According to Bloomberg, the price of the AI server system is increasing in part, with some products increasing by more than 15 per cent. This adjustment is mainly for equipment delivered early next year, and price changes will vary depending on the chip intergenerational and memory configurations.
This change shows that AI ' s cost pressure from infrastructure expansion is being further transferred from the chip link to the complete machine and data centre construction. The products affected included parts of a system based on the Vera Rubin and Grace Blackwell platforms, for which the server manufacturer for clients of large data centres such as Microsoft, Google and Oracle had received notification.
Memory costs become critical stress points
The increasing demand for DRAM and high-bandwidth memory (HBM) by AI servers has made the whole machine increasingly vulnerable to memory costs. With the increase in the number of accelerators on a single server, memory has become an important factor affecting delivery and pricing.
Currently, Samsung electrons, SK Hercules and Light dominate the global supply of DRAM. As the AI data centre continues to expand, more capacity is used for high-performance computing scenarios. U.S.A. previously indicated that the industry could still supply significantly less than the demand in 2026 and some time thereafter, as the AI task load was pushing up the memory of a single server.
Data centre construction costs continue to rise
The increase in memory prices not only affects the GPU server offer, but may also further increase the overall construction cost of the data centre. Currently, inputs from GPU, network equipment, power infrastructure and heat-dispersion systems are at a high level, and new calculator deployment costs are likely to continue to rise if critical components remain strained.
Reuters had previously reported that US$ 10 billion had been invested in US$ 10 billion to build a new R & D facility in Boysi, United States, focusing on the deployment of next-generation storage technologies. This also reflects the continuing expansion of market demand for high-performance memory.
Pre-fiscal markets are concerned with changes in profitability
The price adjustment took place before the forthcoming new season of the New Year's Journal, and the market became more concerned with its profitability performance. Young Weida recorded $81.6 billion in collections during the last financial season, with $75.2 billion in income from data centre operations and 75 per cent of the non-GAAP Māori rate.
If Young Wida transfers more of the memory price increase pressure to the customer rather than self-absorption, it may help to maintain the level of profitability. At 5 p.m. Eastern U.S. time on 26 August, the company will hold a call for the second fiscal year of 2027, at which time the market will further observe its cost transfer and order.
Last Friday, the share price of Weeda was about 214.77 dollars, falling back a few days. What is more of concern for its large clients is not short-term equity prices, but the deployment costs of the next generation of AI infrastructure may remain high if memory supplies continue to be tight.