A fight over the AI memory shortage has moved from supplier roadmaps into Washington policy. SEMI, the semiconductor industry association whose members span chipmakers, equipment suppliers, materials companies, and other manufacturing firms, sent a July 1 letter to senior Trump administration officials warning that attempts to steer memory-chip prices or production capacity could make the shortage worse.
The letter, addressed to Secretary of State Marco Rubio, Defense Secretary Pete Hegseth, Treasury Secretary Scott Bessent, and Commerce Secretary Howard Lutnick, is a response to a widening concern: AI data centers are pulling so much high-end memory into accelerator systems that other industries are worried about getting squeezed out of ordinary DRAM supply.
SEMI’s argument is not that the shortage should be ignored. The group says AI infrastructure demand is placing “unprecedented pressure” on the memory supply chain and outpacing current production. As chipmakers shift wafer capacity from commodity DRAM toward high-bandwidth memory, or HBM, for GPUs and other AI accelerators, SEMI warns that memory used in laptops, smartphones, tablets, cars, appliances, defense systems, and aerospace products will face tighter supply.
The policy dispute matters because memory is no longer just a component-cost problem for PC builders. HBM has become one of the most important constraints in AI infrastructure, sitting beside GPUs, advanced packaging, networking, power, and data center capacity. But the same manufacturers that supply HBM also make the DRAM and NAND that go into consumer devices, telecom equipment, medical devices, industrial systems, and vehicles. When AI customers sign long-term, premium supply agreements, the rest of the market can feel the shortage through higher prices, longer lead times, or delayed products.
Why SEMI is pushing back
SEMI is urging the administration to avoid policies that distort pricing or capacity decisions. Instead, it wants the government to support long-term purchase agreements, expand trade coordination with allies, reduce targeted regulatory obstacles, extend the 48D Advanced Manufacturing Investment Credit beyond 2026, and broaden the credit to cover design, advanced materials, component suppliers, and packaging facilities that support memory production.
The group also wants short-term tax measures that could help affected industries absorb supply-driven price increases. Its examples include consumer-facing credits or deductions for certain electronics, as well as business-focused tax measures meant to keep equipment and materials costs from crowding out capital spending.
The technical point underneath the policy language is straightforward: fabs are not infinitely flexible. HBM uses advanced DRAM dies and packaging, and its growth pulls on upstream materials, chemicals, gases, contamination-control systems, equipment, and packaging capacity. SEMI says overall memory bit-output is projected to grow 19 percent annually, but AI infrastructure demand is rising quickly enough that added output may not arrive where every downstream buyer needs it.
SEMI’s own figures show how much money is already moving toward the bottleneck. The group says worldwide 300mm fab equipment investment in memory is projected to rise 29 percent to $52 billion in 2026, then increase again to $57 billion in 2027. DRAM equipment spending is expected to reach $37 billion in 2026, driven by HBM and DDR5 demand for GPUs and AI accelerators, while 3D NAND equipment spending is projected to reach $14 billion as AI deployment increases data-storage requirements.
The earlier warning came from memory buyers
SEMI’s letter follows a June 3 letter from a cross-sector coalition representing broadband providers, medical device manufacturers, automakers, and retailers. That group warned Treasury and Commerce officials of an “urgent imbalance” in the memory market that could raise household prices and disrupt critical U.S. supply chains.
The June letter was signed by NCTA, ACA Connects, NTCA, the Telecommunications Industry Association, the Medical Device Manufacturers Association, AdvaMed, the Alliance for Automotive Innovation, the Retail Industry Leaders Association, and the National Retail Federation. Their shared concern is that memory allocation is becoming a national supply-chain question, not just an AI industry procurement issue.
Those groups have different exposure points. Broadband operators need memory for network equipment and customer hardware. Automakers cannot ship vehicles if specific electronics modules are missing. Medical-device companies face stricter qualification cycles and cannot always swap memory architectures quickly. Retailers and consumer-electronics sellers feel shortages through higher prices and weaker availability. None of those sectors can easily outbid AI data center buyers if the market’s highest-margin demand is concentrated around HBM and server memory.
That is why the administration faces a difficult choice. A heavy-handed allocation or price-control effort could discourage investment, complicate long-term contracts, or redirect supply in ways that create new shortages. Doing nothing could leave non-AI sectors competing against hyperscale data center buildouts that have stronger balance sheets and more urgent capacity needs.
What readers should watch next
The first sign of whether this becomes a larger tech-policy issue will be how Treasury and Commerce respond. If officials treat the shortage as an industrial-policy problem, the next moves could involve tax credits, permitting changes, trade-agreement work, or sector-specific relief. If they treat it as a market allocation problem, the debate could shift toward priority access for domestic industries or limits on how memory makers structure supply deals.
The second signal is pricing outside the AI data center market. HBM shortages are expected in a GPU-heavy AI buildout, but sustained pressure on commodity DRAM would be more visible to consumers and businesses. That could show up in laptop and phone pricing, delayed network upgrades, higher costs for connected devices, or renewed stress in automotive and medical-device supply chains.
The third signal is whether memory makers can expand without running into upstream bottlenecks. New fab capacity takes years, and SEMI’s letter points beyond chips themselves to materials, specialty chemicals, high-purity gases, equipment, packaging, and regulatory approvals. That means the shortage cannot be solved only by announcing more AI data centers or more domestic fabs. The supporting manufacturing chain has to scale too.
The AI infrastructure boom has already turned GPUs and data centers into boardroom issues. Memory is now joining that list. The harder question is whether policymakers can support supply expansion without breaking the long-term contracts and price signals that memory makers say are funding the expansion in the first place.
