AMD Advancing AI event reinforces the dual-track AI growth narrative for CPUs and GPUs

Overview

This report summarizes the key content from AMD's Advancing AI event held on July 22-23, 2026. Bernstein views the event as overall positive, with key highlights including the launch of Helios (MI450), updates to the rack-scale GPU roadmap, customer partnerships with Anthropic and Microsoft, upward revisions to CPU and accelerator TAM, the Venice CPU roadmap, the AMD-Cerebras inference partnership, and the expansion of ROCm.AI and enterprise/personal/physical AI products.

Core views

The core view is that although market expectations are already high, AMD's exposure to AI demand is simultaneously driving both its CPU and GPU growth stories. The report remains positive on the company's fundamentals, seeing upside in both CPU and GPU trajectories, and maintains its Outperform rating and USD 600 price target on AMD.

Analysis framework

The report combines event commentary with product roadmap analysis, assessing AMD's medium- to long-term growth potential through customer partnerships, TAM outlook, GPU/CPU product cadence, software ecosystem, and application expansion, supplemented by Bernstein's financial forecasts and valuation tables.

Methodology notes

• Event-driven researchReview of key takeaways from the company's AI event

  Assess changes in fundamentals through product launches, customer partnerships, and management TAM guidance.

  The report focuses on extracting new information from the Advancing AI event and comparing it with the prior roadmap and competitive landscape to evaluate its impact on AMD's investment narrative.

• Market size analysisAssessment of TAM revisions

  Use management's 2030 TAM figures for CPUs, accelerators, and total compute to assess long-term revenue potential.

  AMD raised its 2030 CPU TAM to USD 220B and expects accelerator TAM to reach USD 1.4T; the report views this as growth support from AI inference and agentic AI demand.

Asset mapping & comparison

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

• AMD

  Core covered name

  Strengths

  Dual beneficiaries of AI accelerators and server CPUs; Helios has received endorsement from important customers such as OpenAI, Meta, and Anthropic; 2030 TAM has been significantly raised; ROCm.AI strengthens the software ecosystem narrative.

  Weaknesses

  The report notes that performance benchmarks should be viewed with caution; technical details of some partnerships remain limited; market expectations are already high.

  Comparison

  The report compares Helios performance with NVIDIA Vera Rubin and MI350/MI355, and notes that the AMD-Cerebras partnership is similar in direction to NVIDIA and Groq, though the degree of integration may differ.

  Risks

  Product ramps may fall short of expectations, customer deployments may be delayed, the CUDA ecosystem moat may persist, and inference demand or TAM realization may come in below management expectations.

• NVIDIA

  Primary competitive benchmark

  Strengths

  Strong CUDA software ecosystem moat, and its low-latency inference solutions may be more integrated.

  Weaknesses

  AMD is attempting to narrow the ecosystem and inference performance gap through ROCm.AI, Helios, and its Cerebras partnership.

  Comparison

  AMD views ROCm.AI as one answer to NVIDIA's CUDA moat, and Helios benchmarks are also compared with Vera Rubin.

  Risks

  If AMD delivers on its roadmap, NVIDIA could face stronger competition in certain AI inference and rack-scale system opportunities.

Key data

• RatingOutperformBernstein's equity rating on AMD.
• Price targetUSD 600.00Both the table and the main text show AMD's price target at USD 600.
• Current priceUSD 539.69The ticker table uses the closing price on July 23, 2026.
• 2030 CPU TAMUSD 220BRaised from USD 120B about three months earlier.
• 2030 accelerator TAMUSD 1.4TThe report says this grows from USD 200B in 2025 at about 45% CAGR.
• 2030 total compute TAMApproximately USD 2TThe report says this grows from USD 365B in 2025 at about 40% CAGR.
• Anthropic partnershipUp to 2GW of compute deployment, with AMD investing up to USD 5BThe first 1GW is planned to begin deployment in 1H27, with a target of deploying close to a full 1GW in 2027.
• 2027E revenue forecastUSD 82.189BBernstein AMD Income Statement table.
• 2027E adjusted EPSUSD 14.61Adjusted EPS 2027E as shown in the ticker table.

Impact & implications

The implication of the report is that AMD's AI growth no longer depends only on a single GPU product, but instead on a more complete growth path spanning rack-scale GPUs, CPU host nodes, inference partnerships, software ecosystem development, and enterprise/personal AI use cases. If Helios customer deployments, the MI455X ramp, ROCm.AI ecosystem improvement, and CPU share gains materialize, AMD's upside in revenue and margins could expand further.

Risks

• AI demand or inference spending may grow more slowly than management's TAM assumptions.
• Product launches or ramps for Helios, MI455X, MI500/MI600, and others may be delayed.
• ROCm.AI may fail to effectively narrow the gap with NVIDIA's CUDA ecosystem.
• Deployment scale and timing from Anthropic, Microsoft, or other customers may come in below expectations.
• Performance benchmarks may lack comparability, and real-world workload performance may be weaker than what was shown at the event.

What to watch

• Actual progress of MI455X starting in 3Q and ramping in 4Q.
• Milestones for deployment of Anthropic's first 1GW of compute capacity beginning in 1H27.
• Specific scale and timeline for Microsoft Azure's large-scale deployment of Helios.
• Developer adoption of ROCm.AI for optimizing training and inference workloads.
• Market share changes for different Venice CPU versions across GPU servers, agentic CPU servers, and general-purpose servers.
• Whether the 2030 CPU TAM and accelerator TAM assumptions continue to be validated by customer demand.