Nvidia has firmly established itself as the leader in the AI chip market.
However, AMD and the development of proprietary chips may challenge the company’s position, both in the short term and particularly in the medium term.
In 2024, Nvidia made a great fortune, firmly establishing itself as the leading provider of AI chips. The GPU manufacturer has become essential for major technology companies, at least for now. The landscape may shift in 2025.
Microsoft buys twice as much as its rivals. According to data from tech consultancy Omdia, Microsoft has emerged as Nvidia’s largest customer in the U.S. In 2024, the company acquired 485,000 chips based on Nvidia’s Hopper architecture, including H100 and GH200. In contrast, Meta ordered 224,000.
Omdia analyzed the capital expenditures of each company, as well as server shipments and channel and vendor information.
The reason: OpenAI. Microsoft’s substantial investment in Nvidia is likely influenced by its own $13 billion stake in OpenAI. As the creator of ChatGPT, OpenAI is a leading force in AI models, resulting in a high demand for resources that Microsoft provides. However, Microsoft’s massive chip purchases are also driven by Copilot in all its versions.
Beware of Chinese Big Tech. The trade war restricts Chinese tech giants from purchasing Nvidia’s most advanced chips, at least in large quantities. However, companies like ByteDance and Tencent have become significant customers in 2024. They acquired approximately 230,000 AI chips, including the H20 model, which is slightly less powerful than the H100. In comparison, Amazon purchased 196,000 of these chips, while Google acquired 169,000.
Data centers. Another noteworthy finding from Omdia’s study is the spending on servers. In 2024, tech companies globally invested $229 billion, with Microsoft ($31 billion) and Amazon ($26 billion) being the leading spenders. The top 10 companies account for 60% of total data center investment this year. Two recent companies making significant investments in this area are Elon Musk’s xAI and CoreWeave, which “rents” GPUs to AI developers, although it initially provided them for cryptocurrency miners.
Nvidia is everywhere. According to Vlad Galabov, director of cloud and data center research at Omdia, 43% of spending on servers in 2024 went to Nvidia. The company has clearly emerged as the major victor in the AI sector, but this trend could shift in the future.
AMD is slowly growing. Although at a far distance from Nvidia, AMD’s AI chips are starting to gain significance. In fact, Meta purchased 173,000 of its MI300 chips, while Microsoft acquired 96,000. After introducing its Instinct series in October, AMD has an encouraging roadmap planned for 2025.
Big Tech companies are developing their own chips. In addition to AMD’s advancements, it’s especially noteworthy that major tech companies are increasing their investment in and utilization of proprietary chips for their data centers and AI models.
- Google has been developing its Tensor Processing Units for over a decade and utilizes them extensively in its data centers. However, the company hasn’t released specific deployment figures.
- Meta launched its first training and inference chips in late 2023 and has already deployed 1.5 million of these chips for its AI models.
- Amazon is making significant investments in its Trainium and Inferentia chips, having already deployed 1.3 million of them in its data centers. Additionally, the company’s investment in Anthropic includes creating a new cluster featuring thousands of its next-generation Trainium chips.
- Apple has partnered with Amazon instead of Nvidia, a decision that positively surprised Amazon.
- Microsoft is working towards reducing its reliance on Nvidia. However, it’s only deployed 200,000 of its own Maia AI chips so far.
Image | BoliviaInteligente
See all comments on https://www.xatakaon.com
SEE 0 Comment