Five Things the Market Is Missing About AMD's Data Center Share Gains

AMD's server CPU share has grown from roughly 15% at the start of 2022 to north of 30% on the most recent channel data. Crossing 30% is a psychological and practical threshold. Above 30%, enterprise IT procurement teams stop treating AMD as the alternative vendor and start treating it as a primary vendor with full ecosystem support.

That shift has a 12 to 18 month tail of compounding benefits. Historically, when a server CPU vendor crosses the 30% share mark, the following 18 months have delivered an average of 400 basis points of additional share gain before the pace plateaus. Applied to AMD's current trajectory, that implies 34 to 36% share by the end of 2026. Every share point in the server CPU market is roughly $600 million of revenue.

The TD Cowen note on Intel this week confirms the dynamic from the other side. CPU demand is genuinely recovering, but the share trajectory still favours AMD at the margin. That combination, total market growth plus share gain, is the cleanest fundamental setup in large-cap semiconductors.

AMD's MI300 series has moved from a curiosity to a meaningful revenue contributor. The 2025 revenue contribution from the data centre GPU line was above $6 billion. The 2026 consensus puts it above $9 billion. That is a multi-billion dollar business growing at 50%, inside a company where the base consumer PC business is flat.

The MI300 will not compete with Blackwell on performance. It does not need to. The customer profile for MI300 is the workload that is cost-sensitive, the workload that wants a second source, and the workload that is specifically avoiding CUDA lock-in. That is a real market with a real growth trajectory, and it is structurally separate from the premium GPU segment where Nvidia dominates.

Oracle, Microsoft, and a growing roster of sovereign AI deployments have all committed to MI300 volumes. Those commitments are locked through 2026 and most into 2027. The revenue visibility is higher than the market typically assigns to a new GPU architecture.

The AI agent workload pattern is CPU-heavy in a way that the training workload is not. Agents orchestrate many small inference calls and data operations. Those operations run best on fast CPUs with tight GPU coupling. That pattern favours AMD's EPYC plus MI300 stack in a way that the pure-training workload pattern did not.

This is the reason EPYC is being specified into AI agent server SKUs at a rate the 2023 forecast did not anticipate. The installed base of AI agent deployment is still small but growing fast, and the CPU attach is structurally higher than the training attach rate.

Historically, when a workload shift advantages a vendor's architecture, the revenue benefit accrues over 18 to 24 months. AMD is in month 6 of that cycle. The next 12 months should show accelerating EPYC attach in AI-configured servers.

AMD's client PC business is running at roughly flat revenue. Most AMD bulls treat that as a drag. We treat it as a free option. The PC replacement cycle from the 2020-21 demand surge is overdue. When it inflects, the incremental dollars land at roughly 30% operating margin.

Historically, the PC refresh cycle runs 4 to 5 years. The clock on the pandemic surge is now at 4 years. Intel's CPU demand commentary this week was the first signal that the refresh is starting to materialise. AMD gets a meaningful share of that refresh at a higher margin than the pre-2020 cycle delivered.

A recovery in the client segment adds $2 to $3 billion of revenue at 30% margin, or roughly $0.50 per share of earnings power. That is not in any sell-side model we have reviewed.

AMD trades at 34x forward earnings. That is high in absolute terms and low relative to its own history. At the 2021 peak, AMD traded above 55x. At the 2022 trough, it traded at 23x. The median over five years is 38x. The current multiple is below the five-year median despite earnings growth reaccelerating.

The setup is analogous to Broadcom in 2023, where the multiple compressed during the AI narrative because Nvidia sucked up the attention. Broadcom subsequently outperformed Nvidia over a 15-month window as the narrative caught up with the fundamentals. AMD is in a similar structural position now.

The multiple arithmetic is straightforward. Mean reversion to the five-year median delivers roughly 12% upside from here. Earnings revisions for 2026 are trending higher. The combination is a stock priced for disappointment that is fundamentally set up to deliver the opposite.