Google’s TPU Expansion May Open New HBM Markets for Samsung and SK Hynix

Google’s decision to expand its custom artificial intelligence chip business to external customers could create new high-bandwidth memory (HBM) opportunities for South Korean memory makers Samsung Electronics and SK hynix. According to industry reports, this move helps memory suppliers diversify demand beyond Nvidia’s dominant graphics processing unit ecosystem.

The US technology major is currently broadening sales of its tensor processing units (TPUs) to cloud providers and companies building AI infrastructure. Google stated that it now counts nearly 100 TPU customers, including businesses deploying the processors within their own data centers. Additionally, Oracle plans to offer Google’s chips through its cloud platform next year, while Alphabet anticipates substantial growth in TPU system sales following contributions to Google Cloud’s third-quarter revenue.

As part of this hardware push, Google recently unveiled two new TPU systems—the TPU 8i Max and the TPU 8tx—bringing its eighth-generation lineup to four systems. The 8i Max integrates two chips into a single package to double computing performance and HBM capacity, whereas the 8tx focuses on AI workloads requiring enhanced data security.

Industry analysts note that the core business shift lies in Google transforming TPUs into products that serve clients outside its proprietary data centers. This growing demand for customized AI chips is expected to benefit memory suppliers significantly. Counterpoint Research forecasts that HBM bit demand for custom AI server chips will increase 35-fold between 2024 and 2028.

South Korean manufacturers are actively positioning themselves to capture this demand. Samsung is pursuing opportunities alongside Broadcom, which serves as Google’s major TPU development partner. Both companies signed a memorandum of understanding in July covering HBM, foundry manufacturing, and advanced packaging. The potential cooperation is valued at over USD 200 billion through 2030, though this financial estimate encompasses multiple business areas and does not represent confirmed HBM orders.

Meanwhile, SK hynix has proposed the joint design of customized HBM for future TPUs, aiming for earlier involvement in the chip development cycle. Google’s procurement strategy could further alter supplier dynamics. Its collaboration with Taiwan’s MediaTek on the TPU 8t could help the company reduce reliance on Broadcom’s model, where the design partner oversees memory purchases. Gaining tighter control over design and procurement may enable Google to engage directly with memory suppliers and lower overall costs.

Despite these prospects, translating the opportunity into firm orders presents hurdles because current production capacities remain heavily committed. SK hynix reported that existing agreements with 14 major customers account for volumes exceeding half of its projected production capacity for 2027 and 2028. Consequently, analysts emphasize that early participation in future chip architectures will be vital for Samsung and SK hynix to secure long-term supply contracts.

"Google's expansion of its custom TPU business beyond its internal data centers marks a notable shift in the AI hardware supply chain. For major memory manufacturers like Samsung and SK hynix, diversifying beyond Nvidia's ecosystem is a positive strategic step. However, with production capacities heavily booked for the coming years, securing long-term contracts will depend heavily on early design collaboration and operational readiness to meet the rising demand for customized AI server chips." — Dr. Shishir Gupta, Founder & CEO, StartupLanes