Goldman Sachs Upgrades AMD to Buy at $450

Overview

Goldman Sachs upgrades AMD from Neutral to Buy, hiking the 12-month price target from $240 to $450, implying 27% upside. The core thesis: 1) Agentic AI (autonomous AI agents) will rapidly penetrate enterprises, expanding the server CPU market, with x86 dominant and AMD gaining share; 2) Meta’s 6GW GPU deployment and potential orders from Microsoft and OpenAI create significant upside for data center GPUs beyond 2027.

Core views

Demand: Agentic AI inference scenarios explode, with CPUs participating in 'thinking, routing, tool calling,' driving higher CPU counts per server. By 2030, Agentic AI tokens are expected to account for 90% of total tokens, accelerating server CPU TAM over the next 3–5 years. Supply: AMD gains server CPU share; x86 dominates enterprise Agentic AI due to IT compatibility; data center GPU shipments forecasted to grow +66%/+98% YoY in 2026/27, with AI racks driving CPU co-sales. Revenue: Data Center revenue expected at $6.4B/$8.4B in 2027/28, with Server CPU at $21.1B/$27.4B (24%/38% above consensus) and Data Center GPU at $42.5B/$56.3B (28%/26% above). Traditional segments fall below 25% by 2027, with most cyclical risks priced in. Margins: Higher server CPU margins and operating leverage lift non-GAAP EPS to $13.50/$18.00 in 2027/28, 20% above consensus.

Analysis framework

The firm uses a dual-method validation: top-down demand modeling and bottom-up customer order decomposition. 1) Token economy model estimates total AI inference demand, then maps CPU/GPU ratios to derive server CPU growth. 2) Quantifies GW-level deployments from Meta, Microsoft, OpenAI to project GPU revenue scenarios. 3) Targets $450, implying 30x 2027 normalized EPS ($15), up from prior 28x and $8.50 EPS.

Methodology notes

• Industry/Market Analysis FrameworkSupply-demand framework

  Token Demand Modeling

  Estimates AI inference token volume and backcasts server CPU demand using CPU/GPU ratios, a common method for sizing AI hardware markets.

• Valuation MethodPE Valuation

  30x 2027 Normalized EPS

  Applies a reasonable P/E to next-year earnings to derive the target; the forward-looking P/E is moved to 2027 due to clear earnings visibility.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• AMD (US.AMD)

  Upgraded to Buy; benefits from Agentic AI-driven server CPU expansion and Meta’s 6GW GPU order

  Strengths

  Growing x86 server CPU share; Meta’s large order secures GPU revenue beyond 2027; Data Center high-margin segment to exceed 80%

  Weaknesses

  PC and gaming segments remain in cyclical decline; GPU ecosystem lags NVIDIA

  Comparison

  Relative to NVDA, AVGO, AMD’s valuation reflects growth but offers greater upside on EPS revisions

  Risks

  Slower Agentic AI adoption; Meta GPU deployment delays; ARM displaces x86 in enterprise AI

Key data

• 2027 Server CPU Revenue Estimate$21.1B24% above consensus
• 2027 Data Center GPU Revenue Estimate$42.5B28% above consensus
• 2027/28 Non-GAAP EPS$13.50 / $18.0020% above consensus
• Meta 6GW GPU Deployment Revenue$25.0B in 2027Assuming $20k per GW
• Target Implied 2027 P/E30xUp from prior 28x

Impact & implications

The firm views Agentic AI as the second wave of AI infrastructure spending after training, repricing server CPU demand. AMD benefits from x86 ecosystem advantages and GPU order wins, offering over 20% upside for both earnings and valuation.

Risks

• Slower-than-expected Agentic AI adoption
• Delays in Meta’s 6GW GPU deployment
• x86 share erosion in enterprise AI servers to ARM architectures
• Higher-than-expected operating expenses, compressing operating leverage

What to watch

• Meta 6GW GPU deployment milestones
• OpenAI follow-on GPU order execution
• Enterprise Agentic AI penetration data