Making chips—electronic chips, not casino chips or edible potato chips—is not for the penny pinchers. 

Samsung estimated that in 2011, it cost about $48 million to produce a 28-nanometer chip (meaning 

each transistor is just 28 billionths of a meter). Today, the South Korean electronics giant says in a new 

white paper on mobile networks in the age of artificial intelligence (AI) that producing an ultra-advanced 

2-nanometer chip costs about $725 million. Costs have risen dramatically as designers cram more 

computing power onto chips.

This cost climb is terrifying for mobile network developers. According to Omdia, a sister company of Light

 Reading, total annual sales of radio access network (RAN) products are projected to fall by $10 billion

 between 2022 and 2024, to just $35 billion. Specialized RAN chip designers can't possibly keep up with 

the much larger general-purpose chip industry, which continues to churn out central processing units

 (CPUs) and, more recently, graphics processing units (GPUs), made famous by Nvidia. Samsung noted 

that the release cycle for specialized system-on-chip (SoC) technology in the mobile space is three to

 four years, compared to the one to two-year cycle for CPUs and GPUs.

Sachin Katti, Intel's current CTO and AI officer, once predicted that custom chips would eventually fall

 too far behind to be competitive. About three years ago, when he was CTO of Intel's Networking and 

Edge (NEX) division, he told Light Reading, "Universal technology will eventually receive so much investment

 that it will surpass custom chips."