Mizuho raises Arm Holdings stock price target on AI CPU growth outlook

Mizuho raised its price target on Arm Holdings to $360 from $290 while reiterating an Outperform rating, implying further upside from the current $302.52 share price. The firm cited internal CPU ramps in calendar 2027, continued strength from customer CPUs, and agentic AI CPU opportunities; the new target is based on a 2.7x fiscal 2028 P/E-to-growth multiple versus 2.2x previously. The article also notes multiple analyst target increases and a positive revisions trend, though InvestingPro says the stock remains overvalued.

Analysis

The market is beginning to price a more important shift than a simple Arm re-rating: a broadening of AI monetization away from accelerators and toward CPU content per server. That matters because CPU attach rates are levered to every AI deployment stage, so the second-order winner is the ecosystem that captures design wins in inference-heavy and agentic workloads, not just the GPU names that dominate headlines. If this migration persists, it should support a longer-duration multiple expansion across IP holders and server CPU incumbents, while pressuring legacy x86 share faster than consensus expects.

The key competitive implication is that share gains in server CPUs can compound more slowly than investors want but then accelerate abruptly once hyperscalers standardize around a platform. That creates a classic “winner-take-more” setup for Arm-enabled custom silicon, but it also raises the bar for Intel to defend sockets: every 100 bps of lost share is high-margin revenue forfeiture with limited near-term offset. AMD benefits too, but the bigger incremental upside is likely on custom and semi-custom designs where customers can differentiate workloads; this is a favorable backdrop for any company tied to bespoke cloud silicon and a negative read-through for commoditized CPU vendors.

The contrarian risk is that the market may be extrapolating a 2027–2030 adoption curve into 2025 multiples too aggressively. The near-term bottleneck is not architecture enthusiasm but validation cycles, power/cooling constraints, and the willingness of hyperscalers to diversify away from proven stacks during a capital-efficient phase. If AI capex slows, or if agentic workloads prove less CPU-intensive than expected, the stock moves tied to this theme can unwind quickly because the narrative premium is already elevated.

From a trade perspective, this is a better relative-value than outright-long setup: the cleanest expression is long the beneficiaries of CPU share gain versus short the highest-risk incumbent. The upside is strongest if cloud TAM breadth continues improving into next earnings season; the downside is mainly multiple compression if analysts stop upgrading on the same three catalysts. Near term, options are attractive because implied volatility should remain supported by repeated target hikes, but the risk/reward is best when using defined-risk structures rather than spot equity longs.

Key Decisions for Investors

• Long AMD / short INTC for 1-3 months: best way to express ongoing server-CPU share migration; target 10-15% relative outperformance if share gains keep compounding, with INTC downside cushioned only if guidance stabilizes.
• Buy ARM call spreads 3-6 months out: use upside defined-risk exposure into the next validation cycle; favorable if the market keeps capitalizing 2027-2030 earnings power, but cap risk if the narrative cools.
• Add EVR on pullbacks only as a momentum proxy for design-win revisions; trade it tactically over 2-4 weeks rather than as a core long, since the setup is driven by estimate revisions more than fundamentals.
• Avoid chasing INTC on any short-term bounce unless there is evidence of socket retention in the next quarter; upside is likely to be faded because the market is anchoring on structural share loss.
• If you want a cleaner macro-hedged expression, pair long ARM / short QQQ for 1-2 months: the long leg benefits from AI-CPU optionality while the short leg reduces index beta if the market starts discounting expensive AI-duration names.