Pre-meeting discussions with AMD indicate that Agentic AI is boosting the server CPU opportunity, while MI450/Helios ramp risks continue to decline

Overview

This report from JPMorgan summarizes key takeaways from discussions with AMD IR and sell-side participants ahead of AMD's 2Q26 earnings quiet period. The report believes Agentic AI is expanding server CPU demand from traditional general-purpose computing into incremental workloads such as AI orchestration, data movement, and tool calling, thereby enlarging the opportunity for AMD EPYC; at the same time, the ramp cadence and customer validation progress for MI450/Helios are improving, and visibility for data center AI revenue extends into CY28. However, the report still maintains a Neutral rating, mainly because the stock price is close to fully valued and long-term share, competition, and execution risks still need to be monitored.

Core views

The core views include: first, server CPU TAM expansion increasingly looks like a unit-driven story, with AMD expecting server CPU TAM to exceed $120B by 2030, and roughly two-thirds or even 70%+ of the growth coming from unit volume; second, the path from initial MI450/Helios volume in 3Q to a meaningful ramp in 4Q continues to be de-risked, with full racks already running in customer labs; third, CY27 server CPU and data center GPU revenue may grow nearly in parallel, better than the market's assumption of a single GPU-driven story; fourth, near-term MI450 gross margin dilution may be partly offset by the EPYC mix, ASP uplift, scale effects, and Embedded's high margins; fifth, improved ROCm usability lowers the barrier to customer adoption and is key for AMD to convert a single-generation product opportunity into multi-year share gains.

Analysis framework

The report is mainly based on company discussions ahead of the quiet period and breaks down scenarios across AMD's data center CPU, data center GPU, supply, gross margin, Client/Gaming, and Embedded business lines; valuation uses a P/E multiple approach, deriving the Dec-26 target price of $385 by applying roughly 35x P/E to about $11 of CY26 exit earnings power.

Methodology notes

• Valuation methodsP/E multiple valuation

  Roughly 35x P/E multiplied by about $11 of CY26 exit earnings power

  The $385 target price is based on an approximately 35x P/E multiple, broadly in line with peers, applied to AMD's roughly $11 of CY26 exit earnings power.

• Demand analysisTAM volume-price decomposition

  Server CPU TAM exceeds $120B by 2030, with growth mainly driven by unit volume

  AMD believes the server CPU TAM forecast is likely conservative, with roughly two-thirds or even 70%+ of growth coming from shipments, and the remainder from ASP and mix improvement driven by higher core counts.

• AI infrastructure trackingMI450/Helios ramp milestones

  Initial volume in 3Q, meaningful ramp in 4Q, continued improvement in 1Q

  Full Helios racks are already running in customer labs, helping compress system integration cycles and reducing the validation uncertainty that previously limited adoption of AMD accelerators.

• Operating qualityGross margin bridge

  Near-term MI450 dilution offset by EPYC, Embedded, and scale effects

  The report believes initial MI450/Helios gross margins are below the company average, but improvements in EPYC mix and ASP, the Client/Gaming mix, Embedded contribution, and scale effects should support the long-term 55%-58% gross margin range.

Asset mapping & comparison

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

• AMD.O / US.AMD stock

  Research target

  Strengths

  Data center CPU and GPU opportunities are improving simultaneously, while progress in EPYC, MI450/Helios, and the ROCm ecosystem enhances mid-term growth potential.

  Weaknesses

  The current stock price is above the target price, and the report believes valuation is close to full; long-term share gains remain uncertain.

  Comparison

  Competes with Intel for server CPU share and with NVIDIA for data center GPU and AI infrastructure opportunities.

  Risks

  PC cycles, intensifying competition, R&D spending pressure, MI450 gross margin dilution, customer-side data center readiness, and supply coordination risks.

• EPYC server CPU

  One of the core growth drivers for AMD's data center business

  Strengths

  Agentic AI brings incremental CPU workloads, the CPU:GPU ratio shifts toward roughly 1:1, and 6th Gen EPYC Venice and Verano reinforce AMD's positioning in AI infrastructure.

  Weaknesses

  ASP improvement comes more from higher core counts and cost pass-through than from pure pricing power.

  Comparison

  Versus ARM-based solutions, AMD says per-socket throughput exceeds leading ARM AI solutions by 2x; versus x86 peers, it offers better performance per watt.

  Risks

  Demand sustainability, TSMC costs, customer deployment cadence, and competitor responses.

• MI450/Helios and Instinct GPU

  Carrier of AMD's data center AI revenue ramp

  Strengths

  The ramp path from 3Q to 1Q is clearer, full racks are already running in customer labs, and customer demand visibility extends into CY28.

