Buy AMD shares as graphics chip sales boost earnings, says Citi

Citi upgraded AMD to Buy from Neutral and raised its price target to $575, implying more than 17% upside from Thursday's close. The bank now expects AMD AI sales of $33 billion in 2027 and $50.8 billion in 2028, citing stronger GPU demand and a larger-than-expected Meta relationship, including a multiyear deal to deploy up to 6 gigawatts of GPUs. Citi also said AMD could surpass its goal of earning more than $20 per share by 2028.

Analysis

This read-through is less about a one-day analyst upgrade and more about a potential reset in the AI hardware duopoly. If a hyperscaler meaningfully diversifies away from a single GPU supplier, the market will start discounting a broader second-source architecture, which compresses the premium embedded in NVDA’s “default choice” status and raises the value of custom silicon economics across the stack. The bigger second-order effect is not just incremental GPU share for AMD, but the signaling value to every large cloud buyer that multi-sourcing at scale is now operationally feasible.

The underappreciated upside is in AMD’s CPU franchise as AI workloads become more agentic and orchestration-heavy. That shifts spending from only accelerators toward general-purpose compute, interconnect, and memory-adjacent infrastructure, where AMD can participate more broadly than pure GPU competitors. If that mix shift persists, the earnings power re-rating could be more durable than the market is pricing, because it broadens AMD from a single-product AI beneficiary into a full-stack compute beneficiary.

The main risk is that this thesis is front-running execution that still has to clear product ramp, yield, and deployment timing. The market may be extrapolating the headline customer win faster than the revenue recognition curve, creating room for volatility if near-term bookings or supply cadence disappoint. Over a 3-6 month horizon, the key reversal trigger is any indication that the hyperscaler is using AMD as leverage rather than as a true strategic secondary platform.

Consensus still looks too anchored to NVDA’s share dominance and too skeptical that customers will accept lower switching costs through custom designs. That makes AMD the cleaner momentum trade, but also a crowded one after a >100% year-to-date move; the better risk/reward may now be relative value rather than outright longs. The setup favors owning the beneficiary of diversification while expressing caution on the incumbent if the market starts pricing in margin pressure from second-sourcing.

Key Decisions for Investors

• Long AMD into the next 1-3 month catalyst window; target continuation if hyperscaler commentary confirms multi-year deployment cadence. Risk/reward is attractive on a structural rerating, but trim if the stock fully discounts 2028 growth assumptions before execution data shows up.
• Pair trade: long AMD / short NVDA for 2-6 months to express share-reallocation risk without taking directional AI beta. This works best if investors start valuing second-source optionality and custom GPU economics rather than only absolute demand growth.
• Buy META on weakness as a beneficiary of lower infrastructure TCO and supply-chain optionality over 6-12 months. The setup improves if capex efficiency becomes a focal point in guidance, but the trade should be sized assuming a delayed revenue payback.
• Use AMD call spreads instead of outright stock for a 3-6 month tactical expression; upside is tied to confirmation of deployment ramps, while downside is cushioned if the market has over-advanced the story.
• If AMD/NVDA relative strength stalls for two consecutive weeks, reduce exposure and rotate into broader AI infrastructure names with less execution risk. That preserves gains if the market shifts from narrative to proof-of-ramp scrutiny.