Google Pitches Custom TPU Chips to South Korea for Sovereign AI

Google enters the sovereign AI chip race

Google met with South Korean government officials in early September to discuss supplying tensor processing units for the country's sovereign AI infrastructure, according to a report from Seoul Economic Daily. The September 9 meeting marks a strategic shift for Alphabet, which has historically reserved its custom AI accelerators primarily for internal use and Google Cloud customers.

The discussions position Google as an alternative to Nvidia in a market where governments increasingly view compute capacity as critical national infrastructure. South Korea's interest reflects a broader desire among nations to reduce dependence on single foreign suppliers and gain more direct control over AI compute strategies.

Why it matters

This represents a fundamental expansion of Google's chip business model. If TPUs move beyond cloud services into on-premise sovereign installations, Google enters direct competition with Nvidia in the one domain where the GPU maker has maintained structural dominance: high-performance accelerators outside hyperscaler data centers. For governments building national AI capabilities, a credible second option could reshape procurement decisions and spending patterns worth billions of dollars.

The technical and economic case

Google has spent years refining TPUs to reduce the cost of running its own models and cloud workloads. The company reportedly argues that its chips can deliver material cost savings for AI compute, though actual performance depends heavily on specific workloads, software compatibility, and utilization rates.

The challenge lies in interoperability. Nvidia has invested years building CUDA, networking tools, and systems software into a developer platform with deep market penetration. Even governments motivated by diversification may determine that migration costs and software maturity make GPUs the lower-risk choice. Google must also balance supporting external customers against preserving the economic advantages that differentiate its cloud business.

Implications for Nvidia's market position

Sovereign AI represents a significant growth opportunity for Nvidia as countries treat compute as strategic infrastructure. However, that same strategic logic creates pressure to avoid single-vendor dependence. A Korean TPU deployment would not displace Nvidia outright, but it could establish a precedent for splitting national clusters across competing chip architectures.

Investor positioning reflects confidence in both companies. Hedge fund holdings of Alphabet rose from 265 funds in Q1 2025 to 275 in Q2, with Berkshire Hathaway increasing its Class A position by approximately 45 percent during the quarter. Nvidia's holder count climbed from 275 to 285 funds, while Fisher Asset Management raised its stake by roughly 3 percent. These filings predate the September Korea discussions.

Alphabet's short interest stood at 77.70 million shares as of August 31, representing about 0.72 percent of float with 3.48 days to cover.

These details were first reported by Seoul Economic Daily.