The explosive growth of artificial intelligence (AI), high-performance computing (HPC) and hyperscaling requires fundamental architectural changes in data centres, significantly impacting semiconductors. According to market research and strategy consultancy firm, Yole Group, this market is at an inflection point, as presented in its latest report ‘Data Centre Semiconductor Trends 2025’.
In 2024, the total available market for semiconductors for data centres reached $209bn, which includes compute, memory, networking and power devices. By 2030, Yole predicts this figure to reach $492bn. AI and HPC are driving the demand, with generative AI requiring even more processing power and accelerators. Here, hyperscalers pursue vertical integration and better costs.
GPUs remain the cornerstone of AI infrastructure, considered key devices in AI training and inferencing. Yole sees GPU revenues more than doubling by 2030, from $100bn in 2024 to $215bn. In 2024, the lion’s share of this market was taken by GPU provider, Nvidia, capturing 93% of the server GPU revenues.
Adding to this dynamic environment are AI ASICs, which are increasingly gaining momentum. Here, Google, Amazon and Microsoft are investing in domain-specific silicon to optimise performance and reduce dependence on Nvidia. Based on the market standing of these companies, AI ASICs revenue is expected to reach $84.5bn by 2030.
Memory architectures are evolving, too
Alongside the growing demand for more processing power, memory architectures are also rapidly evolving, with a greater adoption of the fifth generation of Double Data Rate Synchronous Dynamic Random-Access Memory, DDR5. This type memory provides more speed, capacity, power efficiency and data integrity – as required by the evolving data centres – and thus represents a significant upgrade in computing memory technology.
R&D has been focusing on resolving bottlenecks in memory abut also new interconnects. AI training requires high-bandwidth memory and new types of interconnects, such as the Compute Express Link, or CXL, an industry-supported Cache-Coherent Interconnect for Processors, Memory Expansion and Accelerators, that promises to solve memory disaggregation and latency challenges in new server architectures.
AI workloads are expected to continue to drive interconnect architectures.
Chief market players
Leadership in data centre silicon is also shifting. US players remain dominant, especially Nvidia, AMD and Intel, but Yole’s analysts also look to China as a growing player in this sector, as the country is rapidly scaling up its domestic capabilities through strategic investments and policies. Export controls continue to impact supply chains but also reinforce sovereign development goals in China.
There are also startups and newcomers in this market, and they continue to shape it with innovations. Chip designs from startups like Groq, Cerebras and Tenstorrent are pushing what AI inference hardware can do. This is a healthy competition in the market and for the progress of silicon technology, as novel solutions challenge established players, not only with innovations, but costs, performances and energy efficiency, too.
A more in-depth analysis is available in the Yole Group’s new report, ‘Data Centre Semiconductor Trends 2025’ , which provides further detail of how AI, HPC, and hyperscaler demand is driving the new semiconductor trends and technologies.
By Eric Mounier, Chief Analyst, Photonics at Yole Group
