The chief executives of Intel and Broadcom have both warned that the pace of artificial intelligence expansion now hinges on electricity supply rather than semiconductor manufacturing capacity, India Today reports.

The message from two rival CEOs — one leading the company that pioneered x86 processors and co-invented the modern foundry-model contest, the other supplying custom AI accelerators and networking silicon to the largest hyperscalers — carries weight precisely because it comes from opposite ends of the AI silicon stack. When the people whose job is to ship more chips say chips are not the binding constraint, buyers and investors tend to recalibrate.

The warning reframes a debate that has dominated the semiconductor industry since generative AI demand began consuming wafer capacity at record rates. For most of the past two years, analysts, executives and policymakers have treated advanced packaging, high-bandwidth memory and leading-edge logic as the scarce inputs limiting AI infrastructure buildouts. The two CEOs now point at a different chokepoint: the grid.

Why does the power argument matter commercially?

Data center power consumption has become the operational ceiling for AI expansion. Hyperscalers that once competed on GPU allocation now compete on megawatts — securing interconnects, land, and long-term power purchase agreements before placing silicon orders. If electricity, not silicon, sets the pace, then the value chain shifts:

• Utility-scale generation and grid interconnection timelines move to the critical path of AI revenue
• Power-efficient silicon, advanced packaging and interconnect gains command a pricing premium
• Site selection for new data centers increasingly follows energy availability rather than network topology

For chipmakers, the implication is subtle but material. Demand visibility for accelerators and server CPUs now depends on customers' ability to energize facilities, which introduces a non-semiconductor variable into order forecasts.

What does this mean for the supply chain?

Neither company is signaling weakness in its own product lines. Intel sells data center CPUs and is racing to build out its foundry business, where AI-related demand for advanced nodes is a central pillar of the turnaround plan. Broadcom designs custom AI accelerators for major cloud providers and supplies the networking chips — switch silicon, optical interconnects — that move training and inference traffic inside AI clusters. Both companies' orders remain tied to data center construction schedules.

The CEOs' shared concern is that those schedules slip when power does not arrive on time. A fab can be built in roughly two to three years; grid infrastructure, transmission approvals and new generation frequently take longer. That mismatch, not wafer output, is what they identify as the drag on AI growth.

Is this a shift from the shortage narrative?

Yes, and a significant one. The industry's recent history — pandemic-era shortages, allocation regimes, multi-year capacity expansion announcements — trained customers to treat semiconductor supply as the scarce resource. Governments reinforced that framing with subsidy programs aimed at reshoring leading-edge manufacturing.

The CEOs' position does not dismiss those pressures. It ranks them. In their assessment, the binding constraint has moved downstream from the fab to the substation.

For component suppliers, that ranking affects roadmap prioritization. Performance-per-watt becomes a sales argument rather than an engineering footnote. For customers, it affects total cost of ownership models, where energy now rivals capital expenditure as the dominant lifetime cost of AI infrastructure.

What happens next?

If the power bottleneck persists, competition among silicon vendors will increasingly be fought on energy efficiency per workload, and data center siting decisions — not chip allocation — will determine how quickly AI capacity actually comes online.