3 AI Chipmakers With Far More Upside Than AMD

The article argues AMD faces intensifying competition in AI chips from Nvidia, Broadcom, and Amazon, with AMD's data center revenue up 39% YoY versus Nvidia's 75% and Broadcom's AI semiconductor revenue up 106%. Broadcom's management expects more than $100 billion of revenue from this division by 2027, while Amazon's chip business has surpassed a $20 billion annual run rate. The piece is bearish on AMD's upside because much of its growth is already priced in, though it remains constructive on the broader AI infrastructure trade.

Analysis

The real message is not that AMD is losing share; it’s that the AI capex stack is fragmenting from a single-gpu winner-take-most model into a multi-vendor procurement regime. That is structurally bullish for the platforms that own the customer relationship and workload specification — hyperscalers and vertically integrated ASIC providers — because they can monetize design, software integration, and capacity reservation simultaneously. In that world, AMD is exposed to the worst part of the value chain: it carries product risk without the same pricing power or ecosystem lock-in.

The second-order implication is that Nvidia’s moat is being attacked less by a direct substitute than by workload-specific alternatives that can win on TCO for narrowly defined tasks. That is more dangerous for AMD than headline market-share charts suggest, because every custom-chip win by Broadcom or Amazon reduces the addressable pool of “good-enough GPU” demand in future data-center refresh cycles. The risk horizon is months-to-years, not days: the market may continue to bid AMD on AI optionality, but the conversion of inference workloads away from general-purpose GPUs is the lever that compresses multiple expansion over the next 4-8 quarters.

The contrarian angle is that AMD’s premium valuation may be less about current growth and more about being the cleanest public lever to a broad AI acceleration trade after Nvidia became crowded. That positioning can hold until guidance starts missing the implied narrative, but once custom silicon spend is recognized as additive rather than complementary, AMD’s relative multiple should de-rate faster than fundamentals. The biggest upside surprise would be a materially better software stack and/or a surprise win in high-volume inference, but absent that, the stock looks vulnerable to being the funding leg in AI relative-value trades.

Near term, the cleanest risk is not a collapse in AI spending but a slower-than-expected mix shift that keeps AMD relevant just long enough to maintain its premium. The longer-term risk for the bulls is that capacity expansion at hyperscalers shifts bargaining power away from merchant GPU vendors entirely, making gross-margin durability the key variable rather than unit growth.

Key Decisions for Investors

• Short AMD vs long NVDA in a 3-6 month pair: Nvidia remains the cleaner beneficiary of broad training demand while AMD is more exposed to multiple compression if custom ASIC adoption continues to accelerate.
• Long AVGO on any post-earnings consolidation over a 6-12 month horizon: the market still underestimates how much recurring design-in revenue plus manufacturing slots can re-rate the business versus pure-play accelerator vendors.
• Long AMZN as a hidden AI infrastructure owner, not as a retail consumer play, with a 9-18 month view: if custom chip volumes keep scaling, the valuation gap versus merchant chip vendors should widen.
• Consider buying AMD put spreads 2-4 months out, funded by selling downside farther out: thesis is not a crash, but a multiple reset if guidance fails to re-accelerate relative to hyperscaler-driven peers.
• If you need AI exposure but want lower execution risk, rotate capital from AMD into a basket of NVDA/AVGO/AMZN; the expected return per unit of competitive risk is materially better over the next 12 months.