Palo Alto-based startup Inflection AI has successfully raised $1.3 billion led by Microsoft, Reid Hoffman, Bill Gates, Eric Schmidt, and Nvidia.
As part of the plan, the new funding will be allocated to build the Nvidia H100 Tensor GPU cluster consisting of 22,000 units, which the company claims to be the largest in the world. These GPUs will be used to develop large-scale artificial intelligence models, estimated to achieve 22 exaFLOPS at 16-bit precision and potentially even more with lower precision.
Additionally, Inflection AI is also developing a personal AI assistant system called “Pi.” The company explains that Pi is “a teacher, coach, empathetic friend, creative collaborator, and soundboard” that can be accessed directly through social media or WhatsApp.
Inflection AI has successfully raised a total of $1.525 billion since its inception in early 2022 from prominent investors both within and outside the industry. Notably, the company boasts a team of top experts in the field of artificial intelligence who have previously worked at DeepMind, Google, Microsoft, OpenAI, and Meta.
Jensen Huang, the founder and CEO of NVIDIA, has expressed his support for Inflection AI’s “groundbreaking work,” acknowledging that the world-class team at Inflection AI is helping the company become a pioneer in this breakthrough endeavor.
Despite the increasing investment in AI, experts have warned that the practical training effectiveness of AI models may be severely limited by current technology.
In an example provided by Singapore’s Foresight Ventures, researchers cited a case of a large AI model with 175 billion parameters and a storage size of 700GB.
Assuming we have 100 computing nodes and each node needs to update all parameters in each step, it would require transmitting approximately 70TB of data per step (700GB * 100). If we optimistically assume that each step takes 1 second, then we would need to transmit 70TB of data per second. This bandwidth requirement exceeds the capabilities of most networks.