Google is reportedly negotiating sales of its tensor processing units to rivals like Meta and Anthropic, potentially disrupting Nvidia's lead in AI hardware. These custom chips, designed specifically for AI tasks, offer efficiency gains that could save companies millions. As demand for AI computing surges, this shift highlights growing diversification in the sector.
Google pioneered tensor processing units (TPUs) in 2016, creating chips optimized for the matrix multiplications central to training and running large AI models. Unlike general-purpose graphics processing units (GPUs), which Nvidia has dominated, TPUs focus exclusively on these parallel computations, reducing inefficiencies in AI workloads.
This year, Google unveiled the seventh-generation TPU, named Ironwood, which supports its own models such as Gemini for language processing and AlphaFold for protein modeling. Experts note that while TPUs share similarities with GPUs, their specialization makes them more efficient for specific AI applications. "They can be much more efficient for these jobs and save potentially tens or hundreds of millions of dollars," says Simon McIntosh-Smith at the University of Bristol, UK.
Historically, Google used TPUs internally for over a decade, but recent reports indicate a pivot toward external sales. Companies including Meta and Anthropic are reportedly planning billion-dollar investments in Google's TPUs, seeking alternatives to Nvidia's scarce and pricey GPUs. This move could foster competition, allowing buyers to negotiate better terms with Nvidia while diversifying supply chains.
Francesco Conti at the University of Bologna in Italy explains the appeal: TPUs excel in parallelism suited to AI, though their niche design risks inflexibility if models evolve rapidly. Other hyperscalers, like Amazon with its Trainium chips, are also building custom hardware amid GPU shortages. As McIntosh-Smith observes, "What we haven’t heard about is big customers switching, and maybe that’s what’s starting to happen now." This trend signals maturation in AI infrastructure, potentially reshaping the industry's hardware landscape.