The U.S. Needs More Memory Chips, More Free Trade, and Market Flexibility

Imagine a tiny piece of technology which temporarily stores the information a processor needs to function. That in a nutshell is a dynamic random-access memory (DRAM) chip, a memory chip that helps your phone run an app, your refrigerator keep food cold, and ATMs process transactions. Memory chips are easy to overlook, but they are embedded in much of the technology that keeps the economy running.

Today, demand for memory chips is surging as artificial intelligence (AI) and data centers require increasingly large amounts of advanced memory. While most of the economy relies on traditional DRAMs, data centers require High Bandwidth Memory (HBM), a specialized form of DRAM that stacks multiple memory layers to move data quickly between memory and processors. This infrastructure is raising demand for memory—data shows data centers accounted for 32 percent of global DRAM consumption in 2020, rising to roughly 50 percent in 2025 and projected to exceed 60 percent by 2030. As manufacturers shift more capacity toward HBM, less capacity is available for conventional DRAM used in laptops, smartphones, appliances, and other products.

But the problem is not simply that the U.S. needs more memory capacity. The bigger challenge is that the market cannot respond to this surge in demand overnight. Markets must be permitted to respond and meet the demands of the moment.

Making memory chips is a highly complex process, which means existing manufacturers cannot simply expand production when demand rises. It requires huge amounts of capital, specialized equipment and materials, advanced packaging, and years of technical expertise. Manufacturers also depend on a specialized network of suppliers for the equipment, materials, and components needed to make those chips, which adds more complexity when they need to expand production. Today, three companies lead the market, including Micron in the United States and Samsung and SK Hynix in South Korea—accounting collectively for more than 90 percent of the market. This market leadership, combined with the industry’s accumulated expertise and supplier relationships, has built out the basic investment and infrastructure needed to meet demand.

But building new semiconductor manufacturing facilities (fabs) is a slow and expensive process. A TSMC executive noted it takes two to three years to build a fab, and another one to two years to boost production. And expanding capacity requires massive investments. For example, SK Hynix recently approved a $38.3 billion investment in two new memory fabs. But even these investments will take years to translate into additional supply. Existing manufacturers can shift some capacity toward HBM but doing so puts pressure on conventional DRAM rather than adding new supply.

These conditions create a timing problem. Even when demand changes quickly, physical capacity cannot follow, and new supply can take years to arrive.

Policymakers have already recognized the importance of chip supply, and have attempted to bolster supply through legislation such as the Creating Helpful Incentives to Produce Semiconductors (CHIPS) and Science Act. But taxpayer subsidies for new fabs are often poorly targeted, and perhaps most importantly, cannot quickly fix a shortage overnight if demand is raising faster than new factories can be built. The goal should therefore be to make it easier for manufacturers to respond when demand changes.

That means keeping memory chips, manufacturing equipment, and other inputs moving across borders and avoiding unnecessary trade restrictions that could further reduce supply. It also means making it easier for manufacturers to substitute components when safe alternatives are available, particularly in industries where regulatory or certification requirements can make switching suppliers slow and costly. One example is restrictions to PFAS, which could create challenges for semiconductor manufacturers because these chemicals are used in chip production, and alternatives are not yet available for all uses.

As AI reshapes demand for memory faster than a heavily regulated industry can adjust, the lesson is not simply that the economy needs more chips. It is that when demand changes quickly, supply needs room—and market flexibility—to respond.