Google Expands TPU Chips to Outside Customers

Google is turning its tensor processing units into a business serving clients beyond its own data centers, drawing in Korean memory suppliers.

Updated on Nov. 11, 2026 in Semiconductors

Isometric editorial illustration of a tall stack of memory modules beside a square processor die on a dark slate field.
Google said Monday it is selling its TPU AI chips to outside customers, with Samsung and SK hynix pursuing expanded high-bandwidth memory orders tied to the push. AI Illustration. Upload story photo >

Live Poll

Do you expect electronics prices to rise over the next two years?

Google unveiled two new TPU systems, the TPU 8i Max and TPU 8tx, on Monday, bringing its eighth-generation lineup to four systems. The company says it now has nearly 100 TPU customers, and Samsung and SK hynix are pursuing expanded high-bandwidth memory orders tied to the expansion.

Why it matters

Google's push to sell TPUs externally diversifies AI infrastructure demand beyond Nvidia's GPU ecosystem and creates opportunities for Korean memory makers. Google is also moving chip design and HBM procurement in-house, starting with the TPU 8t, to bypass the intermediary fees in Broadcom's turnkey model.

Google reports nearly 100 TPU customers, while Counterpoint Research projects a 35-fold increase in HBM bit demand for custom AI server chips between 2024 and 2028. Broadcom's turnkey model adds a 15 to 20 percent markup to HBM costs, and the scale of any Samsung-Broadcom deal beyond its memorandum remains unresolved.

The players

Google

Alphabet's technology company that designs the Tensor Processing Unit, a custom chip built for AI workloads, and runs one of the largest cloud platforms.

Samsung Electronics

South Korean electronics conglomerate and one of the world's largest memory chip makers, pursuing expanded high-bandwidth memory supply for AI chips.

SK hynix

South Korean semiconductor manufacturer and a leading supplier of high-bandwidth memory used in AI accelerators.

The details

Samsung and Broadcom signed a memorandum of understanding in July covering HBM, foundry manufacturing and advanced packaging. SK hynix proposed jointly designing customized HBM for future TPUs after a February meeting between SK Group Chairman Chey Tae-won and Google CEO Sundar Pichai, and Google is working with MediaTek on its TPU 8t.

Timeline

  1. February: SK Group chairman met Google CEO on HBM customization.

  2. July: Samsung and Broadcom signed a memorandum of understanding.

  3. October: SK hynix said agreements with 14 major customers cover more than half of its 2027 and 2028 production capacity.

  4. Monday: Google unveiled TPU 8i Max and TPU 8tx.

The Tech Race

Google's external TPU sales present a direct alternative to Nvidia's GPU ecosystem, which has dominated AI accelerator supply. The move also reshapes the memory supply chain, as Korean makers like SK hynix and Samsung seek to diversify their HBM demand away from a single chip ecosystem.

Broader TPU availability could expand the range of cloud providers offering AI compute, including Oracle's planned offering next year. Users of AI services may eventually see differences in pricing and performance as the chip ecosystem diversifies beyond a single supplier.

The takeaway

Competition in AI chips is widening beyond Nvidia, and memory suppliers in South Korea are positioning to benefit. Watch whether projected HBM demand growth translates into finalized supply agreements.

Further reading

For broader context on the chip supply chain behind this shift, explore Semiconductors coverage on Wisevoter.

Source note: This article includes information reported by The Korea Herald.

Live Poll

Do you expect electronics prices to rise over the next two years?