AWS plans to offer access to Nvidia's H200 chips, following Azure's similar dual-pronged approach of offering its own Maia chips as well as Nvidia's latest
- Amazon Web Services announced Trainium2, a chip for training artificial intelligence models, and it will also offer access …
Context & Ripple Effects
AWS’s move extends a custom-silicon strategy that began with its earlier Trainium training chip, rather than treating in-house hardware as a wholesale replacement for outside accelerators. Its earlier consideration of AMD’s MI300 chips also showed that AWS was keeping its AI-compute catalog open to multiple suppliers.
The significance is the cloud-service model: AWS is pairing a proprietary training option with Nvidia capacity, mirroring Azure’s stated Maia-and-Nvidia approach. That gives customers a choice while AWS builds its own chip platform.
First-order effects
- AWS customers gain access to Nvidia H200 capacity alongside the newly announced Trainium2 option for AI model training.
- AWS must position and support two distinct training-accelerator paths: its Annapurna-designed silicon and Nvidia’s hardware.
Second-order effects
- Nvidia remains embedded in AWS’s AI offering even as Trainium2 gives AWS a product it can differentiate and develop independently.
- Competing cloud providers face greater pressure to offer mixed accelerator portfolios rather than make customers choose between proprietary chips and Nvidia GPUs.
Third-order effects
- If this model persists, cloud AI infrastructure will become more heterogeneous: providers will compete on the breadth and integration of accelerator choices, not solely on access to one chip vendor.
- Custom cloud silicon can gain adoption without displacing Nvidia outright; the durable contest shifts toward which workloads customers place on each platform.
The trend: This is one data point in the shift toward heterogeneous AI clouds that combine proprietary accelerators with Nvidia hardware to serve different customer and workload needs.