Summary Bullets:

• Google Cloud and open-source startup Hugging Face signed a deal to share hardware, cloud infrastructure, open data, and open-source models and libraries.

• The partnership meets growing enterprise demand for generative AI (GenAI) software that is optimized for specific tasks and reflects the increasing popularity of open-source applications.

Google Cloud and Hugging Face announced an agreement that will enable developers to access the Google Cloud infrastructure to fine-tune and operate Hugging Face’s open-source models without the need for a Google Cloud subscription. The partnership will also enable Google Cloud customers to train and deploy Hugging Face models within Google Kubernetes Engine (GKE) and Vertex AI, the company’s ML platform offering Gemini, a multimodal platform from Google DeepMind. Vertex AI and GKE will be available on the Hugging Face platform during H1 2024.

The deal means the developer community can train and operate Hugging Face models on the Google Cloud infrastructure running on the company’s proprietary tensor processing unit (TPU) architecture, entailing access to very powerful hardware (for more, please see Google Leverages AI and Analytics, Partners with Nvidia, to Gain Advantage in the Cloud Wars). The developers will also be able to use virtual machines running on Nvidia H100 GPUs, and on Intel Sapphire CPUs, having their pick of silicon architectures. This can drive innovation by giving open-source developers access to high quality storage and compute. Google says that unlike general purpose graphics processing units, which are designed for parallel computing and are versatile across various compute workloads, TPUs are purpose-built for AI and ML workloads, focusing on tensor operations to achieve higher speeds and energy efficiencies.

Hugging Face is adept at striking partnerships and made a similar deal with Amazon almost a year ago to help developers access the Amazon Web Services (AWS) infrastructure as well as to work on and extend Amazon’s AI platform. In 2023, the startup, based in New York (US), also signed agreements with Nvidia, Microsoft, and Dell Technologies, which offers customers the option to train and implement the open-source models on their own premises as well as on hybrid environments.

In a similar fashion to the Amazon deal, this agreement allows Google to grow the capillarity of its AI ecosystem while getting ready for the forthcoming release of Gemini Ultra, Google’s enterprise GenAI model, at some point in 2024. Google’s move caters to growing demand for customization by enterprises eager to find practical uses for the technology. The GenAI ecosystem is increasingly looking to the open-source world because these models give customers more transparency than black-box models such as OpenAI’s. The opportunity for customization and community collaboration is also greater. The partnership also opens a potential avenue for Google to drive penetration of its cloud market as an increasingly large community of open-source AI users grows to rely on its powerful hardware and services, while categorizing the move under the lofty goal of ‘democratizing AI.’

This agreement means Google will be well-positioned to help enterprises with the new wave of customization of GenAI applications. As the market slowly matures, companies will demand to get more bang for their buck and adopt increasingly customized applications for specific business cases. This type of partnerships can help them build their own AI systems leveraging the innovations of the open-source community while maintaining the security and performance of established hyper scale deployments.