Google Takes Aim at Nvidia With New Tensor Chips to Power AI Boom

Google revealed 2 AI processors at its Cloud Next 2026 conference in Las Vegas on Wednesday, marking the business’s 8th generation of custom-made silicon developed to challenge Nvidia’s AI chip supremacy.

The training-focused TPU 8t provides almost 3x the calculate efficiency per pod compared to its predecessor, with a single superpod scaling to 9,600 chips and providing 121 ExaFlops of calculate capability. The architecture likewise provides 2.8 x much better price-to-performance, according to Google.

The TPU 8i takes a various method, enhancing for reasoning work with 3x more on-chip SRAM than previous generations– 384 MB of on-chip SRAM coupled with 288 GB of high-bandwidth memory. The chip provides up to 80% much better efficiency per dollar and 2x the efficiency per watt, the business declared.

Both chips take advantage of Google’s brand-new Boardfly architecture, which attains as much as a 50% enhancement in latency for communication-intensive work by lowering network size, the technical paperwork programs.

The hardware statement follows Google’s broadened collaboration with Anthropic previously this month, which will offer the AI start-up with several gigawatts of next-generation TPU capability. The offer highlights how Google is leveraging its custom-made silicon to bring in significant AI business looking for options to Nvidia’s GPUs in the progressively competitive facilities market.

Google CEO Sundar Pichai placed the chips as purpose-built for AI representatives, specifying they provide the huge throughput and low latency required to simultaneously run countless representatives cost-effectively. The business has actually currently protected adoption from Castle Securities, with the monetary services firm picking TPUs to power their AI work.

The dual-chip technique shows the diverging computational requirements of contemporary AI systems: huge parallel processing for training frontier designs versus fast, memory-intensive operations for releasing those designs as interactive representatives.

Pichai stated Wednesday that Google is on track to invest as much as $185 billion this year alone to power AI facilities for the “agentic period,” with the company currently producing almost 75% of its brand-new code with AI under the careful eye of engineers.