Why Nvidia has lots of room to run

As of 2025, there are approximately 1,200 hyperscale data centers worldwide, operated by the major cloud and technology providers—Amazon AWS, Microsoft Azure, Google Cloud, Meta, Apple, and a few others. While hyperscale data centers represent only a small share of total data centers globally, they are massive in scale and critical to cloud computing, AI, and intensive workloads.

Typical rack counts per hyperscale facility vary:

  • Smaller sites: 5,000 to 10,000 racks
  • Mega sites: 30,000 to 50,000+ racks

Using a conservative average of 15,000 to 25,000 racks per data center, the global hyperscale rack count is estimated at 18 million to 30 million racks.


Potential NVIDIA NVL72 Adoption

If 3% of these hyperscale racks were to migrate to NVIDIA’s new NVL72 Blackwell racks (priced at $3 million per rack), the market opportunity becomes clear.

  • 3% of global racks = 540,000 to 900,000 racks
  • NVIDIA’s capture of 80% of that migrating market = 432,000 to 720,000 racks

Revenue Projection

At $3 million per rack:

  • Total revenue = $1.296 trillion to $2.16 trillion

Assuming the migration occurs evenly over three years:

  • Annual revenue = $432 billion to $720 billion per year

This scenario suggests that even modest market penetration—just 3% of racks upgrading—could deliver hundreds of billions of dollars in annual revenue for NVIDIA over a three-year cycle. Given the push by hyperscalers like AWS, Azure, and Meta to upgrade infrastructure for AI workloads, this level of adoption is within the realm of possibility.


Discover more from Priory House

Subscribe to get the latest posts sent to your email.

Leave a Reply