A Global Race against Nvidia’s Stranglehold on the Chip Market

In view of the explosion in the AI chip market, the shortage of Nvidia GPUs and their exorbitant prices, many American, European and Taiwanese companies dream of overtaking Jensen Huang’s company. Nvidia’s hegemony is set to endure, given the excellence of its GPUs for AI and the software empire it has built around them. However, there is room for alternative designers and manufacturers. Rémi Bourgeot is an Economist, Engineer, and Associate Fellow at IRIS. Estelle Prin is the Founder of The Semiconductors Observatory.

Beyond investors’ focus on Nvidia’s AI chip empire, tentative alternatives are beginning to take shape. Various credible options are emerging to push back the limits of existing chips. However, Nvidia’s competitors, whether Big Tech giants or cutting-edge start-ups, are faced not only with the technical supremacy of Jensen Huang’s company, but also with the closed environment it has developed around Cuda, its proprietary platform.

Its market capitalization has exploded to around $3,000 billion, making Nvidia the third most valuable American company in the world… Since January 2023, its share price has jumped by almost 450 %. Sales for the last quarter of 2023 reached $22.6 billion, compared with $6 billion for the same period the previous year. As fanciful as Nvidia’s share price may seem, it is in line with the company’s near-monopolistic business reality.

Nvidia rides the AI wave

The company controls between 70 % and 95 % of the design of the various AI chips, positioning itself at the forefront of the current boom. It is crushing competition from AMD, Qualcomm, Amazon and Google. Some Big Tech companies have started designing their own chips for their data centers, in line with the AI boom. But this is a recent phenomenon compared to the long experience of a pure design company like Nvidia.

The latter owes its success primarily to its decade-long focus on AI, the result of a visionary gamble. Parallel computing on GPUs has proven to be well-suited to the countless linear algebra operations that underlie the training of giant neural networks. This resolute reorientation towards AI was not an obvious choice for a company originally specializing in GPUs for video games.

Nvidia also benefits from another major asset: Cuda, its software platform, which enables customers to adapt their own AI models very quickly using the company’s chips. Huang describes Cuda as the operating system (OS) of AI. Owned by Nvidia, it makes customers captive. Developed since 2007 and constantly upgraded, this software platform is now used by the majority of AI model developers worldwide. Cuda has become an international standard.

Emerging alternatives

Alternative approaches to Cuda are emerging. The Triton platform was launched by OpenAI in 2021. Meta, Google and Microsoft are contributing. Intel and AMD are also investing in it to bypass the Nvidia ecosystem.

Given the explosion of the AI chip design market, the shortage of Nvidia chips and their exorbitant prices, many North American, European and Taiwanese companies are dreaming of dethroning the company headed by Jensen Huang. In addition to the efforts of Intel, AMD and giants like Microsoft, Meta and Amazon in the AI cloud, start-ups are demonstrating their boundless creativity.

In California, Cerebras and Groq are developing alternative architectures to increase chip speed at lower cost. The aim of these rival companies is to surpass the efficiency of Nvidia chips, with architectures that are more efficient, less expensive… and consume less energy. For example, Cerebras is developing large chips rather than stacking GPUs, in order to reduce latency.

Nvidia’s hegemony is set to continue, given the excellence of its AI GPUs and the software empire it has built around Cuda to exploit them. However, the demand and interest from investors and Big Tech is such that alternative designers and manufacturers of AI chips can exist. It’s a matter of betting on original, even disruptive approaches, focusing on chip efficiency, as well as availability and price.

 

This article was originally published by Les Echos in French.