AMD’s Server CPU Growth Is Real but Partly Cyclical; GPU Remains a Show-Me Story

Core Conclusion

AMD’s March-quarter results and June-quarter guidance confirm that server CPU demand is accelerating faster than previously expected, but at least some of this strength is cyclical (Intel’s supply shortfall and likely double-ordering) rather than purely share-driven structural gains. Meanwhile, GPU revenue declined sequentially in Q1, and the company offered no visibility that MI455 rack-scale products will close the competitive gap with NVIDIA or ASIC alternatives. The Equal-weight rating is maintained, with a $410 target—driven almost entirely by a higher target multiple (37x vs. 32x previously) rather than upward EPS revisions. At ~$355, the stock offers limited upside without material GPU proof points in the second half.

Evidence Chain

1. Server CPU TAM growth doubled, but sustainability is questionable.

Q1 Data Center revenue of $5.775bn beat consensus by ~$275mn, with AMD guiding for 70%+ year-over-year server growth in Q2. Management more than doubled its five-year server TAM CAGR assumption from 18% to 35%. However, Intel’s CPU volumes fell 5% in Q1, and component shortages likely drove customers to double-order from both suppliers. A portion of AMD’s surge is therefore a tactical response to constrained supply, not a pure structural shift in the compute architecture. Our own forecast falls below the new 35% CAGR — we see a higher probability of demand normalization once Intel resolves its capacity issues, which may occur toward late 2027.

Investment implication: Server revenue estimates are being pulled forward. The multiple assigned to AMD’s CPU business has expanded (Intel’s P/E rose meaningfully, lifting the peer-group ceiling), but if TAM growth reverts to the mid-teens, today’s ~37x valuation on 2027E EPS becomes hard to defend.

2. GPU carries no positive catalyst until the second half.

AMD acknowledged that GPU revenue fell sequentially in Q1, as China exposure exited the model. Q2 guidance suggests double-digit sequential growth, but the company would not confirm that GPU returns to Q4 levels. Customer feedback on MI455 remains “positive but inconclusive” — a descriptor that signals limited conversion to committed orders. The rack-scale platform promises a step change in performance and ease of deployment, but its competitive positioning against NVIDIA’s Blackwell family and hyperscaler-built ASICs will not be clear until late 2026.

Investment implication: The consensus narrative on GPU is optimistic without hard evidence. Investor enthusiasm may persist on AI theme momentum, but any disappointment in MI455 performance or customer adoption could compress the stock’s multiple rapidly, as GPU revenue was a primary driver of bulls’ terminal value assumptions.

3. Non-data center segments provide a buffer, not a boost.

Client revenue (PCs) grew 25.8% year-over-year but is expected to peak in the first half, with a much softer PC TAM in H2. Gaming revenue declined 14.6% sequentially and will fall more sharply year-over-year in 2026. Embedded continues to track the broader analog recovery (+6.1% y/y) but adds only ~$873mn per quarter to total revenue. Together, these three segments account for about 44% of revenue but are unlikely to grow faster than single digits for the next 12 months.

Investment implication: Revenue growth must come almost entirely from Data Center. Any softening in CPU momentum or GPU ramp would expose AMD’s overall growth profile and earnings trajectory.

Key Disagreements & Risks

• Double-ordering risk is underappreciated. If Intel resolves its output constraints by 4H27, a portion of AMD’s recent server orders could unwind, compressing the TAM back toward 15–20% CAGR.
• GPU competitive risk is not priced in. NVIDIA is shipping Blackwell at scale; ASIC vendors are gaining share in hyperscale inference. AMD needs MI400 to be a generational leap in 2027, and that outcome is still 18 months away.
• EPS is declining for FY27 even as the stock rerates. Our Non-GAAP EPS estimate for 2027 fell from $13.06 to $12.12 due to higher operating expenses and stock-based compensation. The entire $50 PT increase came from a 5-turn P/E expansion.

Valuation & Trading Implications

Price target of $410 is based on ~37x our 2027E ModelWare EPS of $11.10 (vs. $13.06 Non-GAAP consensus- methodology EPS). The multiple was raised from 32x to reflect peer-group expansion and a more bullish CPU narrative, but the EPS base is now lower. Current share price of $355.26 sits just 15% below that target, leaving a narrow risk/reward band. Upside to $570 (bull case) requires GPU to cement a #2 position in Data Center AI—a scenario that lacks any confirming evidence today. Downside to $210 (bear case) would follow a GPU miss and Intel share recovery. Given the asymmetry, we would not add exposure without a material catalyst for GPU.

Appendix: Key Estimate Changes