As demand for Google's self-developed artificial intelligence (AI) chip surges, cracks are appearing in NVIDIA's dominance in the AI semiconductor market. Amid the accelerating movement among big technology corporations to 'decouple from NVIDIA,' there are assessments indicating that NVIDIA's market dominance is gradually weakening. According to market research firm Omdia, Google's AI chip TPU (Tensor Processing Unit) is expected to generate sales of $6 billion to $9 billion (approximately 8.8 trillion won to 13.2 trillion won) this year. The TPU is a custom chip designed by Google for AI operations, specialized for machine learning and inference. In comparison, NVIDIA's AI chip sales are estimated to exceed $80 billion (approximately 117.74 trillion won), making Google's figures only about one-tenth of that. However, Omdia analyzed that the market demand for Google's AI chips is strong enough to replace NVIDIA's GPU (graphics processing unit) AI chips.

Alexander Harwell, a chief researcher at Omdia, noted, 'Demand for Google TPU is growing very rapidly, enough to take away market share from NVIDIA.' He added, 'Even if this year's TPU shipment sales only hit the lowest expectation of $6 billion, it will have a significant impact on the market share of AI chips dominated by NVIDIA.'

Graphic=Jeong Seo-hee

The high demand for Google's AI chips stems from the concerns of big technology corporations looking to improve cost efficiency. In particular, global cloud service providers (CSPs) such as Microsoft, Alphabet, Amazon, and Meta have started to prefer customized AI chips (application-specific integrated circuits) that are more power-efficient than NVIDIA's GPUs. In the past, big technology corporations that have purchased NVIDIA AI chips in bulk for AI development referred to the high operational costs of NVIDIA chips as the 'NVIDIA tax.' As the cost burden has increased, these corporations have sought to develop their own AI chips or find alternatives.

Apple also used Google's chips instead of NVIDIA chips for AI model training. In a paper published in July, Apple stated that it trained the AI model that underpins its AI system 'Apple Intelligence' on Google TPU clusters. Following this trend, U.S. investment bank Morgan Stanley projected that the growth rate of the customized AI chip market, led by cloud service providers, would surpass that of the GPU market. The market size for customized AI chips is expected to grow from $120 billion (approximately 176.5 trillion won) this year to $300 billion (approximately 441.2 trillion won) by 2027.

Thanks to the demand for Google TPU, the Google Cloud Platform business responsible for it is also thriving. In the third quarter of this year, the revenue of Google Cloud Platform reached $11.4 billion (approximately 16.7 trillion won), a 35% increase compared to the same period last year. This accounts for about 13% of the total revenue of Alphabet, Google's parent company ($88.3 billion), and this proportion has been steadily rising. During the same period, the operating profit of Google Cloud Platform soared by 627% year-on-year to $1.938 billion (approximately 2.8 trillion won), with an operating profit margin of 17%. Harwell noted, 'High-speed data processing services and AI products utilizing TPU are the main drivers of Google Cloud's growth,' adding, 'Especially, the high-speed data processing services based on TPU are highly profitable and will drive Google Cloud's performance in the future.'