Particle.news
Download on the App Store

Nvidia Tightens Hold on AI by Locking Customers Into Its Software and Systems

A two-decade-old software platform and an expanding hardware stack are turning AI spending toward chips, data-center systems, and long multi-year orders that favor incumbents.

Overview

  • Reporting shows Nvidia’s CUDA software, used for about 20 years by researchers and engineers, creates large switching costs because developers and tools are built to run on Nvidia hardware.
  • Nvidia’s Blackwell GPUs are the current data-center standard and the company says its next-generation Vera Rubin platform will begin shipping this year, reinforcing a one-stop hardware-and-software offering.
  • CEO Jensen Huang has publicly guided that Nvidia expects roughly $1 trillion in orders for Blackwell and Vera Rubin across the 2026–2027 period, a projection that has pushed Wall Street revenue models sharply higher.
  • Analysts now model steep revenue growth and a multitrillion-dollar market value for Nvidia while competitors such as Broadcom are shifting from networking chips to custom ASICs to try to capture parts of the infrastructure opportunity.
  • The broader shift from consumer-facing models to physical AI infrastructure—chips, rack-level systems, power and cooling—raises demand for data-center engineering and benefits public infrastructure suppliers more than consumer app makers.