[Click eStock] "Nvidia Evolving into a Physical AI Company"

On December 31, NH Investment & Securities predicted that concerns over Nvidia's product sales restrictions to China and a potential decline in competitiveness in the artificial intelligence (AI) inference sector would gradually be resolved, leading to an upward revision of its earnings outlook. The target price was set at $260.8.

Recently, U.S. President Donald Trump announced a measure allowing the export of H200 chips to China, which is now acting as a factor for raising Nvidia's earnings guidance. Koh Minseong, a researcher at NH Investment & Securities, stated, "Although Nvidia will have to pay 25% of its sales proceeds as a fee to the U.S. government and still requires approval from the Chinese government, an upward revision of earnings estimates is inevitable once product sales resume." The supply of H200 chips is expected to begin in February 2026, with the initial volume estimated at 40,000 to 80,000 units.

Concerns over intensified competition in the AI inference sector are also being addressed through a technology license agreement with Groq. Previously, on December 24, Nvidia decided to acquire the core assets of Groq, an AI accelerator startup founded by developers of Google's Tensor Processing Unit (TPU), for $20 billion (approximately 29 trillion won).

Koh explained, "This is the largest acquisition in Nvidia's history and can be interpreted as an effort to strengthen competitiveness in the AI inference sector." He added, "If Nvidia successfully internalizes Groq's core technology, the language processing unit, it will be able to expand its market share in the inference sector as well."

The upcoming 'CES 2026' is also expected to showcase expanded partnerships in autonomous driving and robotics. Koh stated, "We will be able to see real-world use cases of physical AI solutions," and added, "In particular, the key point to watch will be how differentiated the inference and complex task performance capabilities are for robots equipped with the Jetson Thor chip and trained through the general-purpose model 'Project GR00T,' compared to competitors."