Susana SchwartzTable of Contents
Ericsson pivot toward high-performance indoor networks and differentiated connectivity
Yesterday, Ericsson made a series of major structural, financial, and product announcements that reflect changes to its 5G and AI infrastructure strategy.
For example, the company officially launched its sovereign AI platform, Sferical One (backed by 1,152 NVIDIA Blackwell Ultra GPUs), and announced a major operational overhaul that will include consolidating regional divisions and dismantling its dedicated enterprise wireless unit, Business Area Enterprise Wireless Solutions (BEWS).
Additionally, Ericsson will partner with Verizon, having recently completed its first pre-standard 6G field call over a commercial 7 GHz site in California.
The changes come despite positive Q3 earnings, which beat expectations due to strong 5G demand. Despite that strong demand, company leaders are wary of the uptick in component costs, driven in large part by AI data center demand.
Because AI infrastructure requires massive uplink capacity and ultra-low latency, Ericsson CTO Erik Ekudden has promoted the idea of an “intelligent fabric,” which he says will connect far-flung cloud data centers, regional edge compute, and localized devices. Through custom ASICs embedded directly into the RAN base stations, he believes there will be more efficiency in data processing at the network edge. The goal is to avoid unnecessary cloud backhaul while managing the growing uplink traffic from real-time AI agents and M2M interactions.
Read more about this in our “top story,” below.
Susana Schwartz
Technology Editor
RCRTech
AI Infrastructure Top Stories
Ericsson retools in flat market: A flat RAN market remains a big challenge, says Ericsson CEO Per Narvinger, while enterprise and AI-related opps are yet to deliver 5G monetization. More of its R&D will focus on indoor enterprise environments.
Which AI-RAN investment makes sense?: According to analyst Vish Nandlall, AI-RAN is starting to produce numbers that operators can put into investment models – as with SoftBank- Ericsson tests of AI link adaptation on commercial 5G networks.
AI Today: What You Need to Know
Calculating GPU cluster cost: Jordan Nanos from SemiAnalysis and Roman Luchkov from Nebius explain how to evaluate GPU cloud providers using the ClusterMAX TCO calculator. Listen to this webinar on GPU cluster TCO.
Gartner issues capex warning: Gartner says over $3.5 trillion of AI infrastructure spending could be exposed to impairment by the end of 2030, projecting that AI data center owners will need $8.5 trillion in cumulative revenue to justify investments.
Oracle-Tencent $7B deal: Oracle signed a 5-year lease with Tencent for 100,000 AI chips. According to the Financial Times, that will equal about 4.38B GPU hours (at about $1.60 per GPU hour). Why might this matter? Read here.
Questions grow around CoreWeave debt: According to Bloomberg, CoreWeave’s debt is worrying some equity market investors, but Nvidia CEO Jensen Huang said its “strength is improving every day“ and rejected the need for a financial backstop.
Google eyes Roanoke Co.: Google is considering Roanoke County as a site for a new Virginia DC, with Google’s Clay Allsop stating, “We are committed to growing this collaboration as a long-term partner in the Roanoke region.”
RCR Events
Industry Resources
Report: NTN in motion — evolving standards, expanding services
