Google has joined Amazon in attacking Nvidia's stronghold with AI app developers
News Source
β’Tue, 20 Oct 2026 18:38:24 +0000
π° What Happened
Google has made TorchTPU, its version of PyTorch, open source. PyTorch is the toolkit most AI builders use to make apps. Google's version lets those apps run on its own TPU chips. Developers only need a simple change, with no rewriting.
Amazon made a similar move the day before. It released TorchNeuron for its Trainium chips and AWS. Amazon is also sending code to the main PyTorch project. Both giants want to break Nvidia's hold on the AI chip market.
π The Backstory
Nvidia makes the GPUs that most AI apps run on today. By some counts, PyTorch powers 85 percent of AI apps. Meta created PyTorch in 2016, then gave it to the Linux Foundation. Because PyTorch was built around Nvidia, its tools favor those chips.
Google and Amazon have their own AI chips but fewer users. To win customers, they need the same tools to run on their hardware. Making PyTorch work everywhere weakens Nvidia's advantage. It also gives startups more choices for where to build.
π― Why It Matters
More chip choices mean lower prices and faster AI apps. Right now one company controls most AI hardware. Competition decides how much AI services cost for everyone.
Google has made TorchTPU, its version of PyTorch, open source. PyTorch is the toolkit most AI builders use to make apps. Google's version lets those apps run on its own TPU chips. Developers only need a simple change, with no rewriting.
Amazon made a similar move the day before. It released TorchNeuron for its Trainium chips and AWS. Amazon is also sending code to the main PyTorch project. Both giants want to break Nvidia's hold on the AI chip market.
Nvidia makes the GPUs that most AI apps run on today. By some counts, PyTorch powers 85 percent of AI apps. Meta created PyTorch in 2016, then gave it to the Linux Foundation. Because PyTorch was built around Nvidia, its tools favor those chips.
Google and Amazon have their own AI chips but fewer users. To win customers, they need the same tools to run on their hardware. Making PyTorch work everywhere weakens Nvidia's advantage. It also gives startups more choices for where to build.
More chip choices mean lower prices and faster AI apps. Right now one company controls most AI hardware. Competition decides how much AI services cost for everyone.