Morgan Stanley: What does OpenAI's embrace of TPU mean for Google, NVIDIA, and Amazon?
Morgan Stanley pointed out that OpenAI plans to use Google's TPU for AI inference tasks, and this collaboration is seen as a significant recognition of Google's AI infrastructure capabilities, highlighting its leading position in the custom AI chip field. Although OpenAI has denied plans for large-scale deployment of TPU, this choice may drive growth in Google's cloud business and enhance market confidence in its search technology advantages. At the same time, the tight supply of NVIDIA GPUs has also prompted OpenAI to consider TPU as an alternative
According to Zhitong Finance APP, Morgan Stanley published a research report stating that recent reports indicate OpenAI plans to use Google's (GOOGL.US) TPU (Tensor Processing Unit) for AI inference tasks. The firm believes this collaboration will become a landmark event in the AI chip field and will have a profound impact on the market landscape of tech giants such as Google, NVIDIA (NVDA.US), and Amazon (AMZN.US).
The report pointed out that OpenAI is reaching an agreement with Google Cloud to rent its TPU to support AI inference workloads. Although a spokesperson for OpenAI later denied this report, stating that there are currently no plans for large-scale deployment, they noted that the company is conducting early testing on some of Google's TPUs.
Morgan Stanley stated that although OpenAI cannot use Google's most advanced TPUs (which are reserved for training the Gemini model), this collaboration is still seen as a significant recognition of Google's AI infrastructure capabilities, highlighting Google's leading position in the field of custom AI chips (ASIC).
As Google's self-developed AI hardware for ten years, TPUs have been in use since 2015, with well-known clients such as Apple (AAPL.US) and Safe Superintelligence. Morgan Stanley believes that OpenAI's choice is a "major endorsement" of Google's AI infrastructure capabilities, which may not only accelerate the growth of Google Cloud's business but also enhance market confidence in its technological advantages in the search field.
Morgan Stanley also pointed out that part of the reason OpenAI chose TPU may be due to the tight supply of NVIDIA GPUs. The report stated that NVIDIA's "rack-level products are sold out," while Google's TPU provides a viable alternative. Although NVIDIA still dominates the AI training market (with its revenue at Google expected to triple to over $20 billion by 2025), the surge in demand for inference may prompt companies to explore other options.
On the other hand, this news indicates that OpenAI's AI workloads have covered Google Cloud, Microsoft (MSFT.US) Azure, Oracle (ORCL.US), and CoreWeave (CRWV.US), but Amazon AWS did not appear on the collaboration list. The firm speculated that this may indicate AWS is facing capacity constraints and cannot meet OpenAI's demands. Additionally, OpenAI's choice to use Google's older TPUs instead of AWS's self-developed chip Trainium indirectly reflects its lower technical recognition compared to TPUs.
Overall, Morgan Stanley maintains an "Overweight" rating on Google, with a target price of $185. The firm believes that the potential growth of Google Cloud's business has not been fully reflected in the stock price, and the commercialization of TPU is expected to become a new valuation catalyst
