AI compute demand growing 2x faster than Moore’s Law, with $500 billion a year now needed

AI’s appetite for compute is growing so fast that it’s left Moore’s Law choking in the dust. Over the past decade, the demand for processing power has been expanding at more than double the pace the chip industry followed for over fifty years.

According to data from OpenAI itself, the world now needs to spend $500 billion every single year, until 2030, just to keep up. That doesn’t even include GPUs or servers. Just the physical buildings.

Moore’s Law used to define the future, claiming that transistors on chips would double every two years. That future is now dead. Over the last ten years, the compute needed to train and run large AI systems has exploded past that timeline.

The gold standard of tech progress has been crushed by demand that’s gone completely off the rails. Data centers have become the world’s new oil fields, only way bigger, hotter, and far more expensive.

Data centers now cost more than office buildings in the U.S.

By 2028, data center investment will climb to $900 billion globally, driven by an insane 41% annual growth in AI servers and an overall market growth of 23%, according to OpenAI.

The growth is the fastest the sector has ever seen, though all of this building has created a black hole in revenue. There is now an $800 billion funding gap, and just to hit construction goals by 2030, the market needs nearly $2 trillion in revenue.

Construction costs for just the centers alone (without racks, chips, or anything inside) have hit $43 billion a year, up 322% since 2020. Meanwhile, the U.S. has $40 billion worth of data centers currently under construction, a jump of 400% since 2022.

For the first time ever, these unfinished centers will soon be worth more than all U.S. office buildings under development. This is a permanent real estate reset, and it’s all because of compute.

Power is another problem entirely. AI data centers are now projected to eat up 1,600 terawatt-hours of electricity by 2035, which equals 4.4% of all global power use. And that’s not a plateau.

Power demand will quadruple over the next decade. But no one has a solid answer on where this power will come from, or where the money to build the needed energy infrastructure will be found.

Nuclear and quantum enter the conversation as warning bells ring

Some are pushing nuclear energy as the fix, because plants run 24/7, just like the AI systems drawing the power. Others believe quantum computing will help. Quantum machines use qubits instead of bits and can solve some problems exponentially faster than classical systems.

But this is still early-stage tech. For now, the cost keeps climbing, the power grid is strained, and AI keeps growing.

And while all this capital floods into compute, global financial institutions are getting nervous. The International Monetary Fund and the Bank of England have both issued fresh warnings about market instability.

Speaking ahead of next week’s Washington meetings, IMF chief Kristalina Georgieva told investors to “buckle up: uncertainty is the new normal and it is here to stay.” She pointed out that while economic growth will slow only slightly, there are “worrying signs” that investor confidence could collapse if shocks emerge.

Kristalina pointed to the record $4,000 per ounce gold price this week as a sign that fear is already showing up in markets. The IMF head also pointed to U.S. tariffs and sky-high stock valuations as major risks. “As for easy financial conditions — which are masking but not arresting some softening trends, including in job creation — history tells us this sentiment can turn abruptly,” she added.

Over in London, the Bank of England’s latest meeting minutes raised concern over a “sharp market correction.” They flagged weak AI adoption or rising competition as potential triggers that could sink projected earnings for major AI firms.

They aren’t alone. Sam Altman, Jamie Dimon, and Jerome Powell have also joined the caution chorus, warning that the current levels of AI spending could be laying the groundwork for a hard reset.

If you're reading this, you’re already ahead. Stay there with our newsletter.