AMD: Why I'm Going From Very Bearish To Confidently Bullish

This is less about a one-quarter revenue pop than about AMD converting from a cyclical share-taker into a contracted infrastructure supplier with duration. The second-order effect is that the market will likely start capitalizing AMD on a hybrid hardware-plus-capacity expectation, which matters because long-dated AI infrastructure visibility usually compresses the discount rate on forward gross margin expansion. Meta's scale also creates a reference contract that can pull other hyperscalers toward a multi-vendor strategy, especially if they want to avoid single-source exposure and retain pricing leverage versus NVIDIA.

The margin setup is more important than the top-line optics. If AMD can raise prices while utilization stays tight, the mix shift into inference-heavy deployments should produce a cleaner gross margin inflection than many expect, because inference clusters are typically more cost-sensitive and much more exposed to energy/footprint economics than training. The CPU shortage is an underappreciated read-through: agentic AI increases orchestration overhead and general-purpose compute demand, so the bottleneck may migrate from GPU scarcity to platform-level supply constraints, benefiting vendors with credible CPU+GPU attach and hurting buyers that relied on surplus server capacity.

The main risk is timing mismatch: revenue recognition and installed base monetization will lag headline announcements by quarters, while execution risk around deployment cadence, software stack maturity, and customer concentration remains non-trivial. Another hidden risk is that a cheaper alternative only stays attractive if perf/watt and software portability hold; if the ecosystem friction is higher than expected, the competitive moat remains with the incumbent stack. Also watch for hyperscaler capex normalization in 2H26-27, which could turn today's multi-year visibility into tomorrow's digestion phase.

Consensus may be underestimating how much of this is a supply-chain repricing story rather than a pure product share story. If the CPU market tightens, the broader server complex can re-rate on pricing power, not just unit growth, and that would spill into memory, networking, and ODMs before showing up in headline AI revenue. The overdone part may be extrapolating immediate EPS upside; the underdone part is the structural uplift to AMD's gross margin and bargaining power if it remains one of the few credible alternative platforms at scale.