The rivalry between tech titans Google and Nvidia is said to be heating up as the focus of the artificial intelligence (AI) boom shifts from teaching AI to actually using it to get answers quicker than ever. While Nvidia CEO Jensen Huang has claimed his chips (GPUs) are more versatile than Google’s, Google is preparing a major counter-move. At the Google Cloud Next conference this week, the search giant is expected to double down on its custom-made AI chips, known as Tensor Processing Units (TPUs), to meet a massive surge in demand, as per Bloomberg.
Training vs. inference is the new battleground
In the world of AI, there are two main phases: Training (teaching a model like ChatGPT to learn) and Inference (where the AI actually answering your questions). While Nvidia’s chips are currently the “gold standard” for training, Google believes the future lies in specialised chips built specifically for inference that will allow the models to answer queries quickly.“It now becomes sensible to specialize chips more for training or more for inference workloads,” said Google Chief Scientist Jeff Dean, Bloomberg reported.This matters because more people are using AI daily, which means companies need cheaper and more efficient ways to run these models at scale. That’s why, companies like Google itself is using a mix of its TPUs and Nvidia GPUs.
Jensen Huang’s take vs. Demis Hassabis ’ argument
Nvidia’s Jensen Huang recently argued that his GPUs are superior because they can handle “a whole bunch of applications” that specialised TPUs simply cannot do. However, Google DeepMind CEO Demis Hassabis sees it differently. He noted that the world’s leading AI labs are increasingly desperate to get their hands on Google’s hardware. “A lot of people would like to run on both,” Hassabis said, highlighting that interest in TPUs has reached an all-time high.Google has a decade-long head start in designing its own chips. Analysts say this gives Google a “home-field advantage” as AI agents (programs that can perform complex tasks on a user’s behalf) become the next big thing.“The battleground is shifting towards inference,” said Chirag Dekate, an analyst at Gartner. He noted that Google’s Gemini model is already among the fastest at complex reasoning, largely thanks to the infrastructure Google has built under the hood.Nvidia has already spent a reported $20 billion to bolster its own inference technology by acquiring Groq, that offers inference technology.Moreover, Google has previously touted inference capabilities for its chips. It also considered releasing separate chips for training and inference early on, according to Partha Ranganathan, a vice-president and engineering fellow at Google.
