Apple is scaling up pre-training of its foundation models on Google's Tensor Processing Units. That puts some of the costliest work in its AI strategy on chips designed by one of its biggest rivals. The partnership dates to 2024 and now also includes a Gemini access deal.
Apple is scaling up pre-training of its foundation models on Google's Tensor Processing Units, or TPUs. Pre-training is the compute-heavy phase of building a large language model, so the hardware choice is strategic rather than a procurement detail.
From 8,192 chips to a 2,048-chip slice
TPUs are custom chips Google built specifically for machine learning workloads. Apple's first AFM server model was trained on a cluster of 8,192 TPU v4 chips.
The smaller on-device model, designed to run locally on Apple hardware, was built on a 2,048-chip slice of TPU v5p chips. The move gave Apple more than 2x the floating point operations per second, along with triple the high bandwidth memory. As a result, training time for large language models fell by up to 2.8 times.
How the Google relationship grew
The story starts in July 2024, when Apple revealed it had trained its initial Apple Foundation Models, known as AFM, on Google Cloud TPUs instead of the NVIDIA GPUs most of the industry was using. The shift came amid NVIDIA H100 supply issues and what the research describes as the superior performance and cost-effectiveness of Google's TPUs for Apple's needs.
In January 2026, Apple entered a multiyear agreement with Google for access to Gemini technology, a deal reportedly worth $1 billion annually. Then, in June 2026, Apple announced its third-generation AFM models, a release that marked custom development work done with Google.
Not everything runs on TPUs, however. One of the third-generation models, AFM 3 Cloud Pro, operates on NVIDIA Blackwell GPUs inside Google Cloud.
Apple maintains that no private user data was used to train its foundation models, and it says it honors publisher opt-out preferences.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
