We have been re-introduced to the idea of AI factories, that are primarily gigantic information facilities and the place, Nvidia’s world, tokens generated are equal to income: the presentation talked about a worth, $1 trillion.
The price of constructing information facilities, like mega-farms, will increase exponentially each era as they get extra advanced, and require extra energy to accommodate much more compute.
A prototype of the Newton robotic in Nvidia’s Isaac lab (Picture credit score: Nvidia)
The lofty Stargate project has a finances of $500 billion and Apple has already dedicated to spending $500 billion over the following few years, a big chunk of which might be on information centres.
Relying on what analysts you speak to, the worldwide information centre capital expenditure is anticipated to succeed in $1 trillion both in 2027 (PWC) or 2029 (Dell ’Oro).
So that offers you a measure of Nvidia’s extraordinarily bold targets for itself and for humanity.
An enormous piece of that puzzle that Nvidia launched on the keynote is its Photonics swap methods which is able to enable these AI factories to “scale to hundreds of thousands of GPUs”, seemingly inside the similar bodily perimeter.
A Nvidia GB300 NVL72 server rack (Picture credit score: Nvidia)
Massive, enormous numbers
The most important quantity of all of them on this presentation was when Rev Lebaredian Nvidia’s VP, Omniverse and Simulation Expertise, introduced a slide about how Bodily AI is remodeling $50 trillion industries.
These span throughout two billion cameras, 10 million factories, two billion automobiles and, Lebareian added ‘Future Billion humanoid robots’.
Sure, it’s not nearly robots, however humanoid robots, hundreds of thousands of them. Robots, on the whole, have been round for many years in lots of shapes and varieties.
Nvidia, nonetheless, envisions a world the place billions of robots are constructed on what it calls, its “Three Computer systems”: Pre and post-training (achieved through DGX), simulation and artificial information era (achieved through Omniverse) and Runtime (achieved utilizing AGX).
The world of the longer term could also be house to “a number of sorts of humanoid robots” that can have the ability to post-train – whereas on the job so to talk – with actual or artificial information, permitting them to be taught new abilities or improve their knowledgebase on the fly, actually at (or close to) the velocity of sunshine.
And the primary of them might be based mostly on its new GR00T N1 humanoid robotic basis mannequin, developed in partnership with Google Deepmind and … Disney Analysis.
(Ed: One cannot see the tie-in with GR00T from Guardians of the Galaxy and Wall-E, a doubtlessly prescient film in regards to the affect of robotics and AI on humanity).
The place does this go away us? With a transparent roadmap to a world the place AI turns into an indispensable a part of what it means to be human, maybe crucial utility of all of them. By no means, within the historical past of humanity, will so many rely upon so few.
Oh and the opposite acronym (AGI) was not talked about in the course of the presentation, not as soon as.
You may additionally like 
