Post

Reviewing the AMD earnings call transcript, I found Lisa Su was asked an unavoidable question—Elon Musk just spent tens of billions buying NVIDIA GPUs to build the world's largest training cluster, so why not choose AMD? Her answer was very restrained: AMD's market share on the inference side is growing rapidly, and customer choice depends on the specific workload. To translate: the top-tier large orders for training clusters currently still belong to NVIDIA. But AMD's earnings data is not bad. Q2 revenue was $11.5 billion, up 50% year-over-year. The data center business doubled outright. Looking at these numbers alone, it doesn't seem like a company whose stock dropped nearly 10% that day. The problem lies in the AI revenue guidance. The market doesn't want "fast growth," it wants "growth fast enough to challenge NVIDIA." AMD delivered growth but not disruption. This gap reflects a solidifying pattern: the AI chip market is splitting into two tiers. The first tier is ultra-large-scale training clusters with tens of thousands of cards—NVIDIA currently has no rivals here. The biggest buyers like Musk, OpenAI, and Anthropic default to NVIDIA. The second tier is inference deployment and enterprise AI—AMD is rapidly gaining share here, with MI300 series shipments ramping up. Lisa Su revealed a key piece of information on the call: AMD's AI customer count has tripled year-over-year. This shows it's not that no one is buying, but that the scale of buyers hasn't reached Musk's level yet. Next to watch is whether AMD's AI revenue in Q3 can step up again quarter-over-quarter. If it can, the current pullback is a good observation window. If growth slows, the market's valuation framework for the "AI number two" may be compressed further. The above does not constitute investment advice; chip stocks are volatile, so please base your judgments on your own research.