Kepler Computing Raises $468M to Solve AI Memory Bottleneck
Original: A Stealth Startup Thinks It Just Hacked the Memory Shortage
Why This Matters
A fab-agnostic HBM alternative could ease AI chip supply constraints without requiring new EUV-dependent manufacturing lines.
Stealth chip startup Kepler Computing emerged from seven years of secrecy with a new 3D-stacking architecture for high-bandwidth memory (HBM) and SRAM that avoids costly EUV lithography. The San Jose-based firm has raised $468M from Intel Capital, AMD Ventures, GlobalFoundries, and Bill Gates, plus up to $245M from the US Department of Commerce.
Kepler Computing, founded in 2018 by physicists and computer scientists in San Jose, California, exited stealth mode after more than seven years of development. The company claims its proprietary 3D-stacking approach and a new ferroelectric material can increase memory density without relying on extreme ultraviolet (EUV) lithography—an expensive process used by most leading chipmakers. The technology targets both high-bandwidth memory (HBM), which stacks DRAM for AI accelerators, and SRAM, the high-speed cache embedded in CPUs, GPUs, and XPUs.
Kepler has raised $468 million in total funding. Backers include GlobalFoundries ($50M), Intel Capital, AMD Ventures, UK fund Baillie Gifford, and Bill Gates via his Gates Frontier fund. In July 2026, the US Department of Commerce committed up to $245 million for Kepler to develop AI memory technology using 3D and ferroelectric processes domestically.
Current manufacturing is taking place primarily at GlobalFoundries' Singapore facility and in Burlington, Vermont, using 28-nanometer processes in dedicated "mini fabs." The startup says its approach is compatible with existing semiconductor fabs, reducing the need to build costly new facilities. The announcement comes as SK Hynix and Micron race to construct multibillion-dollar HBM fabs to meet surging AI-driven demand.