Google is continuing to push its AI tensor processing units (TPU) and compete with NVIDIA as the industry grapples with whether to use NVIDIA’s expensive AI GPUs or rely on the relatively cheaper in-house AI chips. Amazon’s shares closed 9.8% lower after its second-quarter earnings saw cloud computing revenue drop, and analysts attribute the fall to the firm’s decision to rely on its in-house Trainium AI chips rather than NVIDIA’s AI GPUs. Now, a report from The Information suggests that Google is approaching smaller cloud computing providers to also provide its TPU chips alongside NVIDIA’s AI GPUs.

Google Wants Its AI TPUs In Data Centers Alongside NVIDIA’s GPUs, Says Report

Today’s report from The Information follows a report in June that claimed that OpenAI was using Google’s TPUs to power its ChatGPT and other AI products. The publication quoted a single source for this report, and even if it is true, it is likely that OpenAI shifted a small fraction of its compute to Google’s products. The Information added that the shift was due to the high costs associated with NVIDIA’s products, as it outlined that Google was seeking other cloud providers to include its TPU chips in their offerings.

Now, the publication builds on its earlier report. This time it quotes seven sources to outline that Google is approaching small cloud computing infrastructure providers to ask them to place its TPUs alongside NVIDIA’s AI GPUs in data centers. These computing companies provide infrastructure to AI software firms for their computing needs, and Google’s purported moves appear to be aimed at creating demand for its in-house chips.

However, The Information adds that competing with NVIDIA might not be the only reason Google is pushing its TPUs to cloud providers. Another reason could be that while Google has an adequate supply of chips, it is unable to build out data center infrastructure rapidly enough to accommodate the GPUs. As a result, the firm might be interested in relying on external infrastructure equipped with the TPUs for in-house AI computing needs.

The discussion surrounding NVIDIA’s AI GPUs and big tech’s attempts to rely on in-house chips intensified after Amazon’s latest earnings report. The results sent the shares lower by nearly 10% as the firm grappled with investor expectations about cloud computing growth and whether Amazon’s attempts to rely on in-house Trainium chips were lowering demand.

A fresh report from New Street outlines that these problems are persisting as it notes that “AWS still faces adoption challenges and we heard that Anthropic (the main Trainium user at scale) would rather have GPUs.” Amazon and Anthropic are key partners as the latter provides the former with a foundational AI model through which it can offer AI services to businesses.

While the cost benefits of in-house chips are attracting big tech, NVIDIA continues to assert that its chips provide the best performance.

https://wccftech.com/nvidia-is-now-battling-for-ai-chip-dominance-on-two-fronts-as-google-steps-up-efforts-to-potentially-replace-its-chips-report/amp/