Despite the recent massive price increases for RAM that have shocked enthusiasts, manufacturers aren’t expected to meaningfully increase production of standard memory to offset the demand from the AI industry. According to TrendForce, memory makers are limiting their capital expenditure on building additional capacity, instead focusing on research and development of process technology, improving density through stacking, and the far more lucrative HBM chips for AI accelerators (HBM requires more wafers to produce than standard memory types).
The three biggest memory chip makers — Micron, SK hynix, and Samsung — are anticipated to spend $54 billion on capital expenditure, but this is focused on improving the performance of HBM modules for AI chips. So, despite the 14% year-on-year growth in investments in the DRAM sector, they will not increase the available bit output of memory chips. Because of this, we can assume the ongoing shortage will continue into next year and well into 2027. In fact, experts say that the massive appetite for AI chips, driven by the infrastructure build-out, will cause a pricing apocalypse that will last a decade.
Even though there is robust demand for both memory and storage chips, especially amid AI companies' multi-year expansion plans, manufacturers are not jumping headfirst into expanding their facilities. Aside from the natural volatility of the semiconductor industry, industry leaders also say that we’re already in an AI bubble, and analysts suggest that it would take at least a $650 billion annual revenue for all the investment into the infrastructure to make sense.
Because of this, memory chip makers are likely hesitant to spend billions on infrastructure, which, if the worst comes to pass, would leave them with multi-billion-dollar investments with practically no demand. And even if they start building new factories right now, it will take a couple of years or more before they can become fully operational and start producing chips for the market.
Instead, they’re focusing their capital outlays on technologies that have a higher ROI and efficiency. To satisfy the demand for these more lucrative chips, memory manufacturers are converting their existing DRAM manufacturing lines instead of building new ones. After all, doing so is much easier, cheaper, and less risky than building an all-new line — but at the expense of reducing the output for consumer memory modules.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.