AI’s proportion of global chip consumption is rising — and the rest of the economy is competing for what’s left.
The United States is building the world’s most advanced AI infrastructure. In the process, however, we are overlooking a critical vulnerability: the basic semiconductor components that make that system usable in the real world.
At the center of that vulnerability is Dynamic Random-Access Memory (DRAM), the essential memory chip at the core of nearly every computing device on earth. Demand for these chips is surging — driven overwhelmingly by AI — but supply is not keeping up. According to Bloomberg Intelligence, data centers already account for roughly 50 percent of global chip consumption in 2025 — up from 32 percent just five years earlier. By 2030, AI servers are expected to command a clear majority of the global memory supply. In the near term, Tom’s Hardware reports that data centers will absorb up to 70 percent of all memory chips manufactured in 2026 alone.
That leaves the rest of the global economy — hospitals, manufacturers, automakers, schools, local and federal governments, and the internet service providers carrying America’s digital traffic — competing for a shrinking share of what’s left. The consequences of these shortages are already cascading across industries. Honda has cut North American production by 110,000 vehicles. Micron has called the situation “unprecedented.” Intel told investors there will be “no relief” for the memory shortage until 2028. And Elon Musk has concluded that Tesla’s only option is to build its own memory fabrication plant.
This is what a structural supply failure looks like in real time — not a dip, not a cycle, not a temporary shortage, but a permanent reallocation of global silicon capacity driven by AI’s insatiable demand.
Nowhere is this more concerning — and perhaps more ironic — than in broadband and Wi-Fi infrastructure, the real workhorse of the internet. The routers, gateways, and network equipment that form the backbone of American internet access — the infrastructure that will ultimately carry the tsunami of AI-generated traffic — are being neglected as chipmakers shift production to more advanced AI chips. This dries up the supply and drives up the cost of the chips needed for internet connectivity. According to Counterpoint Research, memory prices for broadband products have increased nearly sevenfold over the past nine months, far exceeding even the tripling of smartphone memory costs.
Memory now accounts for more than 20 percent of the total manufacturing cost of a low-to-mid-range router — up from just 3 percent a year ago. As Counterpoint warned in February, the resulting “memory winter” will slow broadband deployments and drive up the cost of procuring routers and gateways, precisely when broadband providers were planning to accelerate AI-capable network rollouts. At the same time, the cost of the personal devices we use to connect to the internet is also rising due to chip constraints.
In other words, America is accelerating demand for AI while constraining the very networks required to deliver it.
Three companies control more than 90 percent of global memory chip production. All three are redirecting capacity to AI customers who pay a massive premium, and critical infrastructure users of legacy chips are left with few alternatives. And this is exactly where public policy has a critical role to play.
The United States maintains a Strategic Petroleum Reserve precisely because some resources are too critical to national function to be left entirely to market dynamics. Memory chips — the fundamental building block of every networked device, every hospital system, every connected vehicle, and every router in every home and small business — now occupy a similar position in the digital economy and deserve the same treatment.
This chip shortage is solvable when we work together. What this moment calls for is Washington and industry locking arms — chipmakers, communications providers, network equipment manufacturers, and federal agencies — around a shared sense of urgency and a shared commitment to keeping all of America’s critical infrastructure supplied and operational.
The United States can’t lead the world in AI if the infrastructure itself that carries AI collapses under the weight of its own success. AI depends on networks, which depend on devices, which fundamentally depend on memory. And right now, that memory foundation is under strain. The window to act without lasting damage is narrowing and closing. RAM-aggedon is a present emergency, hidden in plain sight, but with a solution well within reach.
The future of AI depends on the fiber-optic and Wi-Fi networks that carry most internet traffic. Without the chips needed to build and maintain those networks, the future of AI in the U.S. goes nowhere at all.