How the Global AI Boom Is Triggering a Memory Chip Crisis Outside Silicon Valley

Medical-device manufacturers, carmakers, and consumer electronics groups, including Apple, are now lobbying the US government for help securing memory supplies. Some want Washington to require chipmakers receiving federal support to supply a wider range of industries, while others are pressing for the use of the Defense Production Act - a Korean War-era measure that would allow the government to direct production towards sectors outside AI.

What began as a supply-chain problem is becoming a test of how far the US government is prepared to intervene in a market that has acquired strategic importance.

The stakes are unusually high because the administration faces two competing objectives. It wants to keep consumer prices under control, but it also wants to preserve the enormous investment flowing into AI infrastructure - an industry increasingly viewed as part of the US competition with China.

AI can afford what everyone else cannot

The imbalance comes down to economics as much as technology.

AI companies building data centres are willing to pay prices that other industries cannot absorb. Dan Kim, chief strategy officer at TechInsights and a former chip strategist in the Biden administration, expects shortages to persist for another two years. He said there was effectively no ceiling to what AI companies would pay for memory.

For manufacturers producing MRI machines, cars or consumer electronics, there is a very different limit.

That divergence is beginning to feed into the wider economy. Diane Swonk, chief economist at KPMG US, has warned that memory chips are already contributing to US inflation. Consumers have begun reducing purchases of products such as video-game consoles, while higher prices for other electronics are likely to follow.

The economic problem is therefore not simply that semiconductors are expensive. It is that one of the world's fastest-growing industries is capable of outbidding industries that provide products with very different social and economic functions.

“AI boom” has often been shorthand for investment, productivity and technological leadership. The memory shortage introduces another dimension: AI's growth is imposing costs on businesses that are not part of the AI economy.

That is why the lobbying campaign has expanded so quickly.

Washington is being asked to choose who gets scarce supply

The proposals circulating in Washington reveal how unusual the situation has become.

Some industries want access to memory chips to become a condition for companies receiving money under the 2022 CHIPS and Science Act. Others want the Trump administration to use the Defense Production Act to ensure supplies reach sectors outside AI.

Such measures would represent a significant shift from the normal role of industrial policy. Instead of simply encouraging companies to build more factories, Washington would be influencing who gets the output of those factories.

That is a much more consequential intervention.

It would also force the government to make an uncomfortable judgement about priorities. Should scarce memory go first to a data centre supporting America's AI advantage, a medical device manufacturer producing MRI machines, a carmaker or a consumer electronics company?

The political difficulty is obvious. The administration wants to demonstrate that it can bring inflation under control while simultaneously protecting an investment cycle that depends on extraordinary demand for computing infrastructure.

Tim Cook, Apple's chief executive, described the movement in memory prices last month as a “100-year flood”, pointing to the scale of the increase.

The White House has not indicated whether it will intervene. Kush Desai, a White House spokesman, said domestic semiconductor manufacturing remained a top priority for President Donald Trump and pointed to hundreds of billions of dollars in investment secured by administration policies.

But investment today does not solve a shortage tomorrow.

Building American capacity takes years

The industry's structural problem is that memory production cannot be expanded at the speed of demand.

Micron Technology has pledged more than $250bn to the US memory sector, according to company executive Manish Bhatia. Samsung and SK Hynix have also committed hundreds of billions of dollars to expanding production in the US and elsewhere.

The problem is time. Semiconductor factories are extraordinarily expensive and take years to build. Bhatia argues that US projects take longer to complete than facilities elsewhere and says the country needs faster permitting and a stronger workforce.

That creates a gap between Washington's industrial ambitions and the immediate needs of manufacturers.

The US has spent years trying to rebuild domestic semiconductor production. Both the Biden and Trump administrations have encouraged companies to invest locally. Yet a factory announced today cannot immediately relieve a shortage that is already pushing prices higher.

The industry therefore faces a choice between short-term allocation and long-term capacity.

SEMI, a semiconductor industry association, has resisted the more interventionist proposals. It has argued against government efforts to influence memory prices or determine allocation, instead favouring trade agreements, regulatory changes and an extension of a tax credit designed to encourage domestic manufacturing.

The industry's preference is understandable. Government allocation could distort a market that companies are already trying to expand. But simply waiting for new factories to come online leaves other industries exposed to the purchasing power of AI companies.

China offers supply - and a strategic trap

That tension has revived an option Washington has spent years trying to constrain: buying memory from China.

Samsung and SK Hynix produce much of their memory in South Korea and China, and companies affected by the shortage have considered increasing production at their Chinese facilities. Apple and other consumer electronics companies are also pressing for permission to purchase Chinese memory despite existing restrictions.

For Apple, the issue is particularly politically sensitive. The company had explored buying memory from China's Yangtze Memory Technologies Corporation, or YMTC, in 2022, before the effort collapsed amid pressure from US lawmakers.

Now the proposal has returned in a different market environment.

Apple has cited the memory shortage when explaining recent price increases and is seeking access to additional supplies. But many US officials and lawmakers see Chinese memory not as a convenient solution to a temporary shortage, but as a potential strategic dependency.

YMTC was placed on the US entity list in 2022, restricting its access to American technology. ChangXin Memory Technologies, or CXMT, has also come under scrutiny. Both companies are subject to restrictions affecting their use in products sold to the US government.

The political response has been bipartisan. In July, senators urged Cook to commit to avoiding Chinese memory. John Moolenaar, the Republican chairman of the House Select Committee on China, and Democrat George Whitesides separately urged Commerce Secretary Howard Lutnick to add CXMT to the entity list and restrict purchases from companies placed there.

The argument is not simply about today's supply.

Chris Miller, author of Chip War, has warned that buying more Chinese memory could create longer-term dependency if companies establish durable commercial relationships with Chinese suppliers.

That is the strategic dilemma Washington now faces. Restricting Chinese memory protects the logic of semiconductor security but can intensify the shortage and raise costs for American companies. Allowing greater access could ease pressure on consumers and manufacturers while undermining efforts to build a more secure domestic supply chain.

The shortage exposes the limits of industrial policy

The memory crisis is revealing a contradiction at the heart of America's semiconductor strategy.

Washington wants domestic production, technological leadership over China and resilient supply chains. It also wants affordable consumer goods and an AI industry capable of expanding without constraint. Those objectives can reinforce one another when supply is plentiful. They collide when a critical component becomes scarce.

The temptation will be to treat the problem as another manufacturing bottleneck that can be solved with subsidies and faster permits. But the immediate question is harder: who should receive the chips that exist now?

If AI companies continue to pay almost any price, conventional manufacturers will have to absorb higher costs, delay production, or pass those costs to consumers. If Washington intervenes to redirect supply, it will be choosing where the economic gains from America's AI boom should be allowed to flow - and where its costs should land.