Are AI Companies Building Sand Castles

In the grand history of technological booms, from railways to the internet, infrastructure usually precedes profit. It takes money to build the future. But the current rush to construct the physical foundations of artificial intelligence is testing the limits of this historical rule, creating what may be the largest financial gamble in the history of the technology industry.

The scale of the investment is staggering. America's biggest technology firms , among them Microsoft, Amazon, and Google , are burning through capital at an unprecedented rate. Last year, they spent some $400bn on AI infrastructure, largely data centers and the specialised chips required to run them. This year, that figure is expected to surpass $500bn. Goldman Sachs, an investment bank, projects a cumulative bill of $7.6trn by 2031.

Yet the financial returns are, for now, insignificant. In 2025, AI services generated roughly $25bn in revenue , a mere tenth of what was spent building them. Data centers are depreciating in value by about $40bn a year, while bringing in less than half of that in sales. Sequoia Capital, a venture-capital firm, calculates that AI companies must earn roughly $600bn a year just to pay for their infrastructure. Put simply, the industry is digging a hole so deep that it needs revenues to grow a hundredfold, to $2trn annually by 2030, just to justify the digging.

Why is the payoff so elusive? The first reason is the illusion of productivity. Although many workers use AI to save a few hours a week, nearly 40% of that saved time is squandered on a "rework tax", correcting errors, smoothing over clumsy prose, and double-checking facts. Most companies have simply dropped space-age tools into outdated job structures.

Furthermore, running these models is eye-wateringly expensive. One executive at Nvidia, the company that designs the computing chips powering the AI boom, recently admitted that the computing cost for his team exceeds the salaries of the employees using it. One firm accidentally racked up a $500m computing bill in a single month after forgetting to set a spending limit. Rather than replacing expensive human labour with cheap software, firms are often doing the reverse.

At the corporate level, the experiment is failing to move the needle. Studies suggest that 95% of corporate AI pilot schemes deliver no measurable boost to the bottom line. Meanwhile, the expensive hardware sits idle; three-quarters of organisations leave their precious computing chips running at less than 70% of their capacity, bleeding thousands of dollars in potential value per chip every year.

Despite these glaring inefficiencies, the spending continues, driven largely by corporate fear. Stopping the investment means ceding the future to rivals; continuing means burning cash with no clear path to profit. Half of chief executives believe their very jobs depend on getting AI right. This fear has pushed the "capital intensity" of the top tech firms, the proportion of cash flow devoured by physical investment , to nearly 100%. They are spending almost everything they earn, and increasingly turning to debt to fund the rest.

This frenzy is now colliding with the physical limits of the real world. The primary concern for building a new data centre is no longer the cost of land, but the availability of electricity. America's ageing power grid is straining, with operators projecting alarming shortfalls in power supply by next year.

The financial foundations of the boom are also becoming unnervingly complex. Chipmakers are taking equity in AI startups in exchange for hardware; leading developers like OpenAI are raising hundreds of billions while bleeding cash. The ecosystem is tightly coupled. A single unravelling thread could cause the whole ecosystem to collapse.

The period between 2026 and 2027 will bring a reckoning. The tech industry has trapped itself in an arms race where the cost of competing is astronomical and the prize remains largely unproven. Ultimately, the winners will not be those who stockpile the most computing power. They will be the companies that can figure out how to deliver actual, efficient value to customers, before the market's patience runs out. Either the software starts generating spectacular revenues, or the world is about to witness the largest write-down of wealth in the history of modern business.

Yet for all the hand-wringing over bloated budgets and absent profits, this frenzy is neither unprecedented nor unnatural; it is the familiar rhythm of technological progress. Whenever a profoundly disruptive technology emerges, a period of wild, capital-intensive experimentation inevitably follows. Businesses and investors rush in, splashing billions in a crazy race to discover the commercial applications that can transform a novel invention into sustainable revenues. This staggering expenditure is not necessarily a sign of collective madness, but rather the steep, unavoidable price of discovery. In the chaotic early stages of a revolution, the market must spend lavishly to test the boundaries of a new tool, searching for the elusive business models that justify the initial outlay.

One need only look to the internet boom of the late 1990s to see exactly how this cycle plays out. During the dot-com era, eager investors poured fortunes into laying vast networks of fibre-optic cables and funding profitless startups, enduring immense losses in a desperate scramble to dominate the digital frontier.

The subsequent crash was brutal, but it served as a necessary, Darwinian correction. It ruthlessly purged the market of speculative fluff and flawed business models, leaving behind abundant infrastructure and the resilient few, firms like Amazon and Google, that would eventually make huge profits. The artificial-intelligence industry is racing toward a similar reckoning, where a painful but healthy collapse will wash away the misfits, leaving only those who have truly cracked the code of profitability to inherit the future.

Ultimately, the artificial-intelligence revolution will remain a glorious, but a huge loss-making experiment until the same brilliant minds who engineered the technology can master a far older art: figuring out how to generate profit out of it.