Find AI Tools
two new AI chips to compete with Nvidia screenshot

What is two new AI chips to compete with Nvidia?

Google Cloud has introduced two new tensor processing units (TPUs) designed to compete with Nvidia's dominance in AI chip hardware. These custom-built processors offer improved performance and lower costs compared to Google's previous TPU generations, making them an alternative for organisations running large language models and other AI workloads on Google Cloud. The chips integrate directly into Google Cloud's infrastructure, allowing users to rent computing capacity rather than purchase hardware outright. Whilst Google continues to support Nvidia GPUs on its platform, these new TPUs represent an effort to reduce dependency on external chip manufacturers and offer customers more pricing options for their AI projects.

Key Features

Custom AI chip architecture

Purpose-built tensor processors optimised for machine learning workloads

Integration with Google Cloud Platform

Direct access through standard cloud infrastructure without additional setup

Cost reduction

Lower per-unit pricing compared to previous TPU versions

Performance improvements

Faster processing speeds for model training and inference tasks

Flexible capacity

Rent compute resources on demand rather than purchasing hardware

Multi-chip support

Google Cloud continues offering Nvidia GPUs alongside TPUs for choice and flexibility

Pros & Cons

Advantages

  • More affordable alternative to Nvidia chips for AI workloads on Google Cloud
  • Faster processing speeds than earlier TPU generations
  • No hardware procurement needed; scale up or down compute capacity on demand
  • Tight integration with Google Cloud services and tools
  • Reduces vendor lock-in by offering options beyond Nvidia

Limitations

  • Limited to Google Cloud Platform; cannot be used with other cloud providers or on-premises
  • Smaller ecosystem and third-party software support compared to Nvidia's established market position
  • Still early adoption phase; long-term reliability track record not yet fully proven in production environments

Use Cases

Training large language models and other deep learning models at scale

Running inference for AI applications with high throughput requirements

Organisations seeking cost savings on GPU expenses for existing Google Cloud deployments

Companies building custom AI models that need dedicated compute resources

Research institutions requiring significant computational capacity for AI experiments