Google Announces Inference-Focused TPU Chips to Challenge NVIDIA; Gemini 4 Expected at I/O

Google revealed plans for new tensor processing units (TPUs) specifically designed for AI inference workloads, partnering with Broadcom, MediaTek, Marvell, and Intel in a four-partner supply chain to challenge NVIDIA's dominance. At Google I/O (May 19-20), the company is expected to unveil Gemini 4 with trillion-parameter MoE architecture and 2M+ context windows, plus Veo 4 with 30-second video generation capabilities.