Google's TPU Chips Set to Transform AI Hardware Landscape, Boost South Korean Semiconductor Titans

Google is advancing AI hardware with its tensor processing units (TPUs), which could spur demand for high-bandwidth memory, benefiting companies like Samsung and SK hynix. Google's TPUs, cost-effective and efficient, pose a strong challenge to Nvidia's dominance, potentially reshaping market dynamics and boosting the semiconductor supply chain.

In a bold move to redefine the artificial intelligence hardware market, Google is leveraging its tensor processing units (TPUs) to accelerate demand for high-bandwidth memory, according to sources cited by The Korea Herald. The move is anticipated to benefit South Korean semiconductor giants Samsung Electronics and SK hynix significantly.

The report highlights that Google is working to supply its TPU chips—currently used in its AI model Gemini 3—to other Big Tech firms like Meta, which is considering adopting TPUs for its upcoming data centers set to launch in 2027. Co-developed with Broadcom, Google's TPUs offer efficient AI performance, positioning them as a compelling alternative to Nvidia's market-leading GPUs.

TPUs are recognized for their cost-effectiveness, reportedly being up to 80% cheaper than Nvidia's H100 GPU, industry estimates suggest. Even though Google's Ironwood TPU might trail Nvidia's future Blackwell chips in raw computing power, it surpasses Nvidia's H200 in performance. Industry analysts predict that broader adoption of TPUs could break Nvidia's stronghold on the AI chip market, where it commands over 90% market share, and improve profitability across the semiconductor supply chain.

The inclusion of six to eight high-bandwidth memory (HBM) modules in each TPU also directly impacts memory demand. SK hynix, already providing fifth-generation HBM3E chips for Google's Ironwood, is projected to deliver 12-layer modules for the next-generation TPU, nicknamed '7e.'

Increased HBM usage by Google is expected to exacerbate the current supply shortage, boosting average selling prices and shipments for memory firms like SK hynix and Samsung Electronics. Anticipated growth in AI data centers is also likely to escalate demand for products such as DDR5 and LPDDR5 DRAM, further enhancing memory sales.

Concurrently, as Taiwan's TSMC escalates prices for advanced processes, Samsung's foundry is gaining attention due to improved yields in 3-nanometer and 2-nanometer technologies. Samsung's integration of memory, foundry, and advanced packaging solutions is viewed as a strategic advantage.

Quoting industry experts, The Korea Herald notes that Samsung's Texas fabrication plant, nearing readiness to produce sub-2-nanometer chips, stands to capitalize significantly on TPU market expansion. This advancement could not only increase memory shipments but also elevate foundry use and bolster sales of AI-enhanced Galaxy smartphones.

(With inputs from agencies.)