AI is hitting a wall, and the wall is power. Gas is the only thing breaking through

Last month Sam Altman made a comment that should have set off alarms across the entire energy sector. He said most new AI data centres will be powered by natural gas, at least for now.

People focused on the AI angle, but the real story is energy. Altman is effectively telling the market that the limiting factor for AI is not GPUs, it is electrons. And the only scalable, dispatchable source that can be rolled out fast enough is gas.

Here is what that means.

1. Natural gas is the only firm power that can scale on AI timelines

AI data centres are not small. The next generation builds are running into the hundreds of megawatts per site, with entire campuses creeping toward multi-gigawatt demand. US analysts are now forecasting triple-digit percentage increases in data-centre load this decade.

The bottleneck is not technology. It is the grid.

• Nuclear has multi-decade lead times.

• Renewables are constrained by land, intermittency and slow transmission approvals.

• Battery storage will help but cannot support baseload AI clusters running continuously.

• Grid interconnect queues in the US are already years long.

Gas turbines are the only thing that can be deployed at the speed the AI sector requires.

2. AI power demand is already outstripping supply

Multiple operators are reporting situations where they have GPUs on pallets that cannot be energised because there is not enough power at the data-centre site. Major US utilities are flagging structural power deficits later this decade as AI-related demand ramps faster than new generation can be built.

This is why Altman is leaning on natural gas. He is not making a philosophical statement. He is stating the physical constraint the AI boom is running into. If the aim is to stand up massive compute clusters quickly, gas is the only practical pathway.

3. Implications for Australian investors

Australia will not host the bulk of global AI compute, but the global LNG market ties us directly into the same demand curve. If the US, Middle East and parts of Asia accelerate gas-to-power builds for AI, that increases competition for LNG and reinforces the value of long-life Australian gas assets.

Likely beneficiaries over a 5 to 10 year horizon include:

• Upstream gas producers with low-cost, flexible supply.

• Infrastructure operators positioned for expanded gas transport and storage.

• Electricity gentailers that can justify new gas-fired firming capacity on the back of higher baseload requirements.

If the US oil majors have started building dedicated gas plants specifically to power AI data centres, that should be taken as a forward indicator of medium-term gas demand.

4. Do not confuse a decade-long boom with a permanent structural shift

Altman is also heavily invested in advanced nuclear. It is clear that his long-term view is a transition away from gas once new nuclear and storage-heavy renewable systems mature. That timeframe is not short, and gas will dominate until then, but investors need to understand the cycle.

This is a window, not a forever trend.

• Gas dominates the next 5 to 15 years.

• Beyond that, political and technical pressures will force a shift toward lower-carbon baseload.

• Any company sinking capital into multi-decade gas assets without acknowledging that risk is ignoring the longer-term trajectory.

5. The takeaway

AI is not floating in the cloud. It is anchored in real-world energy systems. And right now, the only energy source that can keep up with AI scale is natural gas.

If Altman is right, the world is heading toward a structural gas squeeze over the next decade. The winners will be companies that can supply flexible, reliable gas into global markets without over-extending themselves into assets that will face headwinds once the nuclear and storage technologies reach commercial maturity.

Short version: AI needs power. Power needs gas. Anyone ignoring that linkage is not paying attention.

Not financial advice. Use your own judgement.