TL;DR — Key Takeaways
- Infrastructure shortages will constrain AI growth. IDC expects limits on power, memory and advanced chip packaging to persist through 2027, with memory shortages continuing into 2028.
- Lower AI costs will drive greater consumption. IDC forecasts a roughly 95% decline in cost per agent action between 2025 and 2030, but expects increased usage to push total spending higher.
- Capacity constraints will reshape AI deployment. Open-weight models, private clouds, sovereign infrastructure and edge computing will help spread workloads, while specialized AI cloud providers face consolidation.
The AI industry faces a prolonged shortage of computing infrastructure as enterprise adoption accelerates, with limits on power, memory and chip production expected to restrain growth through 2027, according to new research from IDC.
The research firm says that concerns about an AI investment bubble overlook the fact that demand for AI computing resources continues to exceed available capacity.
IDC forecasts that 2.5 billion AI agents will be deployed by 2030, creating massive demand for the infrastructure needed to run autonomous applications. Already, 95.3% of enterprises have at least one AI agent workflow in production, with an average of approximately 11 workflows per organization.
This widespread adoption is driving spending beyond initial forecasts. IDC found that 67% of enterprises exceeded their AI agent budgets over the past year, including 24% that overspent by at least 25%.
Not that any of this will slow down spending: IDC expects lower costs per AI action to encourage far greater usage, driving total spending higher despite improvements in efficiency.
Falling AI Costs Won’t Reduce Spending
The firm predicts that the cost of performing an individual AI agent action will decline approximately 95% between 2025 and 2030. However, the total cost of delivering AI tokens is forecast to grow an enormous 487-fold during that period.
The reason is that lower operating costs enable businesses to deploy far more AI applications, offsetting savings from improved efficiency. IDC estimates that model routing, which directs AI tasks to the most cost-effective models, can reduce token costs by 30% to 60%.
Among the most serious challenges for the AI buildout is the availability of electricity for data centers. IDC projects a 45-gigawatt power shortfall in the U.S. by 2028. Nearly 7 gigawatts of planned U.S. data center capacity for 2026 has already been delayed or canceled because sufficient power was unavailable.
Globally, data center electricity consumption is forecast to climb from 519 terawatt-hours in 2025 to 1,308 terawatt-hours by 2030.
Semiconductor shortages present another obstacle. IDC expects global DRAM supply to grow just 16% in 2026, while NAND supply expands 17%. Advanced chip packaging is also constrained. Although advanced chip packaging capacity is expected to expand approximately 80% in 2027, demand for this capacity is forecast to nearly double.
These limitations are part of what’s driving massive infrastructure investment. IDC forecasts that annual AI infrastructure buildout by service providers will zoom upward from $401 billion in 2025 to $1.45 trillion in 2030.
Spending on data center facilities, excluding IT hardware, is expected to rise from $276 billion to $655 billion.
The capacity shortage is also reshaping where businesses deploy AI. IDC found that 31% of companies have moved some workloads from proprietary AI models to open-weight alternatives over the past year. IDC expects open-weight models to support greater AI deployment in private clouds, sovereign infrastructure and edge computing, shifting some workloads away from public cloud.
IDC forecasts that AI infrastructure spending by enterprises and other non-service providers will leap from $28.6 billion in 2025 to $96.4 billion by 2030.
The research firm also expects consolidation among specialized AI cloud providers as infrastructure costs and financing pressures reshape competition. In other words, neoclouds may face challenges ahead.
