By Light Reading
Publication Date: 2026-02-19 08:30:00
Swedish inscrutability makes it hard to figure out what Ericsson’s executives really think of the $1 billion deal that Nokia struck with Nvidia last year. The Finnish rival accepted investment, welcomed the giant chipmaker as its second-biggest shareholder and announced plans to design 5G and 6G radio access network (RAN) software that will run on Nvidia’s graphics processing units (GPUs), the chips more commonly used to train AI’s large language models (LLMs). But there is suspicion within Ericsson that Nokia would struggle to extricate itself from that partnership or use that AI-RAN software with anybody else’s chips.
No doubt, Nokia would reject the argument that it is now in a state of vendor lock-in. Yet Ericsson has also flatly denied that GPUs are an essential component of AI-RAN. This week in London, its pre-MWC Barcelona catwalk paraded its own initial wave of AI-RAN products, designed to work on Ericsson’s own purpose-built silicon. The company has defended its investment in silicon while others predict an industry pivot to the general-purpose processors of virtual or cloud RAN. At the same time, Ericsson looks more determined than ever to have complete software freedom – a base of algorithms and code it can deploy on any silicon platform.
On that, it looks in full alignment with service providers such as Orange. The main reason for the drive toward independence is that nobody can realistically…
