Nvidia's rise to the top of the tech world is due to explosive advances in artificial intelligence (AI) technology. Known for its gaming-oriented graphic processing units (GPUs), Nvidia has positioned itself well for the future as its GPUs have proven ideal for the complex calculations required for AI tasks such as training large-scale learning models.
Companies like Google, Microsoft, Amazon and Meta are all relying heavily on Nvidia's H100 AI chip, which is designed specifically for AI applications and offers significant performance gains over traditional CPUs.
However, with demand high and supply limited, some major tech companies have already announced or are considering launching their own AI chips to gain an edge. Below is a list of companies that either have their own AI chips or are looking to introduce them in the near future to reduce their reliance on Nvidia.
Google's TPU chip series
Google has built and relied on custom-designed tensor processing units (TPUs) to run its AI models. Its latest chip, Trillium, announced at the Google I/O event in May, is said to be five times more powerful than its predecessor, the TPU v5e. Companies including Assembly AI, Hugging Face, and Anthropic use Google's TPUs.
Microsoft Azure Maia 100
Last year, Microsoft unveiled its Azure Maia 100 AI chip, designed to run cloud-based AI workloads. It's currently being tested with Bing AI chatbots, GitHub Copilot and OpenAI's GPT-3.5-Turbo language model, and Microsoft is already developing its successor.
Amazon's Trainium chip
AWS Trainium is a machine learning (ML) chip developed by AWS specifically for deep learning (DL) training. Last year, the company unveiled Trainium2, which is said to deliver up to four times the training performance and three times the memory capacity compared to the first-generation Trainium chip. Anthropic, which received a $4 billion investment from Amazon in March, plans to leverage the tech giant's AI chip.
Meta's Project Artemis
Facebook's parent company Meta is focused on AI, and in April revealed plans for its second-generation AI chip, called Artemis, which will surpass the first Meta Training and Inference Accelerator (MTIA) product released last year.
“The chip's architecture is fundamentally focused on delivering the right balance of compute, memory bandwidth, and memory capacity to serve our ranking and recommendation models,” the company said in a blog post.
AMD M-series processors
Advanced Micro Devices (AMD) unveiled its latest AI processors at Computex earlier this month. AMD CEO Lisa Su unveiled the MI325X accelerator, due to be released in the fourth quarter of 2024. AMD also unveiled a series of chips called the MI350, due to be released in 2025. The chips aim for a 35x performance boost in inference, the process of computing generative AI responses. AMD unveiled the MI400 series, due to be released in 2026.
Intel Gaudi AI chipIntel also unveiled its first-generation Intel Gaudi AI deep learning processor, which the company says has an architecture designed for deep learning performance and efficiency, while also enabling flexible system scaling.
