Kepler Computing Emerges from Stealth with $468M to Solve AI Memory Crunch

After seven years of quiet development, Kepler Computing has announced a novel approach to manufacturing memory chips that could help address one of the semiconductor industry's most pressing bottlenecks: the shortage of high-bandwidth memory (HBM) needed to power AI workloads.

The San Jose–based startup, founded in 2018 by physicists and computer scientists, claims its new architecture enables higher memory density without relying on expensive extreme ultraviolet (EUV) lithography. Instead, Kepler is combining proprietary 3D stacking methods with a novel ferroelectric composite material—a two-pronged strategy aimed at squeezing more capability out of existing chip factories.

A Different Path to Density#

Rather than pursuing the conventional route of shrinking transistors to pack more circuitry into the same space, Kepler is stacking memory components vertically and introducing materials that operate at lower voltages than standard semiconductors. The company says this ferroelectric approach achieves the density of 2–3 nanometer chips while working within existing fabrication plants—a potentially significant cost advantage.

For high-speed cache (SRAM), Kepler reports its material innovation solves density constraints through a composite material developed over 35 design iterations. For HBM—the stacked DRAM memory crucial to data centers—the 3D manufacturing technique packs more chips into a tighter footprint, allowing data to travel shorter distances and consume less energy moving between compute and memory.

According to Sasi Manipatruni, the company's chief technology officer, the breakthrough came from solving "the physics problem—the physical limitations for HBM"—and then translating that into materials science. GlobalFoundries, one of Kepler's manufacturing partners, reports that the startup converted an existing fab to production-ready status in eight months, versus the typical 24-month timeline.

Heavy Backing and Aggressive Timeline#

Kepler has raised $468 million from a roster of heavyweight backers including Intel Capital, AMD Ventures, GlobalFoundries (which invested $50 million), the private Gates Frontier fund, and British investor Baillie Gifford. In July, the U.S. Department of Commerce committed up to $245 million in support specifically for developing advanced AI memory technology using the company's 3D and ferroelectric innovations.

The company is currently testing around 2,000 wafers in collaboration with GlobalFoundries' Singapore facility, with plans to ship HBM samples later this year, scale Singapore production in 2026, and launch U.S. manufacturing in 2028.