A Stealth Startup Thinks It Just Hacked the Memory Shortage
| Source: Wired AI
Tags: Kepler Computing, high-bandwidth memory, HBM, chip design, GlobalFoundries, AMD Ventures, Intel Capital, CHIPS Act
Kepler Computing exits stealth with $468M raised — including $245M from the US Department of Commerce — claiming its 3D-stacking approach builds high-bandwidth memory without expensive EUV lithography, directly targeting the HBM shortage bottlenecking AI data centers.
Details
After seven years in stealth, San Jose-based Kepler Computing has emerged publicly, backed by $468M from investors including GlobalFoundries ($50M), Intel Capital, AMD Ventures, Baillie Gifford, and Bill Gates' Gates Frontier fund. The US Department of Commerce has separately committed up to $245M to Kepler under what appears to be a CHIPS Act-aligned program. Kepler's core claim is that its 3D stacking architecture and a proprietary new material can increase memory density — for both HBM and SRAM — without relying on extreme ultraviolet lithography (EUV). EUV machines from ASML cost over $300M each and are in severely limited supply, making EUV-free paths to density gains strategically valuable. The company says its approach works with existing 28nm GlobalFoundries fabs, which lowers the barrier to production. Current testing runs in Singapore at GlobalFoundries' facility and in Burlington, Vermont. The startup has been building 'mini fabs' and producing chips in conjunction with GlobalFoundries' existing processes. If the claims hold at production scale, Kepler could meaningfully ease the HBM supply crunch that has driven up AI infrastructure costs and slowed GPU deployments at hyperscalers. The big uncertainty: scaling from mini fabs to volume production is the hardest part of semiconductor manufacturing.