Price target adjustment: U.S. IT hardware - Server OEM: Agent AI drives the next phase of AI infrastructure construction - What does this mean for Dell, HP Enterprise, and SuperMicro?

Server OEMs: Agentic AI driving next leg of AI infra build - whatdoes this mean for DELL, HPE, SMCI? We believe that Agentic AI will be significantly more CPU intensive driving upside totraditional server TAM, which in turn should benefit the server OEMs. In addition, challengesand pushouts from SMCI could portend the beginning of share losses from SMCI from itsmost recent reputational challenges (the most recent being the smuggling case see here). Mark C. Newman+1 212 845 7822mark.newman@bernsteinsg.com April Li+1 917 344 8339april.li@bernsteinsg.com The rise of more complex inference workloads, and in particular Agentic AI, isexpected to drive increased demand for CPU-based (aka “traditional”) servers. Phoebe Sun+1 917 344 8481phoebe.sun@bernsteinsg.com Agentic AI is more CPU intensive than training or simple chatbot inference with CPUs andtraditional servers orchestrating workloads. Recent commentaries from Intel/AMD/ARMsuggest the early inning of CPU demand upcycle.We see potential for a 3x-4x expansion Meanwhile, SMCI’s recent earnings showed a higher mix of enterprise and non-AI revenue, with AI server mix declining for the first time in 10 quarters.This likelyreflects 1) strength in traditional servers and 2) potential share loss in AI serversfollowing recent reputational challenges.We believe Dell and HPE are positionedto benefit from both dynamics. While these are longer-term themes and may not yet bereflected in the upcoming earnings,we see the current setup as supportive of upside heading into the prints -Dell reports on May 28th Dell is our more preferred beneficiary,given its scale, financing capacity, and robustsupport/services across the AI infrastructure.We consider Dell to be the real AI winner.Being the scale leader with the broadest financing capacity and deepest customerrelationships, Dell is best positioned to absorb displaced SMCI demand across cloud, HPE should also benefit from stronger server demand and incremental share gains in enterprise and sovereign accounts,justifying our estimates increase and multipleraise from 8x to 12.5x FY27 EPS and a $35 target price, Market-perform. SMCI's compliance overhang creates a real share shift opportunity as cloud, enterprise, andsovereign customers reassess vendor risk.SMCI’s FQ4 and beyond faces a much hardersetupas pre-scandal enterprise momentum fades, unusual data center delays raise the riskof soft cancellations, and potential risks around Nvidia GPU supply -leaves risk-reward BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We value Dell at ~17x (up from 14x previously) our FY 2028 EPS, or $280.We value HPE at ~12.5x (up from 8x previously) our FY 2027 EPS, or $35. Table Of Contents Part I: Agentic AI driving next leg of AI infrastructure build - Traditional Servers to benefit..........................................................................3Part II: Why did SMCI beat FQ3’26? And what could SMCI’s issues mean for DELL, HPE?..........................................................................5Part III: The read across from SMCI hints upside for DELL & HPE into the next print and beyond............................................................. 7Part IV: Valuation and model changes..............................................................................................................................................................................9DELL: AI Infra Winner - Raise estimates and increase TP to $280, Reiterate Outperform.................................................................... 9 DETAILS PART I: AGENTIC AI DRIVING NEXT LEG OF AI INFRASTRUCTURE BUILD - TRADITIONAL SERVERS TO We see AI infrastructure demand beginning to broaden beyond GPU-centric systems. In particular, the rise of morecomplex inference workloads, including agentic AI, is expected to drive increased demand for CPU-based servers. •Agentic inference is more CPU intensive than training or simple chatbot inference(Exhibit 1, Exhibit 2). Trainingworkloads are typically large, batch-oriented processes dominated by GPUs. In contrast, agentic workloads involve multi-step task execution, including tool calls, retrieval, routing, memory and context management, validation loops, and significant •Recent industry commentary points to a broader reassessment of CPU demand within AI infrastructure.AMD raisedits server CPU TAM growth outlook from 18% to over 35% annually, reaching over $120bn by 2030. Intel noted on its Q12026 earnings call that the CPU-to-GPU ratio in inference has already compressed from 1:8 to 1:4 and could trend toward1:1, while guiding to double-digit server CPU unit growth for the full year with momentum extending into 2027. Arm CEO According to above comments, our quick back-of-the-envelope analysis suggests that if assuming CPUs account for20-25% of traditional server ASP, total traditional server TAM could reach roughly $500-600B by 2030, or about 2.5x the$213B TAM in 2025. If instead traditional server T