Phase Space AI

Valuation

Advanced Micro Devices [AMD]

Advanced Micro Devices, Inc. [AMD] — Valuation

Spot $454.54 (2026-07-28) · two outputs, both required · references/valuation.md

---

Verified valuation inputs — and why the share count decides everything

The fully diluted column is the one used in the implied-path test, and the justification is economic rather than accounting. The 320m warrant shares (OpenAI 160m, October 2025; Meta 160m, February 2026) are struck at $0.01 and vest on "AMD Instinct GPU purchase milestones … and achievement of specified Company stock price targets." They vest if and only if the outcome that justifies today's price occurs. Valuing the outcome on 1,630.6m shares values the revenue without the dilution that buys it. GAAP excludes them from today's diluted EPS — correctly, since none has vested — but a five-year valuation cannot.

Not captured in either column: $4.1bn of maximum gross exposure from guarantees of commercial partners' data-centre lease obligations, term up to 15 years, sometimes issued "in exchange for warrants."

---

Output 1 — the 12-month target

Own EV/Sales history, 2022-02-03 → 2026-07-28 (1,123 trading days)

Today's multiple sits at the 95.9th percentile of AMD's own four-and-a-half-year history. Only 25.03x has ever been higher, and 19.55x is 2.3x the median.

Construction as for the cluster: price from Alpaca daily bars against point-in-time TTM revenue lagged to filing dates; current share count and net cash held constant. EV/EBIT is declared UNIDENTIFIED as an anchor — a continuous TTM EBIT series only exists back to the Q1 FY25 filing (120 trading days), far too short for a percentile, and valuation.md forbids substituting a peer median.

NTM revenue

Uniform cluster method: next quarter at company guidance, then +5% sequential for three quarters.

Target

12-month target: $325 on as-reported shares (−28.5%), $272 fully diluted (−40.2%). Base case is the own-history 75th percentile, the same rule applied to all three cluster names.

Bull: $584 / $489 (+28.5% / +7.5%) — requires the multiple to hold at the 95.9th percentile for a further twelve months. Bear: $260 / $217 (−42.8% / −52.3%) at the historical median.

This target is below spot, and the reason is not a house view about the market

Framework item B16 records 16 of 16 house targets below spot at a median 46.1% below Street, and warns that "a process whose every target is below spot is expressing a house view about the market, not valuing companies." AVGO in this same cluster, on the identical method, produces +58.5%. So the method is not biased downward.

AMD's target is below spot because its own multiple is at the 95.9th percentile of its own history while its revenue grows 28.9%. Revenue growth of 28.9% against a required multiple compression of 45% (19.55x to 10.79x) nets to −28.5%. To produce a target above spot the multiple must stay at a level it has reached only once in four and a half years.

The honest caveat, stated rather than buried: AMD's own multiple history is dominated by pre-AI years. Between 2022 and 2024 AMD grew 6–14% and traded at 6–11x sales. It now grows 38% with a 56%-of-revenue Data Center segment. A median drawn from the slow years is not a fair mean-reversion target for the fast ones, and that argument is why the bull case is presented at the current multiple rather than dismissed. What the argument cannot do is make 19.55x cheap: it is the highest EV/Sales AMD has sustained, on the highest revenue base it has ever had.

External sanity band. No professional target for AMD is on file in this project, so no gap is reported.

---

Output 2 — the implied-path test (the Valuation Criteria)

Parameters

On the terminal margin. 18% is deliberately generous, not conservative. AMD's TTM GAAP operating margin is 11.7%; its latest quarter 14.4%; FY2025 was 10.7% (11.9% ex the $440m MI308 charge). 18% assumes the $551m/quarter of Xilinx intangible amortisation largely rolls off and that operating leverage delivers roughly 400bp beyond the current quarter. The screen used AMD's own 10.7% — arithmetically defensible but stale. Using 18% reduces the required CAGR, so the FAIL below is not an artefact of a harsh margin assumption.

On the exit multiple. 21.2x EBIT at an 18% terminal margin is 3.8x EV/Sales — below AMD's all-time-minimum EV/Sales of 3.79x by a hair. That is a stringent exit, so the sensitivity table below does the real work.

