Amazon CEO Warns AI Inference Remains "An Order of Magnitude Too Expensive," Bets on Trainium Chips

Amazon ($AMZN) CEO Andy Jassy highlighted cost barriers in artificial intelligence adoption, stating that AI inference remains "an order of magnitude too expensive" to unlock ubiquitous enterprise deployment, while positioning proprietary Trainium silicon as a core driver to compress unit computing costs.
Despite existing pricing hurdles, Amazon Web Services (AWS) faces severe high-performance infrastructure supply constraints:
• AWS has sold out its major compute capacity commitments for 2026.
• The cloud provider has already booked the vast majority of its 2027 allocations alongside a substantial portion of 2028 availability.
Management emphasized that reducing inference expenses via custom silicon could drive an exponential wave of demand by making generative AI deployment economically viable across broader corporate workloads.