NVIDIA Faces Pricing Pressure as Meta Rents Out GPUs

Meta is shifting from a loyal buyer of NVIDIA GPUs to a direct competitor in the data center market. The company announced plans to rent out its own computing power, which changes the dynamic for cloud infrastructure buyers. This move signals that major tech firms are no longer satisfied with relying solely on NVIDIA for their AI needs. Buyers should watch how this competition affects pricing and availability in the coming years.

Meta shifts from buyer to competitor by renting out computing power

NVIDIA reported $215.9 billion in full-year revenue for fiscal 2026. The data center segment drove $193.7 billion of that total, representing nearly 90 percent of the company's business. This massive revenue stream has historically relied on a small group of enterprise customers. Microsoft, Google, Amazon, and Meta accounted for roughly half of that data center income over the last three years.

Those same cloud providers are now building their own silicon to reduce dependency on NVIDIA. Google uses Tensor Processing Units, Amazon deploys Trainium chips, and Microsoft has introduced the Maia 100. Meta is rolling out its own MTIA chips for internal workloads. OpenAI is also developing custom inference chips, while Apple works with Broadcom on AI hardware. These efforts represent a significant diversification of the AI hardware landscape.

Current custom chip designs focus primarily on inference tasks rather than training models. NVIDIA still dominates the training sector due to its high compute density and mature software ecosystem. However, the shift toward in-house inference solutions reduces the volume of GPUs these companies need to purchase. This trend creates potential pressure on NVIDIA's pricing power as customers balance GPU purchases with internal development.

NVIDIA faces a complex challenge as its largest customers simultaneously buy its products and build alternatives. The company must maintain its technological lead in training workloads to retain market share. Cloud providers are unlikely to abandon NVIDIA entirely given the current software advantages. The long-term impact will depend on how quickly custom chips can match NVIDIA's performance in critical areas.