# Samsung and Broadcom tie AI memory to foundry and advanced packaging Source: TechNewsList (https://technewslist.com) Canonical URL: https://technewslist.com/en/article/samsung-broadcom-hbm-foundry-ai-infrastructure-2026-08-13-night Section: Hardware (https://technewslist.com/en/hardware) Author: TechNewsList Language: en Published: 2026-08-13T17:15:18.332+00:00 Updated: 2026-08-13T17:15:18.514237+00:00 > Samsung and Broadcom’s new MOU reaches across HBM, 2-nanometer foundry, and 2.3D and 2.5D packaging, underscoring how AI silicon is becoming a systems-supply-chain contest. ## TL;DR - Samsung and Broadcom signed an MOU spanning memory, foundry, and advanced packaging. - The companies estimate the collaboration could exceed $200 billion across five years through 2030. - The memory work includes HBM for Broadcom’s next-generation AI accelerators. - The foundry work focuses on Samsung’s 2nm and below processes and 2.3D and 2.5D integration. - The MOU is a strategic commitment, not proof of final volumes or qualified products. ## Key points - Samsung is combining memory and foundry capabilities in one customer relationship. - Broadcom brings AI accelerator and connectivity expertise to the collaboration. - Advanced packaging is treated as a performance and power-efficiency lever. - The partnership reflects demand for integrated AI and networking silicon. - Execution depends on yields, qualification, delivery schedules, and customer deployments. # Samsung and Broadcom tie AI memory to foundry and advanced packaging Samsung Electronics and Broadcom have signed an MOU that spans HBM, advanced foundry processes, and packaging. The agreement is a useful snapshot of where AI hardware competition is heading: the decisive product is increasingly a coordinated silicon system, not an isolated accelerator die. ## What happened Announced at an AI Summit in San Francisco on July 25, the MOU covers next-generation memory and foundry technologies. Samsung and Broadcom said they expect the collaboration to be worth more than $200 billion across memory and foundry over the five years through 2030. On the memory side, the companies plan to pursue HBM solutions for Broadcom’s next-generation AI accelerators. On the foundry side, the work focuses on Samsung’s 2nm and below process technologies for Broadcom products, including wireless broadband communications solutions. The companies also expect the relationship to extend to 2.3D and 2.5D advanced packaging built on Samsung’s 2nm process. ![Samsung Broadcom semiconductor MOU](https://img.global.news.samsung.com/global/wp-content/uploads/2026/07/25142854/Samsung-Semiconductors-Memory-and-Foundry-Technologies-Broadcom-MoU_Thumb932_F.jpg) *The MOU treats memory, logic, and packaging as parts of the same AI infrastructure decision.* ## Why it matters AI accelerators are only useful when the surrounding system can feed them data, cool them, connect them, and manufacture them at volume. HBM helps with bandwidth, but its value depends on packaging, power delivery, thermal design, and the software stack. Foundry capacity matters because the best architecture is irrelevant if it cannot be built reliably. The partnership therefore has strategic logic for both sides. Samsung can offer a broader bundle of memory, logic manufacturing, and packaging. Broadcom can pursue custom AI and networking silicon with a supplier relationship that reaches beyond wafer fabrication. That may reduce coordination friction, although it does not eliminate technical risk. ## Technical details HBM places stacked DRAM close to an accelerator and uses a very wide interface to move data. The approach improves bandwidth per package, but it makes assembly more complex. Defects, warpage, thermal stress, interposer constraints, and yield all influence how many usable products emerge from a manufacturing run. Advanced packaging is similarly important. 2.3D and 2.5D integration can place multiple dies and memory components in a tighter system, shortening interconnects and improving power efficiency. The trade-off is that packaging becomes part of the product architecture and needs its own design rules, testing, and supply planning. ![HBM4E wafer at a semiconductor technology event](https://semiconductor.samsung.com/resources/tech-blog/images/2026/06/computex-2026-hbm4e-wafer.jpg) *HBM performance is inseparable from packaging quality, thermal limits, and qualification.* ## Market / industry impact The agreement signals that AI customers increasingly want fewer gaps between component suppliers. A platform provider may prefer a partner that can coordinate memory, process technology, and packaging rather than negotiate each layer independently. That can speed integration and simplify accountability, but it may also make the market more concentrated. The $200 billion estimate should be read carefully. An MOU is not a purchase order, a shipment schedule, or a guarantee that every planned technology reaches production. Investors and customers will need evidence of qualification, capacity, yields, and delivered systems before treating the estimate as realized demand. The broader competitive effect is already visible. Memory companies are investing in HBM and adjacent technologies, foundries are competing for advanced-node customers, and packaging houses are becoming strategic rather than invisible subcontractors. AI infrastructure is pulling all three into the same conversation. ## What to watch next Watch for named products, tape-outs, HBM qualification, packaging milestones, and customer deployments. Power efficiency at rack scale will matter more than a peak bandwidth figure. Delivery reliability and serviceability will matter more than an attractive roadmap slide. Samsung and Broadcom are making the right systems-level bet. The harder question is whether the partnership can turn a broad MOU into repeatable, qualified silicon before AI infrastructure demand moves to the next bottleneck. There is also a strategic timing question. Advanced-node and packaging capacity are planned years ahead, while accelerator roadmaps can change in quarters. A supplier relationship that is broad enough to cover memory and packaging can reduce one kind of uncertainty, but it can also lock both companies into assumptions about which workloads and interconnects will dominate. Customers will want flexibility as model architectures, cooling designs, and rack standards evolve. That makes the MOU a starting point for coordination rather than a finished competitive advantage. The companies still have to demonstrate that the pieces work together under real workloads, that the economics survive yield loss and testing, and that customers can buy complete systems on a predictable schedule. ## Sources - [Samsung: collaboration with Broadcom](https://news.samsung.com/global/samsung-electronics-and-broadcom-expand-strategic-collaboration-across-memory-and-foundry-technologies) - [Samsung Semiconductor: AI memory and storage at COMPUTEX 2026](https://semiconductor.samsung.com/news-events/tech-blog/samsung-showcases-next-generation-ai-semiconductor-innovations-at-computex-2026/) - [Qualcomm: Dragonwing data-center roadmap](https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent) Mentions: Samsung Electronics, Broadcom, HBM, 2nm, advanced packaging, AI accelerators ## Sources - [Samsung Global Newsroom](https://news.samsung.com/global/samsung-electronics-and-broadcom-expand-strategic-collaboration-across-memory-and-foundry-technologies) - [Samsung Semiconductor](https://semiconductor.samsung.com/news-events/tech-blog/samsung-showcases-next-generation-ai-semiconductor-innovations-at-computex-2026/) - [Qualcomm](https://www.qualcomm.com/news/releases/2026/06/qualcomm-unveils-comprehensive-data-center-roadmap-for-the-agent)