Goldman Sachs Communacopia + Tech Conf 2025: AMD DC & AI Infra Strategy/Outlook

Key Takeaways

TL;DR: AMD reiterates bullish LT AI infra outlook, citing rising demand, improving SW ecosystem, and deepening hyperscaler collabs. Expects major rev. uplift from MI450 in 2026, targets >20% DC GPU share, and sees CPU/AI synergies driving TAM and AMD share higher. Key risks: execution, customer enablement, SW maturity for scale-out AI, but AMD claims significant de-risking at silicon/rack-system levels.

1. AI, Mkt Dynamics, & Addressable Opp

• GenAI & AI infra “early innings” but “super positive” for biz value; AMD seeing “substantial productivity improvements” via AI integration.
• AI monetization/traction accelerating in enterprise, not just consumer. Biggest excitement: productivity gains in engineering/dev processes.
• Bottlenecks: Networking, distributed systems/SW, and power/DC availability now “dominant factors in system perf. going forward.”
• AI HW TAM: Reiterated $500B 2028 AI accelerator TAM (set >1yr ago, now credible given hyperscaler capex). “Growth modulated more by DC/power availability than anything else.”
  • Quote: “We see enough evidence that biz value is there, that we're pretty optimistic this will continue to grow at a rapid pace.”

2. AMD DC GPU Roadmap, MI450, & Competitive Landscape

• GPU sales: Ramp from “very little” to ~$7B this yr.
• Roadmap: Multi-gen—MI300/MI325 (inference, chiplet/mem advantage), MI355 (training), MI450 (2026, AMD’s ‘Milan moment’ for GPUs).
  • MI450: Targeted as “best training, inference, RL solution.” Timed w/ NVDA Rubin to ensure “no excuses” on perf/SW stack for majors.
  • Execution: AMD credits “systematic, thoughtful approach” for progress, acknowledges starting behind NVDA.
  • Customer traction: Positive feedback/public endorsements from OpenAI, Oracle, MSFT, Meta.
    • “Deeply engaged w/ end customers… ramp [MI450] efficiently into prod.”
    • ZT Systems/Helios Rack: 2+ yrs system-level investment for MI450 rack/cluster delivery, focus on DC compatibility, ease of deployment (air/liquid cooling ratios matched to NVDA, larger/easier racks).
  • De-risking: “Very rigorous de-risking plan” (system, mechanical, SI, thermal, etc.).
• Mkt Share Ambitions: “If you’re not at least 20% share, you’re not meaningful. We aspire to be a meaningful player as an intermediate step.”
• Risks: Near-term: SW stack/system-level validation for top ~~20 customers (~~80% of capex), not full “long tail” (vs. NVDA’s approach).

3. SW Ecosystem, Customer Enablement, & Adoption Barriers

• Customer base: Strong CPU relationships w/ all major hyperscalers; growing MI (GPU) engagements (MSFT, Oracle, Meta, OpenAI).
  • “All major customers… already AMD CPU customers, already have DC engagement.”
• SW maturity: AMD “systematic” in building targeted support for frameworks/libs (e.g., “Jack”); prioritizing high-value, high-volume customers over long tail.