Nvidia's Reign At Risk? Dylan Patel Says Google's TPU, Amazon's Trainium Could Outshine GPUs If Sold To Public

Nvidia Corp. NVDA faces pressure from custom silicon competitors as tech giants scale up alternative artificial intelligence chip production, according to SemiAnalysis founder Dylan Patel.

Custom Silicon Orders Surge

Google and Amazon are “massively” increasing orders for their proprietary chips, Patel said during a recent podcast with a16z, also known as Andreessen Horowitz, a popular venture capital firm in Silicon Valley. Amazon.com Inc. AMZN produces millions of Trainium processors while Alphabet Inc. GOOGL GOOG manufactures millions of Tensor Processing Units.

“Google’s TPUs are clearly 100% utilized,” Patel noted. “Amazon’s Trainium isn’t there yet, but Amazon will figure out how to optimize usage.”

Google’s TPU Market Potential

Patel believes Google should sell TPUs directly to external customers rather than just cloud rental services. Such a move could theoretically generate higher market value than Google’s core business.

“Google is even discussing it internally,” Patel said. “It would require a big reorganization of culture and how Google Cloud works, but they could totally do it.”

“I think that’s the biggest threat to Nvidia is that people figure out how to use custom silicon more broadly,” Patel said.

Google’s latest TPU, Trillium, delivers 4.7x peak compute performance compared to previous generations and operates 67% more energy-efficiently. The company currently restricts TPU access to internal use and select cloud customers, including Apple Inc. AAPL.

Market Concentration Determines Winners

The AI chip landscape depends heavily on customer concentration, according to Patel. Concentrated AI development among major tech companies favors custom silicon, while broader distribution benefits NVIDIA’s general-purpose GPUs.

Recent developments support both scenarios. OpenAI continues relying on Nvidia GPUs despite testing Google TPUs.

Amazon’s Trainium2 chips offer 30-40% better price performance than comparable GPU-based instances, positioning AWS as a formidable competitor in the training market.

Read Next: