EquitySemiconductors / AIUSLarge Cap
Advanced Micro Devices logo

AMD

Advanced Micro Devices

Live Quote

······

Avg Cost

$225.20

Total Return

···

In Your Sleeves

Roth IRA
6.3%Below target
···

Position Changes

May 2026

Trim series · 4 sessions

Continued trimming AMD through May as the position appreciated above the target sizing band. Multiple sessions of small trims recycled gains while preserving core exposure to the AI compute thesis.

Apr 30, 2026

Partial trim

Trimmed AMD in the Roth Retirement Account on May 1, 2026 after a significant run to bring the position back toward my 10% max position-size discipline. This was not a thesis reversal. The core AMD thesis remains intact, but the trim reflected concentration control, risk management, and a preference to preserve gains after the position had outgrown its intended role.

Aug 21, 2025

Partial trim

Partial trim to bring the position closer to target sizing.

Why I Own It

AMD is the only credible challenger to NVIDIA in the AI GPU market, and its valuation reflects a significant discount to that positioning. I own it as a structural hedge within the AI hardware exposure: if NVIDIA continues to dominate, AMD's EPYC server CPU business and gaming GPU franchise still compound at a reasonable rate. If AI workloads diversify toward AMD's MI series GPUs — which is beginning to happen at enterprises and cloud customers seeking alternatives to CUDA lock-in — the earnings upside is substantial. The Roth IRA is the right home for this position because the GPU challenge thesis plays out over a multi-year horizon, and tax-free compounding on a long-duration investment is exactly what the account is designed for.

Why This Sleeve

AMD is in the Roth IRA because the AI GPU narrative requires a 5-10 year timeframe to fully resolve — either AMD builds a viable alternative to CUDA or it doesn't — and that holding period makes the tax-free compounding of the Roth account significantly more valuable than managing it tactically in a taxable sleeve.

Investment Thesis

AMD's MI300X GPU has emerged as the primary challenger to NVIDIA's H100/H200 in AI training and inference workloads. The company's x86 CPU share gains in data center and the growing adoption of ROCm as a CUDA alternative in enterprise deployments add multiple vectors of upside beyond the core GPU thesis.

AMD's EPYC server CPU business is a durable earnings engine that often gets overlooked in the AI GPU conversation. EPYC processors have taken meaningful share from Intel in cloud data center deployments — now running in Amazon EC2, Google Cloud, and Microsoft Azure instance families — generating consistent, growing revenue that provides financial ballast for AMD's GPU investments. Even in a scenario where AMD's GPU challenge stalls, the CPU trajectory alone supports a reasonable earnings multiple, which means the AI GPU optionality comes largely at no cost relative to current valuations.

Scenario Analysis

Bull Case

AI GPU Share Gains Accelerate

AMD captures 30%+ of the data center AI GPU market as MI300X adoption broadens and hyperscalers dual-source away from NVIDIA.

  • MI300X and next-generation MI400 outperform NVIDIA H100 on key inference benchmarks

  • Major hyperscalers formalize dual-sourcing strategies to reduce NVIDIA dependency

  • ROCm software ecosystem matures, lowering switching costs for large AI labs

  • Data center GPU revenue exceeds $12B by FY2026

Base Case

Steady Market Share Gains

AMD grows data center GPU revenue at 40-50% annually while maintaining CPU share gains, supporting durable mid-teens EPS compounding.

  • MI300X maintains 15-20% of the AI GPU addressable market

  • EPYC CPU gains to 25-30% server market share by 2026

  • Gaming GPU business stabilizes, reducing cyclical earnings drag

  • Operating leverage drives margin expansion toward 30% non-GAAP operating margin

Bear Case

NVIDIA Widens the Software Moat

CUDA's entrenched ecosystem prevents meaningful AMD GPU adoption in AI, limiting AMD to a niche challenger position.

  • NVIDIA's NVLink and CUDA ecosystem create switching costs that AMD cannot overcome

  • AMD GPU market share plateaus below 10% in AI training

  • PC/gaming market weakness extends, pressuring consumer GPU revenue

  • China export restrictions further limit addressable market

Key Risks

  1. 01

    CUDA ecosystem lock-in — the majority of AI training code is optimized for NVIDIA hardware, creating structural switching costs that AMD must overcome.

  2. 02

    Export restrictions on advanced chips to China remove a significant portion of AMD's addressable GPU market.

  3. 03

    NVIDIA's roadmap cadence (Blackwell, Rubin) may maintain a performance lead that neutralizes AMD's competitive positioning.

  4. 04

    Geopolitical tensions could disrupt TSMC manufacturing relationships critical for advanced-node chip production.

What I'm Watching

  • MI300X and MI400 market share estimates and enterprise adoption announcements from cloud providers.

  • ROCm software ecosystem maturity — developer tooling adoption is the key to unlocking CUDA alternatives.

  • EPYC CPU market share trajectory in cloud instance types each quarter.

  • Quarterly data center GPU revenue as a percentage of total — the most direct thesis progress metric.

  • NVIDIA Blackwell/Rubin performance benchmarks that set the competitive bar AMD must meet.