What does the next generation of AI infrastructure look like inside the world's largest banks?

Why are major financial institutions looking beyond conventional GPU infrastructure, and what makes SambaNova's SN50 and RDU architecture different?