⚠ Design Wins Are Won and Lost Each CycleModerate threat

AMD's marquee AI customers are a validation, not a lock-in, and the nature of this market means every design win must be defended and re-won continually. Hyperscalers and AI labs deliberately dual-source and re-evaluate their accelerator choices generation by generation, playing suppliers against each other to secure supply, price, and performance. A customer running Instinct today can shift the bulk of its next buildout back to Nvidia — or to its own in-house silicon — if the next AMD generation disappoints or if Nvidia's software and networking advantages prove decisive at scale. The revenue is real but the commitment is conditional.

This churn is especially dangerous given how concentrated AI-accelerator demand is among a handful of giant buyers. Losing or winning a single hyperscaler's next-generation order can swing AMD's data-center revenue by billions, making the business lumpy and hard to predict, and giving a few enormously powerful customers great leverage over AMD's pricing and roadmap. The same customers, moreover, are building their own chips, so today's design win may be a bridge to their own future independence rather than a lasting relationship. AMD's execution and the second-source imperative keep the momentum favorable for now. But an investor should not mistake a roster of prestigious deployments for a moat: in AI accelerators, the customer holds the power, the wins are perishable, and the revenue depends on beating Nvidia — holder of 90%-plus of the market¹ — again, and again, and again.