The California startup just landed up to $245 million in CHIPS Act funding for a ferroelectric memory technology it says can outperform current solutions by 15-20x.
The AI boom has a bottleneck problem, and it’s not compute. It’s memory. DRAM prices doubled in Q1 2026, high-bandwidth memory is sold out through at least 2027-2028, and every major chipmaker is scrambling for wafer capacity. Into this chaos walks Kepler Computing, a Moraga, California startup claiming its ferroelectric RAM technology can fundamentally change how chips handle data.
The US Department of Commerce apparently agrees with the pitch. On July 29-30, the agency unveiled a letter of intent to award Kepler up to $245 million under the CHIPS Act, earmarked for research and development of high-performance AI memory. The funding is part of a broader $874 million allocation spread across seven companies, with the government set to take a minority equity stake as part of the deal.
What Kepler is actually building #
Traditional chip architectures shuttle data back and forth between memory and processors, a design limitation known as the von Neumann bottleneck. Kepler’s approach stacks ferroelectric RAM tiles directly on top of computational logic chips using 3D heterogeneous integration. The company’s ferroelectric RAM uses materials like hafnium zirconium oxide to store data, and peer-reviewed research published in April 2026 demonstrated that this architecture can be manufactured using existing CMOS processes.
Kepler claims the resulting compute-in-memory design achieves matrix multiplication speeds 15-20x faster than conventional SRAM and DRAM baselines while consuming orders of magnitude less power. The technology is specifically optimized for inference workloads, the phase where trained AI models actually process new data and generate outputs.
The memory crisis driving the urgency #
HBM production consumes nearly three times more wafer capacity than standard DRAM for equivalent output. The result: DRAM prices doubled in Q1 2026. HBM supply is projected to remain sold out through at least 2027-2028. Samsung, SK Hynix, and Micron, the three companies that dominate global memory production, are all racing to expand HBM capacity, but new fabrication capacity takes years to bring online.
Kepler’s pitch is that ferroelectric compute-in-memory can deliver superior bandwidth density and lower power consumption compared to conventional HBM stacks, potentially offering an alternative path for AI chip designers who can’t secure enough HBM allocation.
A well-funded outsider with serious backers #
Founded in 2018, Kepler has accumulated over $660 million in total funding. A Series D round of approximately $391 million closed in mid-2025, pushing the company’s valuation to around $1.9 billion. The investor list includes AMD Ventures and Baillie Gifford, the Edinburgh-based asset manager known for early, concentrated bets on high-growth technology companies.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our