IBM (IBM) hasn’t built a commercial computer chip of its own in nearly a decade. On Thursday, June 25, the company said its research labs had cracked something the rest of the chip industry hasn’t managed yet.The breakthrough has nothing to do with software, and everything to do with one tiny component.IBM exited semiconductor manufacturing years ago and stopped running its own factories. Instead of building fabs, it kept a research lab in Albany, New York, working on what comes after today’s leading chip designs, according to Fast Company.That bet is why a company best known now for software and consulting just out-engineered TSMC (TSM), Samsung, and Intel (INTC) on the most basic measure of chip progress: how small a transistor can get.New IBM chip packs 100 billion tiny transistors onto a fingernailIBM’s new architecture, called nanostack, builds transistors in three dimensions instead of spreading them flat across a chip’s surface, according to an IBM press release.The result is a 0.7 nanometer, or 7 angstrom, chip, the first technology to operate below the 1 nanometer mark, IBM said.Related: IBM stock just got powerful new price target from Wall StreetIBM Research Vice President Huiming Bu described it to Fast Company as work that happens “between magic and physics.”IBM says the chip can deliver up to 50% more performance or 70% greater energy efficiency than its current 2-nanometer chips, citing its own published technical results.The design also shrinks the memory cells used for AI processing by 40%, based on research IBM presented at the VLSI 2026 conference. Those are lab results, not shipped products, but they explain why the announcement moved markets.
IBM unveiled a sub-1-nanometer chip architecture that could extend Moore’s Law and reshape AI computing costs.Andrew Harnik / Getty Images
7 angstrom chip’s performance math reads like science fictionIBM researchers estimate that AI accelerators built on a 7 angstrom design could run about six times faster than today’s chips, according to the IBM Research blog.That would push AI chip output toward 9,000 trillion operations per second, up from roughly 1,500 today, IBM Research said.At that rate, the time needed to train a frontier AI model could shrink from about three months to two weeks, the same research found.Wall Street moved on IBM chip news before breakfastNone of this came from an earnings report. IBM shares jumped as much as 6% in premarket trading on June 25, according to a Seeking Alpha report. That’s a large move for a century-old technology company on a day with no scheduled financial results.More IBM:IBM and Google team up to launch stunning product for customersWedbush resets IBM stock price target on AI momentumIBM CEO sends strong message on quantum computingInvestors weren’t reacting to revenue. They were reacting to proof that IBM’s research division, the part of the business that rarely shows up clearly in quarterly numbers, still produces things with commercial weight.IBM said it expects sub-1-nanometer chips to reach production within five years, the same statement noted, a timeline that matters more to investors than the lab demo itself.The chip news also didn’t land in isolation. JPMorgan upgraded IBM to overweight just two days earlier, raising its price target to $291 from $270 on the strength of IBM’s software margins.That came the same week President Donald Trump signed executive orders accelerating quantum computing research, an event IBM CEO Arvind Krishna attended in person.By June 25, the chip announcement was the third bullish catalyst to hit the stock in a matter of days.A few details didn’t make the main story but shape how seriously to take it.IBM no longer runs its own chip fabs and instead licenses designs to manufacturing partners, rather than producing chips itself.TSMC, Samsung, and Intel are already mass producing chips based on IBM’s earlier 2-nanometer design, but IBM hasn’t named a manufacturing partner for the new node yet.IBM is separately spinning up Anderon, a standalone company focused on manufacturing quantum computing wafers, according to the newsroom statement.The real race is over who controls the next decade of computingIBM’s announcement lands as the AI industry strains against a shared problem: Chip performance isn’t keeping pace with demand for computing power.Dan Hutcheson, an analyst at TechInsights, said the new architecture could extend the chip industry’s roadmap by 10 to 15 years, easing the conflict between AI’s appetite for compute and its energy cost, MIT Technology Review reported.That tension, more than any single chip, is what has been moving semiconductor stocks all year. IBM just gave the market a reason to think the physics problem has more runway than feared.Whether IBM turns a lab breakthrough into an actual manufacturing partnership, the way it has before with foundry partners on earlier chip generations, will decide if the June 25 stock move holds or fades.The lab result is real. The business model behind it is still being written. What happens next may say more about IBM’s future than the chip itself.Related: IBM handed two major wins within 24 hours