Yee Jiun Song, VP of Engineering at Meta, stated that the company's self-developed chips are more efficient when running AI models compared to NVIDIA's currently shipping products. The company plans to deploy over 1 gigawatt equivalent of power on its self-developed chips within the next 12 months, with accelerated expansion expected thereafter. Meta collaborates with Broadcom for chip design and TSMC for manufacturing. The company has canceled its previously planned training chip project to focus on inference chips to control costs.