Skip to main content

Google Partners With Anthropic

The AI boom continues as AI labs partner with cloud providers in a race towards AGI
Created on February 6|Last edited on February 6
Google Cloud has partnered with Anthropic, and according to the Financial Times, invested 300 million in the company [1]. Google also announced their intentions to build next generation Tensor Processing Unit (TPU) clusters that will power Anthropic’s AI workloads. Google seems to be taking similar steps to OpenAI in terms of responsible development of AI that is aligned with human intentions, and according to James Manyika, the company is “applying responsible principles grounded in human values and safety.” Anthropic launched in 2021, and the founding team has a rich history of developing systems like GPT-3 and Reinforcement Learning from Human feedback (core technologies of ChatGPT). The company launched a ChatGPT competitor in January called Claude, which is also based on Transformers.

Google Versus Microsoft

It should be interesting to see how far the Anthropic partnership continues into the future, as it seems OpenAI and Microsoft are committed to long term AI development. Google does seem to have some benefits over Microsoft in terms of core technology, as Googles proprietary TPU offers cost savings of 35%-50% savings compared to Microsoft powered NVIDIA A100’s in the Azure cloud, and there is a strong likelihood that the next generation TPU’s will improve this margin even more [2]. This simple advantage could lead to a pricing battle between Claude and ChatGPT, where Claude most likely will have the upper hand in terms of operating costs. However, it’s unclear whether Claude will able to compete with the dynamic and sophisticated ChatGPT system. Along with this, it seems as though reducing model inference costs will be a top priority for Microsoft going forward due to ChatGPT’s high usage rates, so a competitor to Googles TPU would not be surprising to see from Microsoft or another player like NVIDIA.

Original Article:

References