Result

Required revenue CAGR: 58.1% for five years.

Implied FY2031 revenue: $370.3bn (from $37.5bn — a 9.9x increase).

Implied multiple compression: today's 201.1x EV/EBIT to a 21.2x exit = −89.5%.

Sensitivity — over the exit multiple

Sensitivity over the terminal margin, at the 21.2x anchor: 14% → 66.3% required; 18% → 58.1%; 22.5% → 51.2%.

There is no cell in this table that produces a PASS. At twice the growth-matched exit anchor — a 42.4x EBIT exit, or 7.6x EV/Sales, which is inside AMD's historical range — the required CAGR is 37.7%, exactly equal to the single best quarter AMD has printed in three years, sustained for five consecutive years. At the anchor it is 58.1%.

Terminal value is 100% of modelled EV, far above the 60% threshold, so this table is the analysis and the reverse DCF is the primary long-horizon output.

Verdict: FAIL

criteria.md allows PASS WITH ARGUMENT where "the price requires more than demonstrated and there is a specific, evidenced reason (a named product cycle, mix shift, pricing action)." AMD has an unusually specific named product cycle — MI450, MI455X, Helios, HBM4 from Samsung, HPE as a platform partner, OpenAI and Meta as named counterparties. On the mechanism test it scores well.

It still fails, for three reasons that the argument cannot reach.

1. Magnitude, not direction. PASS WITH ARGUMENT bridges a gap of the size AVGO has (−6.1pp against trailing, +15.4pp against run-rate). AMD's gap is −42.8pp against trailing and −20.3pp against its best-ever quarter. A named product cycle can justify exceeding a trailing average; it cannot justify requiring nearly 10x revenue growth in five years. $370bn of FY2031 revenue would exceed most estimates of the entire AI accelerator market.
2. The commitment is absent. RPO is $264m — 0.7% of TTM revenue. The 10-K states customers "may cancel orders for standard products more than 30 days prior to shipment without incurring significant fees." "Up to 6 gigawatts" from OpenAI and "up to 6 gigawatts" from Meta are announcements, not obligations, and neither appears in RPO. The contrast with Broadcom's $164.6bn of firmly committed, non-cancellable RPO is the difference between the two verdicts.
3. Success is self-diluting. The 320m warrant shares vest on Instinct purchase milestones. The more of the required path AMD delivers, the more of the equity belongs to the customers who bought it. The upside is capped by construction, which is why the fully diluted EV is the correct denominator and why the as-reported figure (54.9% required) is barely better.

Ranking input for the strategy: margin = −42.8pp against the corrected trailing CAGR, −20.3pp against the latest quarter. Second of three in the cluster on both measures.

A note on what a FAIL does not say. It does not say AMD's business is bad — the Quality Criteria PASS and the concentration profile is the best in the cluster. It says the price already contains an outcome larger than the AI accelerator market is currently expected to be. Per criteria.md, "Cheap cannot rescue a failure" on Quality; the converse also holds — a good business cannot rescue a failure on Valuation.

---

Cluster note

See AVGO_Valuation.md for the full statement. In summary: one shared growth-matched exit anchor is justified for this cluster — TTM YoY growth is AVGO +32.3%, AMD +35.0%, MRVL +34.1%, a 2.7pp band, across three fabless, asset-light, design-win-driven, customer-concentrated names in SIC 3674. A shared terminal-margin ceiling is not, and applying one (15.6% to AVGO and MRVL) is what made all three Tier-1 verdicts INDETERMINATE. AMD's own terminal margin was correctly left uncapped by the screen, which makes the inconsistency in the cap logic visible.

---

Sources

AMD 10-Q (2026-03-28, filed 2026-05-06); 10-K (2025-12-27, filed 2026-02-04); 10-Q (2025-09-27, filed 2025-11-05); earnings releases 8-K EX-99.1 2026-05-05, 2026-02-03, 2025-11-04, 2025-08-05, 2025-05-06, 2025-02-04, 2024-10-29, 2024-07-30; SEC XBRL companyfacts CIK 0000002488; Alpaca daily bars and option snapshots 2026-07-28/29; reports/scan_v3/_scan.json.