  Weaknesses

  Initial gross margins are below the company average, and system ramp-up requires coordination between customers and ODMs.

  Comparison

  The report notes AMD intends to focus first on ramping with two core ODMs to avoid the ramp issues seen when NVDA rolled out across multiple ODMs simultaneously.

  Risks

  Customer data center shells, power availability, capital commitments, HBM, and backend capacity may all affect deliveries.

• Embedded / Xilinx-derived FPGA and SoC

  Gross margin support and auxiliary exposure to AI infrastructure

  Strengths

  This high-margin business can provide a gross margin buffer during the MI450 dilution phase and benefits from AI infrastructure use cases such as networking, storage, and edge inference.

  Weaknesses

  Its short-term contribution to total segment revenue is limited and it is not yet a primary growth variable.

  Comparison

  Compared with the cyclicality of Client and Gaming, Embedded is more high-margin and infrastructure-related.

  Risks

  The pace of conversion of AI infrastructure use cases and insufficient segment scale may limit short-term impact.

Key data

• RatingNeutralJPMorgan maintains a Neutral rating.
• Target price$385.00Dec-26 target price.
• Current price$511.57As of 2026-06-12.
• Implied returnapproximately -24.7%Calculated based on the target price relative to the current price.
• Server CPU TAM>$120B by 2030AMD believes this point estimate is likely conservative.
• Source of server CPU TAM growthroughly two-thirds to 70%+ from unit volumeThe remaining growth comes from ASP and mix improvement driven by higher core counts.
• CPU:GPU ratio changeshifting from 1:4 or 1:8 toward roughly 1:1After the expansion of Agentic AI, CPU demand may carry greater weight.
• MI450/Helios cadenceinitial volume in 3Q, meaningful ramp in 4Q, continued improvement in 1QFull Helios racks are already running in customer labs.
• Data center AI revenue frameworkannualized tens of billions of dollars in 2027Aggregate customer forecasts already exceed the original 2027 plan, with the largest deployments skewed toward inference.
• Revenue per GW framework$15B-$20BThe report believes this framework may have upside bias over the long term.
• Long-term gross margin range55%-58%Despite initial MI450 dilution, management's framework remains unchanged.
• Gaming outlookdown more than 20% HoH in 2HConsistent with the console cycle and component cost pressure.

Impact & implications

In terms of investment implications, the report expands AMD's data center opportunity from a pure GPU substitution narrative into a multi-year opportunity jointly driven by CPU, GPU, and the software ecosystem. If Agentic AI continues to raise CPU attach rates, EPYC could become a more important source of revenue and gross margin support than the market expects; if MI450/Helios ramps as planned, together with lower ROCm adoption barriers, AMD's serviceable customer base in AI accelerators will broaden. However, the target price is significantly below the current price, indicating that JPMorgan believes much of these improvements has already been reflected in the stock price, and more positive ratings will require validation through orders, deliveries, gross margin, and CY27/CY28 visibility.

Risks

• A weaker-than-expected PC end market could reduce microprocessor and GPU shipments, leading to lower revenue and EPS estimates.
• AMD competes with Intel in the microprocessor market and with NVIDIA in graphics and AI accelerators, and share gains or losses could alter the rating view.
• Long-term share gains remain uncertain, and AMD needs to continue increasing operating investment, especially in R&D, to catch up with market leaders.
• Initial MI450/Helios gross margins are below the company average and may dilute corporate gross margin in the short term.
• Industry tightness remains in wafers, backend/OSAT, memory, and HBM, although AMD says it has secured these relatively well.
• The bottleneck in the Helios ramp may shift to the customer side, including data center shells, power supply, and capital commitments.
• Client and Gaming face pressure in 2H, with Gaming expected to decline more than 20% sequentially.
• If ROCm ecosystem improvement is insufficient, MI450/Helios may struggle to convert a single product-cycle win into sustained share gains.

What to watch

• AMD's 2Q26 results and management updates on data center CPU, MI450/Helios, and gross margin.
• Whether MI450/Helios delivers on schedule with initial volume in 3Q, a meaningful ramp in 4Q, and continued improvement in 1Q.
• Whether CY27 and CY28 customer forecasts, orders, and multi-GW projects continue to expand.
• Whether Agentic AI deployment continues to raise CPU:GPU attach rates and validates upside to server CPU TAM estimates.
• Customer adoption of high-core-count EPYC products, Venice, and Verano, as well as ASP/mix changes.
• Whether ROCm enables more customers to deploy independently and reduces reliance on AMD's field support teams.
• Whether HBM, TSMC wafer, OSAT, and memory costs are passed through smoothly into ASP.
• Whether second-half pressure in Client and Gaming can be offset by wafer reallocation to EPYC and Instinct.