Over a century in the past, Henry Ford pioneered the mass manufacturing of vehicles and engines to supply transportation at an inexpensive worth. Right now, the expertise business manufactures the engines for a brand new form of manufacturing unit — those who produce intelligence.
As firms and international locations more and more concentrate on AI, and transfer from experimentation to implementation, the demand for AI applied sciences continues to develop exponentially. Main system builders are racing to ramp up manufacturing of the servers for AI factories – the engines of AI factories – to fulfill the world’s exploding demand for intelligence and progress.
Dell Applied sciences is a frontrunner on this renaissance. Dell and NVIDIA have partnered for many years and proceed to push the tempo of innovation. In its final earnings name, Dell projected that its AI server enterprise will develop not less than $15 billion this yr.
“We’re on a mission to carry AI to hundreds of thousands of shoppers all over the world,” mentioned Michael Dell, chairman and chief govt officer, Dell Applied sciences, in a current announcement at Dell Applied sciences World. “With the Dell AI Manufacturing facility with NVIDIA, enterprises can handle the whole AI lifecycle throughout use instances, from coaching to deployment, at any scale.”
The newest Dell AI servers, powered by NVIDIA Blackwell, supply as much as 50x extra AI reasoning inference output and 5x enchancment in throughput in contrast with the Hopper platform. Clients use them to generate tokens for brand spanking new AI purposes that may assist remedy a number of the world’s largest challenges, from illness prevention to superior manufacturing.
Dell servers with NVIDIA GB200 are transport at scale for quite a lot of clients, resembling CoreWeave’s new NVIDIA GB200 NVL72 system. One in every of Dell’s U.S. factories can ship hundreds of NVIDIA Blackwell GPUs to clients in per week. It’s why they have been chosen by certainly one of their largest clients to deploy 100,000 NVIDIA GPUs in simply six weeks.
However how is an AI server made? We visited a facility to search out out.
Constructing the Engines of Intelligence
We visited certainly one of Dell’s U.S. services that builds essentially the most compute-dense NVIDIA Blackwell era servers ever manufactured.
Fashionable car engines have greater than 200 main parts and take three to seven years to roll out to market. NVIDIA GB200 NVL72 servers have 1.2 million components and have been designed only a yr in the past.
Amid a forest of racks, grouped by phases of meeting, Dell workers shortly slide in GB200 trays, NVLink Swap networking trays after which take a look at the methods. The corporate mentioned its means to engineer the compute, community and storage meeting beneath one roof and positive tune, deploy and combine full methods is a robust differentiator. Velocity additionally issues. The Dell crew can construct, take a look at, ship – take a look at once more on website at a buyer location – and switch over a rack in 24 hours.
The servers are destined for state-of-the-art knowledge facilities that require a dizzying amount of cables, pipes and hoses to function. One knowledge middle can have 27,000 miles of community cable — sufficient to wrap across the Earth. It could possibly pack about six miles of water pipes, 77 miles of rubber hoses, and is able to circulating 100,000 gallons of water per minute for cooling.
With new AI factories being introduced every week – the European Union has plans for seven AI factorieswhereas India, Japan, Saudi Arabia, the UAE and Norway are additionally creating them – the demand for these engines of intelligence will solely develop within the months and years forward.
