A new analysis by Epoch AI projects a significant shift in the cost structure of artificial intelligence chips for major tech giants including Nvidia, AMD, Google, and Amazon between the first quarter of 2024 and the fourth quarter of 2025. The report breaks down the expenses associated with four critical component categories: High Bandwidth Memory, or HBM, logic dies, advanced packaging technologies like CoWoS, and auxiliary components. By multiplying per-chip costs by estimated quarterly production volumes, the team calculated the share of total spending for each category over the forecast period. The findings indicate a dramatic reallocation of costs, driven primarily by the demand for memory. HBM's share of total component spending is expected to surge from 52 percent in early 2024 to 63 percent by the end of 2025. This rise coincides with a sharp increase in overall spending on AI chip components, which is projected to grow from approximately 22 billion dollars in 2024 to 52 billion dollars in 2025. Notably, the 30 billion dollar increase in total spending is largely attributed to HBM, which alone accounts for roughly 20 billion dollars of the growth. In contrast to the memory sector, other component costs are expected to shrink relative to the total. Advanced packaging costs are projected to fall from 19 percent to 15 percent of the total spend. Similarly, auxiliary components are expected to decrease their share from 15 percent to 9 percent. Logic die costs, however, remain relatively stable, hovering around 13 to 14 percent throughout the entire period. This stability suggests that while the physical demand for logic processing is growing, the rapid scaling of memory capacity is outpacing it in terms of cost impact. These trends highlight the intensifying pressure on the semiconductor supply chain, particularly for High Bandwidth Memory manufacturers. As AI models become more complex and larger, the requirement for faster and denser memory storage becomes a primary cost driver. The data underscores that future budget allocations for AI infrastructure will be disproportionately affected by memory procurement. While packaging and auxiliary parts remain essential, their relative economic weight is diminishing as the industry shifts focus toward overcoming memory bottlenecks. The analysis relies on estimates for production volumes and component pricing, which are subject to market volatility and technological changes. Nevertheless, the projection offers a clear roadmap for the coming year, indicating that the majority of capital expenditure in AI chip manufacturing will be directed toward memory solutions. This trend is expected to reshape the financial landscape for chip designers and suppliers alike, with memory vendors likely to capture a growing portion of the industry's revenue. The shift also has implications for the total cost of ownership for companies building AI data centers, as memory expenses will dominate their hardware budgets. Epoch AI, the source of this data, makes its findings available under a Creative Commons license, allowing researchers and industry analysts to freely use and distribute the insights to further inform the technology sector.