Quanta Computer Warns Power Grid is New AI Bottleneck, Eclipsing Chip Shortages

The primary bottleneck for the AI industry is shifting from semiconductor supply, like GPUs and HBM, to securing adequate power for data centers, a process that can take 2-3 years.

Tech giants like Meta, Google, and Microsoft are fueling this demand with massive investment plans totaling around $710 billion in 2026, intensifying the strain on power grids and electrical equipment supply chains.

This power constraint creates a new risk: even with sufficient AI chips, the pace of building and activating new data centers could slow, potentially delaying revenue growth for the entire tech supply chain.

The artificial intelligence boom is running into a very old-world problem: not enough power.

This reality was brought into sharp focus by Liang Chih-chen, the vice chairman of Quanta Computer, a key manufacturer of AI servers. He described AI demand as climbing a staircase, set to rise steadily into the 2030s. However, he warned that the biggest hurdle is no longer the supply of advanced chips, but the 2-3 year wait to secure electricity for new data centers. This shifts the entire conversation about the AI upcycle.

So, what's causing this power crunch? First, it's the sheer scale of investment from hyperscalers. Tech giants like Meta, Alphabet (Google), Microsoft, and Amazon have collectively earmarked around $710 billion for capital expenditures in 2026 alone. This massive spending translates directly into orders for millions of power-hungry AI servers, placing an unprecedented strain on the world's power grids.

Second, our electrical infrastructure wasn't built for this kind of sudden, concentrated demand. Reports from the International Energy Agency (IEA) and grid operators like PJM in the U.S. highlight significant delays. There are long waiting lists, known as interconnection queues, just to connect new facilities to the grid. Furthermore, critical components like large transformers and switchgear have multi-year backlogs. You can have all the servers you want, but they're just expensive boxes without the power to turn them on.

This doesn't mean the old bottlenecks have disappeared. Advanced packaging like TSMC's CoWoS and high-bandwidth memory (HBM) remain in tight supply. But the power issue has emerged as a more fundamental, longer-term constraint. In fact, newer, more powerful chips like Nvidia's Blackwell series only amplify the problem by packing more computational power—and thus higher energy needs—into each server rack.

Ultimately, the game has changed. For companies like Quanta, the challenge is now less about manufacturing speed and more about coordinating shipments with the customer's ability to energize their new data centers. This new layer of uncertainty is something investors are noticing, which may help explain why valuations for market leaders like Nvidia have seen some compression despite record-breaking demand. The AI revolution will be built on servers, but it will be powered by electrons, and getting them where they need to go is the new grand challenge.

Hyperscaler: Refers to the giant companies that dominate cloud computing and data center infrastructure, such as Amazon Web Services (AWS), Google Cloud, and Microsoft Azure.
Capital Expenditure (Capex): Money a company spends to buy, maintain, or improve its long-term assets, such as buildings, vehicles, and equipment. In this context, it's for building data centers.
Interconnection Queue: A waiting list for projects, like new data centers or power plants, seeking to connect to the electrical grid. Long queues indicate a bottleneck in grid capacity.

Quanta Computer Warns Power Grid is New AI Bottleneck, Eclipsing Chip Shortages

The primary bottleneck for the AI industry is shifting from semiconductor supply, like GPUs and HBM, to securing adequate power for data centers, a process that can take 2-3 years.

The artificial intelligence boom is running into a very old-world problem: not enough power.

Hyperscaler: Refers to the giant companies that dominate cloud computing and data center infrastructure, such as Amazon Web Services (AWS), Google Cloud, and Microsoft Azure.
Capital Expenditure (Capex): Money a company spends to buy, maintain, or improve its long-term assets, such as buildings, vehicles, and equipment. In this context, it's for building data centers.
Interconnection Queue: A waiting list for projects, like new data centers or power plants, seeking to connect to the electrical grid. Long queues indicate a bottleneck in grid capacity.