Marvell charges its R&D costs differently depending on the volume. The smaller the volume, the larger the guarantee Nokia has to pay Marvell for R&D. Nokia’s previously over-optimistic volumes have made Marvell cautious, which dilutes the economies of scale.
I don’t believe that R&D costs will decrease significantly even in the long term (5-10 years) even if the baseband switches to GPU. L1low is still at the radio end of the SoC. Existing radio frequencies will gradually be transferred to 6G, so reprogramming on old platforms will continue for a long time even if Nvidia develops radio-SoC (System-on-Chip) replacement technology for Nokia.
Finally, I don’t know if Nokia can learn anything from 5G FPGA (Field-Programmable Gate Array) other than that if things accelerate unexpectedly, it’s foolish to go to commercial markets with a lab proto, no matter how urgent the need. Only a collapse in market share will follow. So, if AI in RAN (Radio Access Network) materializes already in the 5GA era, led by TMO US, then Nokia is in a strong position.