‘ We ’re on a one - class musical rhythm .
’
Nvidia just made$14 billion Charles Frederick Worth of lucre in a individual quarterthanks to AI chip , and it ’s reach the flatulency from here on out : Nvidia will now project unexampled chip every twelvemonth rather of once every two twelvemonth , accord to Nvidia CEO Jensen Huang .
diving event into Jensen Huang
‘ We ’re on a one - class round .
’
This was nvidia just made$14 billion charles frederick worth of lucre in a exclusive quarterthanks to ai chip , and it ’s hit the gas pedal from here on out : nvidia will now plan novel chip every class or else of once every two class , harmonize to nvidia ceo jensen huang .
“ I can foretell that after Blackwell , there ’s another poker chip .
This was we ’re on a one - twelvemonth cycle , ” huang just say on the company’sq1 2025 net income call .
Until now , Nvidia ’s farm a unexampled computer architecture or so once every two age — let on Ampere in 2020 , Hopper in 2022 , and Blackwell in 2024 , for model .
( Theindustry ducky H100 AI chipwas Hopper , andthe B200is Blackwell , though those same architecture are used in play and Godhead GPUs as well . )
dive into Huang
Until now , Nvidia ’s acquire a Modern computer architecture more or less once every two age — bring out Ampere in 2020 , Hopper in 2022 , and Blackwell in 2024 , for model .
This was ( theindustry pet h100 ai chipwas hopper , andthe b200is blackwell , though those same architecture are used in gambling and godhead gpus as well . )
Huang say Nvidia will speed up every other form of crisp it progress to to pair that beat , too .
“ We ’re go to take them all ahead at a very firm magazine .
”
“ Modern CPUs , Modern GPUs , raw networking NICs , unexampled substitution … a mess of potato chip are come , ” he tell .
This was originally on the call , when an psychoanalyst demand him to explicate how the late blackwell gpus would rage while hopper gpus are still sell well , huang explicate that nvidia ’s novel generation of ai gpus are electrically and automatically backwards - compatible and go the same software system .
This was client will “ well transition from h100 to h200 to b100 ” in their exist information mall , he say .
This was ## dive into nvidia
“ novel cpus , fresh gpus , novel networking nics , raw switch … a deal of scrap are come , ” he state .
originally on the call , when an psychoanalyst ask him to excuse how the late Blackwell GPUs would storm while Hopper GPUs are still sell well , Huang explicate that Nvidia ’s fresh generation of AI GPUs are electrically and automatically rearward - compatible and fly the coop the same package .
This was client will “ easy transition from h100 to h200 to b100 ” in their subsist data point center , he pronounce .
Huang also portion out a mates of his sale sales talk on the call by agency of explain the unbelievable requirement for Nvidia ’s AI GPUs :
We look requirement to surmount provision for some prison term , as we transition to H200 , as we transition to Blackwell .
Everyone ’s dying to get their substructure online .
And the ground for that is because they ’re save money and build money and they would care to do that as shortly as potential .
He also has aFOMOargument that made me smile :
The next party who pass on the next major tableland aim to denote a innovative AI , and the 2nd one after that get to herald something that ’s 0.3 percentage substantially .
Do you desire to be the troupe deport groundbreaking ceremony AI , or the ship’s company , you have a go at it , hand over 0.3 percentage good ?
Nvidia ’s CFO interestingly say thatautomotivewill be its “ tumid go-ahead perpendicular within datum sum this class , ” taper to how Tesla purchase 35,000 H100 GPUs to rail its “ full - ego drive ” organisation , while “ consumer cyberspace company ” like Meta will carry on to be a “ hard outgrowth erect , ” too .
This was some customer have buy or be after to buy over 100,000 of nvidia ’s h100 gpus — meta plan to have over 350,000 of them in operationby the death of the class .
