Stealth Startup Kepler Computing Claims a Fix for the AI Memory Shortage

Kepler Computing, a San Jose, California startup that spent more than seven years in stealth mode, has emerged claiming its redesigned memory architecture can ease the global memory-chip shortage, provided it can scale production. Founded in 2018 by a team of physicists and computer scientists, the company says it has built a new architecture for high-bandwidth memory (HBM) that targets the chip supply bottlenecks constraining the computing market.

The core of the claim is a departure from how chipmakers typically pack more capability into silicon. Manufacturers generally rely on expensive extreme ultraviolet lithography (EUV) to shrink transistors and fit more technology into the same space. Kepler says its "3D stacking" approach, combined with a proprietary new material, increases density without any EUV, and can run on existing semiconductor fabrication plants. The company says it has made similar gains for SRAM, the high-speed cache memory embedded in the core of CPU, GPU, and XPU dies to cut data transfer times, which it distinguishes from HBM's stacks of separate DRAM components.

The startup has raised $468 million from prominent backers including GlobalFoundries, Intel Capital, AMD Ventures, the British fund Baillie Gifford, and Bill Gates through his Gates Frontier fund. In July, the US Department of Commerce committed up to $245 million to Kepler to "develop in the US a new class of high-performance AI memory technology, enabled by innovative 3D and ferroelectric technologies."

Much of Kepler's testing is happening in Singapore, where GlobalFoundries, a manufacturing partner and $50 million investor, operates a facility. Over the past two years Kepler has built what it calls "mini fabs," producing its memory chips alongside GlobalFoundries' 28-nanometer chips, and it has also run tests at GlobalFoundries' facilities in Burlington, Vermont.

Kepler and its investors argue the timing is favorable. The global memory shortage has underscored that building new fabs is hugely expensive and slow, and that demand for certain memory types is cyclical. In an AI-driven, data-center-heavy market, HBM has become the memory of choice, and prices have surged as supply has tightened.

The company's claims remain unverified at commercial scale, and its own framing carries the caveat that the technology must be producible in volume to matter. Kepler has not disclosed a timeline for when its memory could reach the market or how its density and cost compare with EUV-based rivals in production. The next test is whether the 3D and ferroelectric approach survives the leap from mini fabs in Singapore and Vermont to full-scale manufacturing.