Inflection AI, a Palo Alto-based technology company, has revealed a successful funding round amounting to $1.3 billion with prominent investors, including Microsoft (NASDAQ:MSFT), Reid Hoffman, Bill Gates, Eric Schmidt and Nvidia.
The company plans to allocate part of the new capital to the construction of a 22,000-unit Nvidia H100 Tensor GPU cluster, a project which, according to the company, would result in the world's largest GPU cluster of its kind.
Developers at Inflection AI said: "We estimate that if we entered our cluster in the recent TOP500 list of supercomputers, it would be the second and close to the top entry, despite being optimised for AI - rather than scientific - applications."
It’s a big week! We’ve raised $1.3 billion and are building the world’s largest AI cluster (22k H100s).We’re grateful for our investors and new funding that will help us accelerate our mission to make personal AI available to every person in the world. https://t.co/l2MPlhgqVl
— Inflection AI (@inflectionAI) June 29, 2023
"Pi" system
In addition to its ambitious GPU cluster project, Inflection AI is also in the process of developing a unique AI system, which it has named "Pi".
The company characterises Pi as "a teacher, coach, confidante, creative partner and sounding board".
This AI system will be accessible directly via social media or WhatsApp.
Pushing boundaries of large-scale AI models
Since its inception in early 2022, Inflection AI has amassed total funding of $1.525 billion.
And yet, as Inflection AI pushes the boundaries of large-scale AI models, some experts have cautioned that the effective training of these models could be hindered by current technological constraints.
A recent report from Singapore-based venture fund Foresight highlighted that large AI models require significant computational resources, which may strain network bandwidth and induce latency issues.
Foresight analysts also cautioned that network congestion and latency could cause the time required for data transmission to significantly exceed the estimated one second, meaning that computing nodes might spend most of their time idle, awaiting data transmission rather than executing computations.
Given these constraints, the analysts concluded that small AI models, which are easier to deploy and manage, might be the optimal solution.
They added, "In many application scenarios, users or companies do not need the universal reasoning capability of large language models but are only focused on a very refined prediction target."