Ericsson is emerging as a preferred option for radio access network (RAN) chip suppliers as industry analysts express doubt regarding the viability of competitors' hardware strategies. Concerns about the practicality of using Nvidia graphics processing units and Intel central processing units in RAN infrastructure are helping Ericsson justify its development of proprietary silicon.
Nvidia has invested heavily in the telecommunications sector, providing $1 billion to Nokia and $2 billion to Marvell Technology, a chipmaker for Nokia. The company aims to expand the market for its GPUs, which are primarily used in data centers for training artificial intelligence models. However, skepticism persists about whether these high-power chips will be widely adopted in RAN systems.
Telcos currently spend approximately $35 billion annually on RAN products, a decline from a peak of $45 billion in 2022. Both Omdia and Dell'Oro Group, which track this market, do not anticipate growth in the current year. Dell'Oro analyst Stefan Pongratz stated that the concept of AI-RAN, promoted by Nvidia, will not significantly increase operator spending or drive purchases of GPU-based RAN products.
Pongratz projected that the GPU RAN market will reach over $1 billion by 2030. This figure represents less than 3% of total RAN spending annually. The analyst noted that predictions can be revised upward, but the current forecast suggests limited immediate impact.
Nvidia is reportedly developing lighter GPUs suitable for RAN baseband units and radios to address power consumption issues. Gabriel Foglander, Ericsson's head of RAN strategy, explained that GPUs designed for RAN would have different properties than those used in data centers. Ericsson is already marketing AI-RAN features supported by its custom silicon, which aim to improve spectral efficiency and throughput.
