1.5; the report's own falsifier admits this. | high |
| Gross margin is 'expanding' toward ~52.8%. | The packet's quarterly gross-margin series is 50.2→39.8→51.7→54.3→52.8%; Q2 2025 also posted a -1.74% operating margin. The 52.8% headline cherry-picks a recent quarter from a volatile, charge-prone series, not a durable trend. | high |
| Moat trend is 'widening' / market will re-rate AMD as a full-stack player. | This rests on a Substack DD whose 'evidence' is +125%/+68% price moves, which the report itself concedes are 'not verified fundamentals.' Narrative-driven re-rating is optionality, not a basis for the widening-moat upgrade. | medium |
| ROIC 7.68% is acceptable once you discount Xilinx goodwill. | Goodwill is real capital paid; excluding it flatters returns. A 7.68% ROIC sits barely above cost of capital with the 10Y at 4.45%, undermining the 'quality compounder' framing at a 168x multiple. | medium |
| Instinct sustains NVIDIA-like accelerator economics to justify the curve. | Report concedes ROCm lacks CUDA lock-in and Instinct must price at a discount to win sockets — directly contradicting the >50% sustained EPS growth the price embeds. Pricing power score of 5 cannot support the implied curve. | high |
The $355 fair value is 37x an unverified consensus forward EPS of ~$9.6, and the bear $250 is 26x the SAME unverified figure — so the entire range pivots on one unconfirmable number (3.1x trailing EPS) that FMP 403's prevented cross-checking. There is no independent DCF or earnings build; '44% above fair value' merely asserts the multiple should be 37x rather than the market's 53x — a multiple opinion, not a fundamentals-anchored valuation. Because forward EPS itself embeds the heroic growth in dispute, haircutting it pushes true fair value well below $355, and the bull $760 (52x) is simply momentum extrapolation.
Independent red-team pass · claude-opus-4-8 · 2026-06-15
Disputes the price-implied >50% sustained EPS growth: CUDA software lock-in caps Instinct pricing power and ROCm share, so AMD's accelerator economics normalize below the embedded curve, compressing the multiple.
Pre-committed, dated checks that would disconfirm the thesis — a review is flagged automatically as each date passes.
| If we observe… | By | …the thesis is wrong because |
|---|---|---|
| Q2 FY2026 print shows MI/data-center GPU revenue growth decelerating below ~30% YoY with no ROCm share gain | 2026-08-12 | Accelerator growth thesis weakening — downgrade to WATCH and re-test the price-implied growth curve. |
| Gross margin falls below 50% for two consecutive quarters | 2026-11-15 | Margin-expansion leg of the thesis broken — re-rate fair value lower. |
| Consensus forward EPS is cut such that forward PEG rises above 1.5 | 2026-08-12 | Growth no longer fairly priced — premium becomes unjustified, move toward PASS. |
| Claim | Source | URL | Retrieved |
|---|---|---|---|
| Price $511.59, +4.74%, market cap ~$834B | finnhub:quote | — | retrieved 2026-06-15T13:29:14.953Z |
| Forward P/E 53.22, forward PEG 0.948, EV/EBITDA TTM 149.58, EV/Revenue TTM 22.21 | finnhub:basic-financials | — | retrieved 2026-06-15T13:29:14.784Z |
| TTM EPS $3.0476; latest-quarter gross margin 52.82%; operating margin 14.4%; FCF/share TTM $5.259; ROIC TTM 7.68% | finnhub:basic-financials | — | retrieved 2026-06-15T13:29:14.784Z |
| 52-week high $546.44 (2026-06-03), low $115.06; 52-week return +331.7% | finnhub:basic-financials | — | retrieved 2026-06-15T13:29:14.784Z |
| Net cash position: cash/share $7.57 (~$12.35B), net-debt-to-equity -0.037 | finnhub:basic-financials | — | retrieved 2026-06-15T13:29:14.784Z |
| Revenue/share TTM $22.80 vs ~$16.99 prior; FY2025 sales/share $21.17 vs $15.75 (~34% growth) | finnhub:basic-financials | — | retrieved 2026-06-15T13:29:14.784Z |
| Analyst ratings 2026-06: 14 strong buy, 32 buy, 11 hold, 0 sell | finnhub:recommendations | — | as of 2026-06-01 · retrieved 2026-06-15T13:29:14.761Z |
| Lisa Su sold ~55,666 shares 2026-06-10/12 at ~$470-476 | finnhub:insider-transactions | — | retrieved 2026-06-15T13:29:14.764Z |
| Peers: NVDA, AVGO, MU, INTC, TXN, MRVL, QCOM | finnhub:peers | — | retrieved 2026-06-15T13:29:14.762Z |
| 10-Q filed 2026-05-06; 8-K filed 2026-05-15 | sec:recent-filings | Link | as of 2026-05-15 · retrieved 2026-06-15T13:29:14.798Z |
| AMD named glass-core substrate qualification customer at Absolics Georgia fab (H2 2026) | x:@aleabitoreddit | Link | retrieved 2026-06-15T13:29:14.549Z |
| Photonics laser standardization for CPO/pluggables/SiPh relevant to AMD accelerator interconnect | x:@aleabitoreddit | Link | retrieved 2026-06-15T13:29:14.549Z |
| Analyst thesis: market re-rates AMD as full-stack player vs GPU second source | substack:damnang | Link | retrieved 2026-06-15T13:29:14.548Z |
| AMD as qualification customer in glass-substrate supply chain | substack:damnang | Link | retrieved 2026-06-15T13:29:14.548Z |
| Macro: 10Y 4.45%, Fed funds 3.62% (2026-06-11) | fred:macro-snapshot | — | as of 2026-06-11 · retrieved 2026-06-15T13:15:57.661Z |
The market misreads AMD as the runner-up that can't beat NVIDIA, which blocks real conviction. The author argues that once the market sees AMD as a full-stack player rather than a GPU second source, the multiple expands again.
The author frames AMD's purchase of MEXT as evidence that leading chip buyers have stopped waiting for memory prices to fall and are instead building 'detour' paths that use less DRAM. Buying a software prediction layer rather than a chip or memory part signals where the memory market is heading over the next few years.