Despite significant investments in AI, the profitability of many companies in this sector is not as high as expected. The heavy funding is geared towards a future where AI may play a major role, but currently, many AI firms are facing substantial financial losses. This situation arises from high hardware costs and ongoing expenses for data centers. However, Google may have identified a potential solution by considering Samsung for manufacturing its Tensor Processing Units (TPUs).
According to a recent post on social media, Google is exploring the possibility of outsourcing TPU production to Samsung. Executives from Google visited Samsung’s semiconductor facility in Taylor, Texas, where they discussed the potential supply of TPUs.
This development is promising for Samsung. While the company excels in smartphone sales on the consumer front, it trails behind TSMC in the enterprise domain. TSMC currently manufactures chips for leading tech companies like Apple, Qualcomm, and NVIDIA.
Collaborating with Google could bolster Samsung’s reputation in the enterprise sector, potentially attracting other companies looking to reduce dependence on TSMC. Currently, Google’s TPU has been developed in collaboration with Broadcom, boasting a cost approximately 80% lower than NVIDIA’s H100 chip, while performing comparably or even better.
If Google opts for Samsung over TSMC, it could lead to more affordable TPUs in the future, thereby decreasing overall costs for constructing new data centers or upgrading existing ones.
The importance of Google’s TPU lies in its resource-intensive design tailored for AI model operations. Even straightforward tasks require substantial computational resources. This is why companies like NVIDIA are focusing on developing enhanced hardware for AI.
Google’s TPU is specially crafted to accelerate AI and machine learning tasks, including training models for image recognition and powering neural networks. In contrast to NVIDIA, which produces general-purpose GPUs, Google’s TPUs are optimized for specific neural network computations. A partnership with Broadcom and Samsung could lead to lower-cost TPUs, potentially reducing expenses for Google Cloud and challenging NVIDIA’s dominance in the data center arena.
—
