In this episode of The Tech Poutine, hosts Ian and George discuss a wide range of recent developments in the semiconductor and memory industries, starting with Micron’s announcement that it will stop selling products under its Crucial consumer brand by February 2026. They clarify that this does not mean Micron is exiting the consumer memory market but rather shifting away from direct-to-consumer sales, with consumer-grade memory now being sold through third-party brands like Corsair and G.Skill. The discussion highlights the complexities of memory manufacturing, binning, and market segmentation between enterprise and consumer products, emphasizing that Micron’s move is largely a corporate cost-cutting measure rather than a response to AI demand.
The conversation then shifts to memory technology trends, including the roadmap for DRAM nodes and the challenges of high-bandwidth memory (HBM) power consumption. They note that HBM stacks consume significant idle power due to the lack of low-power modes in current generations, which impacts data center energy efficiency. The hosts also touch on advances in memory packaging and stacking technologies, such as compression bonding and hybrid integration, and discuss the importance of optical interconnects in high-performance computing. They share insights from recent conferences like IEDM and Hot Chips, revealing ongoing innovations and the complexity of scaling memory technologies.
Supercomputing and high-performance computing (HPC) events are another major topic, with the hosts recounting their experiences at the Supercomputing conference. They describe the scale and scope of the event, noting the growth of the trade show floor and the challenges of navigating such a large venue. Key announcements include new AMD and NEC vector engine developments, advances in optical interposer technology by startups like Light Matter and Celestial AI, and the acquisition of Celestial AI by Marvell. The discussion highlights the critical role of manufacturing scale-up and supply chain robustness in bringing cutting-edge optical technologies to market.
The episode also covers recent acquisitions and corporate moves in the AI hardware space. Qualcomm’s acquisition of Ventana Micro, a RISC-V CPU design company, is analyzed as a strategic move to bolster Qualcomm’s data center CPU ambitions. In contrast, Intel’s potential acquisition of AI startup Sanova is viewed with skepticism due to Intel’s current strategic priorities and past integration challenges. The hosts also critique the controversial trademark lawsuit filed by startup Tachium against Google over the term “TPU,” suggesting it may be a tactic to force a buyout rather than a legitimate legal claim.
Finally, the hosts preview upcoming industry events like CES 2026 and discuss expectations for announcements from major players such as AMD, Intel, Qualcomm, and Nvidia. They share technical insights into new products like AWS’s Tranium 3 AI chip, the Graviton 5 CPU, and Nvidia’s rumored N1X laptop GPU. The episode concludes with reflections on the challenges of memory bandwidth and latency in modern heterogeneous computing systems, the importance of continuous innovation in AI hardware, and the value of community engagement through channels like Tech Potato and Chips and Cheese.