The source of optimism for Nvidia

LEE SOO-HWA
The author is a professor at the AI Convergence Research Institute of Hallym University.

The stock price of Nvidia — the leader in designing and manufacturing GPUs and developing artificial intelligence technology — soared to become the largest market cap in the world at $3.335 trillion on June 18. As the stock price has dropped since then, the market cap of Nvidia is now the third largest after those of Microsoft and Apple.

The slowdown is explained by various arguments ranging from the company selling its shares to realize profit from the short-term surge to bubbles in AI-related stocks. But Nvidia has something to believe in. For one, no other company can surpass Nvidia’s 80 percent market share in the AI GPU category.

As GPU sales are crucial to Nvidia’s stock price, the sales volume is bound to increase. Compared to existing machine learning, a large language model (LLM) requires far more computing resources, especially GPUs, for “inference.” As sales of LLM services such as OpenAI and Microsoft go up, Nvidia’s stock price will rise. Large cloud service companies such as Amazon and Google — called “hyperscalers” — cannot maintain or expand service operation without GPUs.

When multiplied by variables, Nvidia becomes an impregnable fortress. The company is leading in both hardware and software. CUDA — Nvidia’s AI learning tool — is the “One Ring” in the AI field the company got after investing more than $10 billion over the past 20 years. In order to replace CUDA, which emerged as the standard library for AI learning since deep learning in 2010, large-scale AI specialists need to be mobilized. As the global demand for those AI specialists is great, they are in short supply. The most expensive and capable AI specialists should be recruited to make a tool that can beat CUDA. The talent search will nearly be a “mission impossible.”

Hyperscalers are in the early stages of introducing AI development chips, rather than GPUs, to cut costs. But due to conflicts of interest, the possibility of releasing AI development tools as an open source is slim. If not incorporated into the open-source community, it is difficult to improve quality through testing. This stands in contrast with how CUDA was open-source from the beginning.

Will the sales of hardware, software and service companies related to generative AI show an uptick? Will Nvidia’s stock prices continue to bloom as a result? In the short term, worrying about the stock price may be unnecessary when AI improves the quality, cost of goods and services traded in the market.

Will AI work as the “stem cell” to solve the challenges of humanity? The limits AI will face in moonshot projects to resolve the major enemies of mankind — such as the pandemic, climate change, wealth polarization and medical welfare — will pose mid- to long-term challenges to the industry.