Kepler Computing unveils new AI memory chip design after stealth period

In short: Kepler Computing, a startup backed by major chip investors, says it has a new way to build high-speed memory chips for AI using existing chip factories.

What happened

Kepler Computing, based in San Jose, California, has come out of “stealth mode” after more than seven years of quiet development. The company says it redesigned how certain memory chips are built, with a focus on high-bandwidth memory, or HBM. HBM is the fast “working shelf” of memory that AI servers rely on to move huge amounts of data.

Kepler says it can pack more memory into the same space by stacking layers in 3D, like adding extra floors to a building. The company also says it uses a proprietary material to boost memory density without relying on extreme ultraviolet lithography, or EUV, which is an expensive way to print tiny chip features using specialized light.

The company has raised $468 million from backers including GlobalFoundries, Intel Capital, AMD Ventures, Baillie Gifford, and Bill Gates through Gates Frontier. In July, the US Department of Commerce said it would provide up to $245 million to support development of a new class of AI memory technology.

Kepler says it has tested its process on about 2,000 wafers, which are the large round discs chips are made from. It plans to ship first HBM samples later this year, expand production in Singapore next year, and start US production in 2028.

Why it matters

AI systems are often limited by memory supply and cost, not just the main processor. If Kepler can scale its approach safely and reliably, it could help increase the supply of in-demand memory without waiting years and spending tens of billions of dollars to build brand-new factories.

Source: Wired