Jeff Bezos says AI firms are wasting billions by building their own data centres

Jeff bezos

• Bezos: Private AI data centers are flawed and unsustainable
• AI infrastructure will move to centralized cloud platforms for better efficiency.
• AI data centres may match Japan's electricity use by 2030

Did our AI summary help?

Jeff Bezos has warned that the current race among technology companies to build their own data centres for artificial intelligence is fundamentally flawed. Speaking at the 2026 New York Times DealBook Summit, Bezos said the approach mirrors a mistake made by industrial companies more than a century ago.

According to Bezos, companies building private AI infrastructure today resemble early 20th-century factories that generated their own electricity before public power grids became widespread. At the time, factories ran their own generators because there was no alternative. Once centralised grids emerged, that model quickly disappeared.

“Right now, everybody is building their own data centre, their own generators essentially. And that’s not going to last,” Bezos said. He pointed to a visit to a 300-year-old brewery in Luxembourg that once operated its own power plant as an example of how inefficient decentralised energy production eventually became.

Bezos argued that AI infrastructure will follow the same path, shifting away from isolated projects towards large-scale cloud platforms such as Amazon Web Services. Centralised cloud providers, he said, can deliver far better efficiency, utilisation and cost control than companies running their own facilities.

The comments come as AI-driven demand for electricity is rising sharply. Data centres consumed around 415 terawatt-hours of electricity globally in 2024, roughly 1.5 percent of total demand. By 2030, that figure is expected to reach about 945 terawatt-hours, close to the total electricity consumption of Japan. In the US alone, data centres could account for as much as 12 percent of national electricity demand, up from around 4 percent today.

AI workloads are significantly more energy-intensive than traditional computing. A single query to a large language model such as ChatGPT can consume nearly ten times the energy of a standard web search. Training one large AI model can use as much electricity in a year as around 200 average US homes.

These pressures have triggered a global scramble for power. Technology companies are pursuing nuclear energy deals, renewable energy projects and even speculative ideas such as space-based data centres to secure long-term supply.

Bezos is not alone in highlighting energy as the core constraint for AI. Satya Nadella has said that a lack of available power is leaving GPUs idle, while Sam Altman has argued that advances in fusion or low-cost solar are needed to sustain AI growth. Sundar Pichai has also described electricity as the long-term bottleneck for the industry.

Despite these warnings, major players continue to invest in private infrastructure. Meta has secured nuclear power capacity to support its Prometheus AI supercluster in Ohio, while Alphabet recently acquired Intersect Power to develop combined energy and computing projects.

Bezos framed AI as a horizontal technology, comparable to electricity itself. Just as industry eventually standardised around shared power grids, he believes AI computing will consolidate around large cloud providers rather than fragmented, company-specific data centres.