'Our GPUs Are Melting,' OpenAI's Altman Warns — Could US Natural Gas Power AI Revolution?

The artificial intelligence boom is now colliding with America's aging energy infrastructure, but natural gas may be the unexpected wildcard that keeps the AI revolution from overheating—literally.

OpenAI CEO Sam Altman issued a blunt warning Thursday on social media X: "Our GPUs are melting." The comment came as OpenAI imposed temporary usage limits amid surging demand for image generation in ChatGPT.

As tools like ChatGPT become integral to business operations, the data centers powering them are drawing electricity at unprecedented rates. Now, the race is on to secure reliable electricity supplies.

AI Sends US Electricity Demand Soaring: Could Natural Gas Step In To Fill The Gap?

Surging AI adoption is poised to add unprecedented pressure to the U.S. power grid, with data centers emerging as the primary energy guzzlers, according to a report released this week by Natural Gas Intelligence titled ‘AI Rising: How Will Natural Gas Markets Evolve to Fuel the Demand Surge?’

The International Energy Agency projects that electricity demand in the United States through 2027 will rise by an amount equivalent to California's entire current power usage. That's a surge of roughly 400 terawatt-hours (TWh) in just three years.

Wood Mackenzie estimates total U.S. electricity consumption will soar from 4,000 TWh in 2024 to 6,000 TWh by 2050. Of that, between 395 and 660 TWh could be consumed by data centers alone by 2035—more than 10% of total power usage.

The fast pace of AI deployment is leaving little room for slow build-outs of renewable energy or long-term infrastructure upgrades. That's where natural gas steps in.

"Gas will certainly get its share of the new AI-driven power demand given its reliability, which is a must for these facilities," said Patrick Rau, senior vice president of strategy and research at Natural Gas Intelligence, according to the report.

Natural gas provides firm and flexible power, a critical advantage for hyperscale data centers that operate 24/7. Rau said the key bottleneck is how quickly new gas-fired capacity can come online.

Utilities And Oil Giants Jump In

Power producers are already rushing to meet the call. NRG Energy Inc. NRG is fast-tracking 1.5 gigawatts (GW) of gas-fired projects in Texas. CEO Larry Coben said the company has signed multiple letters of intent with data center developers and is advancing 10 natural gas plants totaling about 8 GW in generation capacity.

In February, NRG also partnered with GE Vernova Inc. GEV and construction firm Kiewit Corp. to develop up to 5.4 GW of new capacity explicitly linked to AI data centers.

Other utilities are following suit. American Electric Power Inc. AEP is weighing $10 billion in potential investments to meet what it calls "incremental load," largely driven by AI, manufacturing reshoring and general economic development. The company already has customer commitments for 20 GW of additional load by 2030 across its 11-state footprint.

Oil giant Chevron Corp. CVX is another unexpected entrant in the AI race. In partnership with investor Engine No. 1 and GE Vernova, the energy giant plans to build co-located power plants across the Midwest, Southeast and West to support data centers with dedicated energy generation.

Will DeepSeek Reduce AI Data Centers Demand?

Chinese AI company DeepSeek has introduced a platform that claims to rival current models in performance—at a fraction of the power cost. That innovation, while promising, may add another twist to the AI power narrative.

"DeepSeek's improved computational efficiency may provide a pathway to quell hyperscalers' power requirement," according to a recent report by Rystad Energy experts.

"But it could also escalate the AI arms race and compel companies toward more power-intensive innovation."

For now, the message from the grid is clear: keep the pipelines flowing and the turbines spinning at full throttle.

Read Next:

• S&P Says US Growth To Slow In 2025, But Recession Not ‘Most Likely Outcome Yet’

Image created using artificial intelligence via Midjourney.