Kepler Secures Intel Capital and AMD Ventures Backing as New AI Memory Chip Route Emerges; Micron Stock Weakens

TradingKey - As demand for high-bandwidth and high-capacity memory in AI servers continues to grow, a chip startup backed by Intel (INTC) Capital, AMD (AMD) Ventures, GlobalFoundries, and other institutions is attempting to develop a new AI memory solution beyond HBM.

On September 9, Eastern Time, chip startup Kepler Computing emerged from over seven years in stealth mode to officially unveil its AI memory technology. Kepler stated that the solution combines 3D integration and innovations in ferroelectric materials, allowing it to be produced in existing fabs and reducing reliance on EUV lithography equipment, while increasing memory density and energy efficiency.

[Source: X]

What Kind of Company Is Kepler?

Kepler Computing is a chip startup headquartered in San Jose, California, founded in 2018 by Intel veterans. The company focuses on the R&D of novel AI memory architectures, aiming to break through the density bottlenecks of HBM and SRAM on mature process lines.

To date, the company has raised a total of $468 million, with investors including Intel Capital, AMD Ventures, GlobalFoundries, Baillie Gifford, and Bill Gates' Gates Frontier Fund. In addition to investing, GlobalFoundries serves as Kepler's manufacturing partner.

In July this year, the U.S. Department of Commerce signed a letter of intent with Kepler to provide up to $245 million in incentive funding for the R&D of high-performance AI memory enabled by 3D and ferroelectric technologies in the U.S. Notably, this funding remains a preliminary arrangement, and final disbursement is subject to subsequent due diligence and approval.

Kepler Targets AI Inference Memory Bottleneck

Currently, high-end AI accelerators generally rely on HBM to provide high-bandwidth data access, but as model sizes and inference workloads expand, memory capacity, power consumption, and supply availability still affect system scalability.

Kepler stated that its first AI memory product aims to approach SRAM in terms of bandwidth per unit of power consumption while offering higher capacity than existing SRAM and HBM, leveraging 3D structures and new materials to improve computing system energy efficiency. The company also hopes to utilize existing fabs and mature manufacturing equipment to lower the barrier to capacity expansion.

However, these performance metrics are currently based primarily on Kepler's own disclosures and have not yet been validated through large-scale commercial deployment or long-term mass production. Therefore, it cannot be directly concluded at this stage that its product performance has surpassed mainstream HBM solutions.

According to the company's roadmap, Kepler will begin sampling by the end of 2026, enter capacity ramp-up in 2027, and plans to expand domestic U.S. capacity starting in 2028. The company is currently collaborating with GlobalFoundries to advance process development and scale validation in Singapore and Vermont, U.S.

Intel and AMD Co-Investing Does Not Mean Adoption

Investments from Intel Capital and AMD Ventures have elevated Kepler's industry profile, but an investment relationship does not mean that Intel or AMD has decided to adopt Kepler products in their processors or AI accelerators.

To date, AMD has not announced plans to integrate Kepler memory into its Instinct series, nor has Intel announced any adoption plans for its related products.

What Kepler truly needs to validate remains its manufacturing capabilities. Whether the new ferroelectric material can maintain stable yields, reliability, and cost advantages in mass production, and whether 3D integration technology can successfully scale to commercial levels, will determine if its technology can transition from the experimentation and sampling stages into the mainstream AI hardware supply chain. A WIRED report also noted that integrating new materials into mature manufacturing processes and achieving scale represents a major challenge for Kepler in its next phase.

Micron Shares Weaken as HBM Landscape Remains Unchanged

Kepler unveiled its new AI memory technology on September 9. The following day, shares of Micron (MU) fell by roughly 4.9%. However, U.S. Treasury yields rose on that day, and market concerns over inflation and Fed rate hikes also weighed on tech stocks, so Micron's decline cannot be simply attributed to Kepler.

At this stage, SK Hynix (SKHY), Samsung, and Micron remain the primary global HBM suppliers. In March of this year, Micron announced that its 36GB 12-layer HBM4 for the Nvidia Vera Rubin platform had entered mass production and would begin volume shipments in the first quarter of 2026. The product features a bandwidth exceeding 2.8 TB/s and improves energy efficiency by over 20% compared to its HBM3E.

Meanwhile, demand for AI servers continues to drive shipments of high-end memory. TrendForce data shows that AI server demand propelled shipment growth for HBM3E, LPDDR5X, and high-capacity RDIMMs in the second quarter, while inventories at major memory suppliers remained at historical lows.

Therefore, at this stage, Kepler is better viewed as a medium-to-long-term technological variable in the AI memory market rather than a short-term fundamental threat to Micron, SK Hynix, or Samsung.

Whether it can truly alter the HBM market landscape still depends on future sample performance, mass-production yields, cost advantages, and adoption by major AI chipmakers.