AI chip startup Tensordyne has more than $200 million in expected orders for its new Napier inference system, positioning it as an alternative to Nvidia in AI inference, which is the part of AI where a trained model answers questions or generates outputs. Tensordyne says Napier targets data center constraints by running models faster per watt and fitting more computing capacity into each server rack. The startup built Napier with Broadcom and HPE-owned Juniper Networks, will manufacture it with TSMC, and says customers include Cirrascale, BlueSky Compute, and large tech and AI cloud firms. However, “expected orders” and letters of intent mostly signal interest, as data centers require months of testing and validation before commitments. For markets, the $200 million expected orders must survive the data-center checkout line, with near-term impact potentially showing as pricing and contract pressure in inference rather than immediate market share loss for Nvidia. For Tensordyne, the timing gap matters heading into a planned Series D funding round later this year, as investors will likely focus on how much forecast demand turns into shipped systems and repeat deployments.