IBM unveils tech for chip smaller than 1 nanometer in AI computing push (updated)

International Business Machines

25/06/2026 11:11

Updates shares in paragraph 2, adds information about SRAM memory in paragraph 9
By Anhata Rooprai and Stephen Nellis

    June 25 (Reuters) - IBM IBM.N on Thursday unveiled what it said was the world's first technology capable of producing chips smaller than one nanometer, as tech companies race to build semiconductors that can handle increasingly demanding AI workloads.
Shares of the Armonk, New York-based company rose over 6% in premarket trading, though later pared gains to about 1.9%. They have fallen about 11% so far this year.
The announcement comes at a time when chipmakers are searching for ways to maintain the decades-long trend of cramming more computing power into smaller spaces, a phenomenon known as Moore's Law.
The new chip technology, which bolsters IBM's position to compete with contract chipmakers TSMC 2330.TW and Intel INTC.O, has a transistor architecture of 0.7 nanometers, or 7 angstroms.
Last week, Intel said the new generation of its 18A manufacturing process, which makes 1.8 nanometer chips, moved into risk production, the testing phase before commercial manufacturing. 
IBM said the 0.7-nanometer chip packs nearly 100 billion transistors onto a fingernail-sized surface, about twice the density of its 2-nanometer chip unveiled in 2021, delivering up to 50% higher performance or 70% greater energy efficiency.
To get there, IBM developed a new transistor design called "nanostack." Instead of laying transistors flat, the design stacks them on top of each other in three dimensions, fitting more into the same volume of space.
"With our new nanostack architecture, we’re not just making smaller transistors, we’re reinventing how chips are built to deliver dramatically more power and energy efficiency,” director of IBM Research Jay Gambetta said.
In addition to shrinking computing circuits, IBM said the new nanostack technology will shrink a type of memory circuit called SRAM by 40%, a far better gain than IBM's previous generation of new chip technology. That category of memory is heavily used in Nvidia's NVDA.O new Groq chips and chips from Cerebras Systems CBRS.O, both of which currently rely on TSMC. 
IBM says production could begin within five years. The company has previously licensed chip technologies to Samsung 005930.KS and Japan's Rapidus. It has not announced a manufacturing partner for this technology.

 (Reporting by Anhata Rooprai in Bengaluru and Stephen Nellis in San Francisco. Editing by Varun H K, Devika Syamnath and Mark Potter)
 ((Anhata.Rooprai@thomsonreuters.com))