Memory’s share of AI chip component costs is estimated to have risen to 63 percent in the fourth quarter of 2025.

Online media outlet Gigazine reported on Sunday that Epoch AI said its review of AI chips designed by Nvidia, AMD, Google and Amazon found the share rose to 63 percent in the fourth quarter of 2025 from 52 percent in the first quarter of 2024.

Epoch AI divided AI chip component costs into four categories: memory, logic, packaging and auxiliary components. In the first quarter of 2024, the shares were 52 percent for memory, 14 percent for logic, 19 percent for packaging and 15 percent for auxiliary components. In the fourth quarter of 2025, the shares were 63 percent for memory, about 13 percent for logic, 15 percent for packaging and 10 percent for auxiliary components.

In dollar terms, memory saw the biggest increase. Memory-related spending on AI chips by the four companies rose to about $32 billion in 2025 from about $12 billion in 2024. Over the same period, total spending on AI chip components increased to about $52 billion from about $22 billion. About $20 billion of the increase was memory-related costs.

The bigger memory share was attributed to AI chips’ high demand for memory bandwidth. Training and inference for AI models require large-scale matrix operations, making memory performance that supplies data quickly important as well as compute-chip performance. As a result, AI chips use large amounts of high-bandwidth memory such as HBM3 and HBM3e.

The figures are estimates rather than measured values. Unit costs vary by contract terms, supplier and purchase timing, and there is uncertainty in production volume and product configuration by chip. Applying a 90 percent confidence interval, Epoch AI presented a range for the 63 percent memory share in the fourth quarter of 2025. It put the range at 60 to 67 percent when reflecting only memory costs, and 54 to 73 percent when reflecting uncertainty across all components.

Epoch AI expected the memory share could rise further in 2026 if memory supply tightens and prices increase. It also analyzed that higher component prices influenced Microsoft’s capital expenditure outlook for fiscal year 2026 and Meta’s decision to raise its 2026 capital expenditure range.