Memory by Design

Summary: Beyond Nvidia's GPUs, GenInnov's thesis argues that memory chips are shifting from commodity parts to customised, high-margin subsystems fused with logic and that the market still prices memory makers like SK Hynix as cyclical commodities despite this structural change.

Summary: Beyond Nvidia's GPUs, GenInnov's thesis argues that memory chips are shifting from commodity parts to customised, high-margin subsystems fused with logic and that the market still prices memory makers like SK Hynix as cyclical commodities despite this structural change. The artificial intelligence narrative over the past two years has centred on one name: Nvidia. The market has rewarded the logic side of the equation, treating graphics processing units (GPUs) as the sole engines of the AI revolution. ​But beneath this surface narrative, an architectural shift is underway in the semiconductor industry; one that could mark the next wealth-creation cycle. According to recent research from innovation investing firm GenInnov, the defining investment theme of 2027 will not just be faster math. It will be ‘Memory by Design’. ​The boundary between logic (computing data) and memory (storing data) is dissolving. For decades, memory was treated as a plug-and-play commodity, a cyclical sector prone to price wars. As AI models grow exponentially heavier, memory is becoming a customised, high-margin subsystem. Investors who miss this structural shift risk mispricing one of the most defensible economic moats of the coming decade. ​Leaving flatland: The end of two-dimensional progress For 80 years, chipmaking lived in a two-dimensional ‘flatland’. Progress was measured purely by shrinking the distance between transistors on a flat surface, famously dictated by Moore’s Law. ​An important piece of flat-world planning from the 1940s was known as the Von Neumann architecture. In this setup, computing and m

This article was originally published on October 01, 2026.