The AI Chip War Enters Its Most Dangerous Phase Yet

For three years, NVIDIA's H100 and its successors have been the undisputed currency of the AI boom. Data centres queued for months to acquire them. Startups built entire business models around access to them. Governments stockpiled them as strategic assets. That era may be ending.

A confluence of forces — custom silicon from the hyperscalers, a new generation of challengers from AMD and Intel, and an increasingly assertive Chinese domestic chip industry — is beginning to erode the moat that made NVIDIA the most valuable company in the world.

The Hyperscaler Threat

Google's TPU v5, Amazon's Trainium 2, and Microsoft's Maia 100 are no longer experimental projects. They are production-grade accelerators handling a growing share of the AI workloads that once ran exclusively on NVIDIA hardware. Morgan Stanley estimates that by 2027, custom silicon could account for 35% of AI compute spending at the top five cloud providers.