Verizon and other carriers are resisting Nvidia's AI implementation plans for their networks.
Nokia is part of the game, too.
GPUs excel at AI model training in large data centers, but their use in mobile networks presents challenges. Their high power consumption, relative to standard telecom chips, raises cost concerns unless efficiency improves or revenue opportunities arise.
Nvidia promotes its ARC-Pro as a solution, a compact GPU platform for mobile sites. However, it reportedly consumes around 300 watts, significantly more than some telecom-specific chips. While Nvidia claims competitive efficiency, operators are wary due to data centers' rising energy demands.
This caution seems justified, as operators face pressure to reduce costs amid rising traffic and stagnant revenue. Every expense is scrutinized, making them hesitant to adopt solutions that could increase expenses.
Verizon's Yago Tenorio stated in November 2025 that GPU decisions should prioritize the base station's needs, not complex monetization schemes.
Some telecom executives express doubts about GPUs. Telus and Orange believe converting sites into mini data centers would require substantial upgrades to space, power, and fiber.
The primary argument for GPUs is that AI could enhance spectral efficiency. Yet, the necessity of GPUs remains questionable. Ericsson demonstrated a 10% efficiency increase using AI on regular processors. Nokia previously asserted that its existing chips were sufficient for AI tasks.
If GPUs were essential, more vendors would adopt them. Instead, many favor Intel's CPUs, which appear capable of handling 5G and 6G features without GPUs' energy demands. Vodafone's upcoming open RAN deployment will rely heavily on Intel's processors.
Even AI software developers like Cohere Technology, supported by major operators, report achieving gains using standard servers.
Currently, most operators lack a compelling reason to integrate GPUs into their networks, but AI adoption is expected to grow, particularly in customer service and proprietary applications.
Perhaps this is why Verizon plans to eliminate 15,000 positions in the coming week.
Verizon's Yago Tenorio stated in November 2025 that GPU decisions should prioritize the base station's needs, not complex monetization schemes.