The Great Telecom Bifurcation: One Company Feeds the AI Boom, the Other Gets Eaten by It

The telecom industry has a simple story about the AI boom: memory prices are up, margins are squeezed, everyone is scrambling toward software automation to survive. That story is true. It is also incomplete.

Nokia reported Thursday that its AI and cloud-related net sales climbed 49 percent year-over-year in the first quarter, with comparable operating profit jumping 54 percent to 281 million euros. The companys shares rose nearly 7 percent in Helsinki trading, touching their highest level since April 2010. Reuters Nvidia took a one billion dollar equity stake in Nokia. Business Times The company raised its forecast for the AI and cloud addressable market from 16 percent annual growth to 27 percent through 2028. Reuters This is not a company getting squeezed by the AI boom.

Ericsson posted a 79 percent collapse in first quarter net income on the same day, driven by a 10 percent decline in net sales and input costs that its CEO, Börje Ekholm, explicitly linked to AI-driven semiconductor inflation. EE Times Reuters The numbers are small enough to be precise: 887 million Swedish kronor, about 97 million dollars. Ericsson Interim Report The trend line is large enough to matter. This is not a company thriving in the AI era.

The telecom sector is splitting into two cohorts in real time. Nokia sits on the infrastructure layer of hyperscaler AI buildouts, selling the optical transport systems and fiber networks that connect data centers Nvidia and its peers are filling with GPUs. Ericsson buys chips to run the networks those data centers connect to. When SK Hynix redirects roughly 20 billion dollars of its 2026 capital expenditure toward HBM3E and HBM4, that capacity does not flow to base station suppliers at a competitive price. It flows to the companies that can pay the premiums AI inference demands. DropReference

The numbers are not subtle. Spot prices for memory modules critical to base stations and consumer premises equipment have surged more than 600 percent over the past year. EE Times HP, which shares supply chain dynamics with network equipment makers, saw memory and storage climb from 15 to 18 percent of its PC bill of materials to approximately 35 percent in 2026. Sourceability Micron disclosed in its last earnings cycle that it could fulfill only 55 to 60 percent of core customer demand. Sourceability The AI boom is not a rising tide here. It is a redirect.

Global telecom capital expenditure is projected to decline 2 percent in 2026, with the capex-to-revenue ratio tracking toward 14 percent. DellOro Verizon has already moved, reducing its 2026 capex guidance to 16 to 16.5 billion dollars from 17 billion dollars in 2025. EE Times The operators are not increasing budgets to absorb memory inflation. They are demanding efficiency from their vendors, which is why automation has become the industries preferred vocabulary.

Most operators remain between Level 1 and Level 2 on the TM Forums automation maturity framework, but early adopters are now pushing toward Level 4, where systems reason, decide, and act with minimal human intervention. DellOro At that level, networks detect and correct problems before they escalate, reduce energy consumption autonomously, and optimize spectral efficiency without a human in the loop. Ericsson estimates intelligent RAN automation can improve spectral efficiency by 15 percent. Nokia has reported up to 80 percent efficiency gains with zero-touch radio optimization. ZTE demonstrated roughly 30 percent fault recovery time reduction at Level 2 and 3 automation. DellOro

This is where the 6G paradox becomes unavoidable. The industrys proposed solution to the cost squeeze involves embedding more AI capability into the radio access network itself. AI-RAN, in the framing Nvidia and Marvell are pursuing through their partnership, would allow operators to host enterprise workloads at the cell tower using specialized compute nodes. Telecom Tech News The efficiency gains are real and measurable. But the hardware required to deliver those gains competes for the same HBM capacity that is currently squeezing vendor margins. The cure requires the same expensive medicine causing the disease.

The skeptics have a case. Ericssons first quarter included headwinds from a North American market unwind that some analysts attribute partly to 2025 tariff-related pull-forward. The 79 percent net income collapse includes restructuring charges and foreign exchange pressure. Nokia remains a sub-scale player in AI and cloud at 8 percent of group revenue. The divergence could be cyclical rather than structural. Both companies are still operating in the same memory cost environment.

But cyclicals leave traces. If Nokia is winning because it owns optical transport and fiber infrastructure for hyperscaler AI campuses, that positioning does not reverse when memory prices correct. If Ericsson is losing because its wireless-heavy portfolio puts it in direct competition with Nvidia for HBM allocation, that structural tension does not resolve when AI inference demand moderates. The bifurcation is real and the dividing line is clear: who owns the infrastructure layer of the AI buildout, and who buys into it.

The 6G rollout will stress-test both positions. The equipment that makes AI-RAN work will compete for the same chips that make AI inference work. The operators spending 2 percent less on capex will demand 15 to 80 percent more efficiency from their vendors. The companies that solved that equation by building smarter software will survive. The companies that only built better hardware will not.