Google’s AI chip push boosts Samsung, SK hynix demand
Google’s push to expand its AI chip offerings is creating fresh opportunities for Samsung Electronics and SK hynix in the high-bandwidth memory (HBM) market, moving beyond Nvidia’s dominant GPU ecosystem. The development follows Google’s growing sales of its custom tensor processing units (TPUs) to cloud providers and companies building AI infrastructure, diversifying demand for Korean memory manufacturers.
On Monday, Google unveiled two new TPU systems—the TPU 8i Max and the TPU 8tx—expanding its eighth-generation lineup to four models. The TPU 8i Max combines two chips in a single package, doubling computing performance and HBM capacity, while the TPU 8tx targets AI workloads requiring stronger data security. A key development for Samsung and SK hynix is Google’s growing base of external customers, now nearly 100, including companies deploying TPUs in their own data centers.
Oracle plans to offer Google’s chips through its cloud platform next year, while Alphabet expects TPU system sales to increase substantially after contributing to Google Cloud’s third-quarter growth. Industry analysts highlight that Google’s strategy now treats TPUs as a business serving customers outside its own data centers.
Samsung and Broadcom team up on AI memory deals
Counterpoint Research forecasts that HBM bit demand for custom AI server chips will grow 35-fold between 2024 and 2028, driven partly by Google’s TPU expansion. Samsung is pursuing opportunities through Broadcom, Google’s major TPU development partner. The companies signed a memorandum of understanding in July covering HBM, foundry manufacturing, and advanced packaging, with potential cooperation valued at more than $200 billion through 2030. The estimate spans multiple businesses and does not include confirmed HBM orders.
SK hynix is seeking a role earlier in chip development. Following a February meeting between SK Group Chairman Chey Tae-won and Google CEO Sundar Pichai, the memory maker proposed jointly designing customized HBM for future TPUs. Chey also met Broadcom CEO Hock Tan to discuss next-generation AI chips.
Read Also: Posco Expands Gwangyang LNG Terminal to Boost Energy Security
Google shifts TPU procurement to cut costs
Google’s changing procurement strategy could create another opening. According to Counterpoint, Google is working with Taiwan’s MediaTek on its TPU 8t, partly to move away from Broadcom’s turnkey model, where the design partner also handles memory purchases. Counterpoint estimates that Broadcom’s model adds a 15 to 20 percent markup to HBM costs. By taking greater control of chip design and procurement, Google could reduce costs and deal more directly with memory suppliers.
“By bringing chip design and HBM procurement in-house, starting with Zebrafish, Google eliminates the intermediary fee and lowers its overall cost,” Counterpoint research associate David Wu said, referring to the TPU 8t’s code name.
Customized HBM presents another opportunity. With HBM4 and HBM4E, suppliers can tailor memory components to specific AI processors, potentially securing higher-value, longer-term contracts through earlier involvement in chip design. Counterpoint research director Hwang Min-sung called the shift “a major opportunity for memory suppliers to create a high-value business.”
Yet securing new orders may be difficult as production capacity is already heavily committed. SK hynix said in October that agreements with 14 major customers cover volumes equivalent to more than half its total production capacity for 2027 and 2028, including products beyond HBM.
“Winning a new customer doesn’t necessarily mean shipping more HBM right away when so much future capacity is already committed,” another industry source said. “The real competition is over who gets into the next chip design and secures production slots early enough to turn that relationship into long-term orders.”
