Google Custom Slides: Changing the Game in AI

At Google’s headquarters in Mountain View, California. Tests It runs on the company’s own microchips, called Tensor Processing Units (TPUs). These custom chips were originally designed for internal workloads, but have been available to cloud customers since 2018. In July, it was announced that Apple was using Google’s TPUs to train its AI models. Google also trains and runs its Gemini chatbot.

Developing these proprietary AI accelerators is a complex and expensive process. Even hyperscalers like Google can’t do it alone. They work with companies like Broadcom, which has spent more than $3 billion on these partnerships. The chip developer also helps Meta design its AI chips and handles all the peripherals, including I/O and packaging.

Google’s TPUs are designed to bring powerful computing capabilities to the cloud. By working with Broadcom, Google can focus on providing the computing power and leave the groundwork to the chip developer. This collaboration is essential to making these complex chips a reality.

The importance of energy efficiency

Energy efficiency is critical to the development of AI accelerators. According to Google’s latest environmental report, emissions increased by nearly 50 percent between 2019 and 2023, in part due to the growth of data centers to power AI. By 2027, AI servers are expected to consume as much energy annually as a country like Argentina.

Google is committed to reducing carbon emissions from its infrastructure to zero. The company has started using direct-to-chip cooling in its third-generation TPU, which uses significantly less water than traditional cooling methods. This approach is also used by Nvidia.

The role of devices

Hardware plays a crucial role in the development of generative AI tools. Google’s push to manufacture its own chips is essential to creating powerful and efficient AI accelerators. The company has announced that its sixth-generation TPU, called Trillium, will be released later this year.

Google’s partnership with Broadcom allows the company to focus on delivering powerful computing capabilities while leaving the groundwork to chip developers. This approach is critical to making complex chips a reality.

The future of artificial intelligence

The future of AI looks bright. With the development of powerful and efficient AI accelerators, the possibilities are endless. From training AI models to running them on devices like the iPhone and Mac, the potential applications are vast.

Google’s commitment to reducing carbon emissions from its infrastructure is a key step in making AI a more sustainable technology. The company has begun using direct-to-chip cooling in its third-generation TPU, which uses significantly less water than traditional cooling methods.

The future of AI will be determined by the development of powerful and efficient AI accelerators. And with Google committed to making its own chips, the possibilities are endless.

Key points

• Google’s own chips are revolutionizing the world of artificial intelligence.
• Developing these proprietary AI accelerators is complex and expensive.
• Google is partnering with companies like Broadcom to focus on computing capabilities and leave the core work to the chip developer.
• Energy efficiency is key in the development of AI accelerators, as Google aims to reduce carbon emissions from its infrastructure to zero.

If you want access to all articles, take advantage of our special promotion and subscribe here!