SemiAnalysis: Most New AI Compute by 2028 May Belong to Two Firms

The projection of compute concentration into two firms underscores a structural risk for decentralized AI networks, which depend on distributed supply to stay competitive. If the bulk of new capacity is locked under centralized control, the marginal cost advantage and reliability of open networks could erode, making their value proposition harder to defend. For crypto projects building on such infrastructure, the watch item is whether they can secure access to high-end chips or alternative sources before the gap widens. Whether decentralization can offset this scale advantage is the open question.