Summary

  • Google said it has unveiled two next-generation artificial intelligence (AI) chips as it moves to expand competition with Nvidia.
  • The report said the new products strengthen data protection features, large-scale information processing capacity, and memory performance, raising the possibility of a shift in the competitive landscape of the AI chip market.
  • As Nvidia faces stock-price pressure from controversy involving senior management, reports that OpenAI is considering buying Google TPUs show Google is pushing to expand its ecosystem.

Forecast Trend Report by Period

Google unveiled two next-generation artificial intelligence chips with enhanced security and memory performance, stepping up its challenge to Nvidia.

Decrypt reported on November 9 that Google will add TPU 8tx and TPU 8i Max to its eighth-generation Tensor Processing Unit lineup. The new chips are aimed at improving data protection and large-scale information processing for corporate AI model training and inference.

TPUs are chips Google developed in-house for AI computing and compete with Nvidia’s graphics processing units, or GPUs. Google, which has expanded TPU access beyond its own services, said it now has about 100 customers, including AI labs and universities.

TPU 8tx features confidential computing technology designed to protect sensitive corporate data. It is built to isolate customer information and model parameters at the hardware level even while AI models are processing data. Google said the chip combines fifth-generation AMD EPYC processors with its own networking technology, cutting latency by as much as 50% versus the earlier TPU 8t.

TPU 8i Max is designed to boost memory performance for the large-scale data processing required by AI agents. Google combined two TPU 8i chips into a single package, doubling computing power and memory bandwidth from the previous version. The company also open-sourced TPU Sync, software that optimizes data movement between chips and servers, and said it can cut data-transfer latency and response start times by more than 90% in some workloads.

Google’s product announcement came as Nvidia faces pressure on its stock amid controversy involving senior management. Nvidia shares fell 9.3% on November 5 and were down about 5% again in trading on November 9. Reports that the board is investigating issues related to Chief Executive Officer Jensen Huang’s private life appear to have weighed on investor sentiment. Nvidia said, however, that it has found no evidence that company secrets were leaked.

The competitive landscape in the AI chip market could also be shifting. Reports have said OpenAI is considering buying Google TPUs to reduce its reliance on Nvidia, though that has not been officially confirmed. Google is also seeking to expand its ecosystem by releasing software that allows developers to run AI models built on PyTorch and Hugging Face on TPUs.

Still, the performance gains Google presented were measured against its own previous-generation products, not directly against Nvidia’s latest chips. The company also did not disclose pricing or a detailed launch schedule for the new products.

Suehyeon Lee