Nvidia Drives AI Capacity Build-Out; Rival Vendors Scramble for Share

Nvidia's dominance in GPU supply is not creating a monopoly; instead, it is triggering a bottleneck that is forcing enterprise data center operators and cloud providers to diversify across multiple vendors and specialized chipmakers. CoreWeave, a bare-metal cloud specialist backed by Nvidia capital and expertise, has become the focal point of this diversification play. Nvidia's founder Jensen Huang stated that 'if we didn't help CoreWeave exist, they would not exist,' signalling that even Nvidia sees the need for multiple AI infrastructure options to sustain the capex cycle.

Arm Holdings and Broadcom are benefiting from the design-win race as OEMs seek faster time-to-market for custom AI accelerators. AMD is positioning its MI300 and MI400 chips as credible Nvidia alternatives, with CoreWeave CEO hinting that Nvidia risks losing customers if it does not expand memory and compute capacity faster. Smaller infrastructure players like Iren (CoreWeave parent) and NBist are gaining traction as they position themselves as neutral third-party builders in the AI compute stack. Wells Fargo recently raised its price target on CoreWeave, validating the infrastructure-diversification thesis.

The implication is a broadening of the AI capex cycle beyond pure Nvidia beneficiaries. While Nvidia remains the clear leader, the ecosystem is maturing into a multi-vendor environment. This favours semi-niche players with differentiated architectures (Broadcom, Marvell, Xilinx competitors) and infrastructure-as-a-service platforms that can aggregate compute across multiple chip sources. Capital intensity is rising across the board: each of the major cloud providers (Amazon, Google, Microsoft) is building custom silicon and partnerships to de-risk Nvidia dependency.

Risk: If generative AI adoption plateaus or capex discipline tightens, the infrastructure build-out stalls and specialist vendors face margin compression. However, the current consensus is that AI inference and fine-tuning will sustain data center demand for 3 to 5 years, justifying current capex outlays.