HEATH MUCHENA | The market is mispricing the real bottleneck in AI - TimesLIVE

For all the talk about artificial intelligence (AI) replacing lawyers, programmers and office workers, the most important thing about AI right now may be surprisingly old-fashioned.

It needs an extraordinary amount of stuff. Chips. Electricity. Memory. Land. Cooling systems. Transformers. Data centres. Construction workers. Financing.

We still talk about AI as though it were another software revolution, but economically it is starting to look more like an industrial one. That distinction could matter enormously for investors.

The debate over whether AI is a bubble has become too simplistic. Parts of the market may certainly be overpriced. History suggests transformational technologies and speculative excess frequently arrive together. Railroads changed economies and produced railway manias. The internet changed the world and still gave us the dot-com crash.

AI can be revolutionary and contain bubbles at the same time. The better question is where the bottleneck lies. The largest technology companies are spending sums on AI infrastructure that would once have been associated with national infrastructure programmes.

AI is turning electricity into something economically useful that did not previously exist at this scale: machine intelligence. That creates a new kind of commodity chain

Nvidia’s data centre business has reached a scale that would have seemed implausible only a few years ago. Microsoft, Meta, Amazon and Google are pouring tens of billions of dollars each quarter into computing capacity. Eventually that spending must earn a return. But something more interesting is happening underneath it.

AI is turning electricity into something economically useful that did not previously exist at this scale: machine intelligence. That creates a new kind of commodity chain. Models sit at the top, but beneath them are graphics processing units (GPUs), advanced memory, networking equipment, data centres and electricity. If any one of those becomes scarce, economic value can migrate toward it.

This is familiar commodity economics. When shipping capacity is scarce, freight rates rise. When oil supply tightens, producers capture the scarcity rent. When semiconductors are scarce, fabrication capacity becomes unusually valuable. AI should be viewed through the same lens.

The most important question may be: what prevents the next dollar of AI demand from being served? Today the answer could be GPUs. Tomorrow it may be high-bandwidth memory. In another region it could be electricity, transformers, grid connections or simply permission to build.

There is also a paradox at the heart of the AI boom. Models are becoming more efficient and inference is becoming cheaper. Normally, efficiency reduces demand for a resource. AI may do the opposite.

Investors have spent three years asking which AI model will win. The bigger opportunity may be identifying what the AI economy cannot build fast enough

When something becomes dramatically cheaper people often use much more of it. A company that once made 100 expensive AI requests may make 100,000 when the price collapses. Autonomous agents could increase that consumption again by orders of magnitude.

Cheaper intelligence could therefore produce greater demand for computing infrastructure, not less. Open-source AI creates a similar possibility. Cheaper models could pressure the margins of frontier AI companies while expanding total usage. Value would then migrate down the stack toward compute, memory, networks and power.

This does not make the boom risk-free. The danger comes if AI revenues fail to catch up with the capital being committed. Rising interest rates or widening credit spreads could make data centre projects uneconomic. GPU utilisation could fall. Electricity constraints could trigger political opposition. A genuine breakthrough in efficiency could eventually create excess capacity.

That is why falling AI stocks alone will not tell us whether the boom is over. The more useful signals may be:

Investors have spent three years asking which AI model will win. The bigger opportunity may be identifying what the AI economy cannot build fast enough. AI may ultimately make intelligence abundant. Getting there could make physical scarcity one of the defining investment stories of the decade.

• Muchena is founder of Proudly Associated and author of Artificial Intelligence Applied and Tokenized Trillions.