AI’s trillion-dollar sustainability question
The technology is dazzling, but the economics look less so
Would you use ChatGPT or Gemini if it cost you Rs 20,000 a month? I suspect the answer from most readers would be no. We are accustomed to thinking of software as inexpensive, if not free, services, including email, social media, maps, cloud storage, and video calls, all of which cost almost nothing to use. AI seems like just another such product. Yet the uncomfortable truth is that the cost of delivering AI is not in the same league at all.
More than a year ago, in this space, I wrote about what I called “the unpredictable revolution of AI”. My conclusion then was that while AI is undoubtedly revolutionary, revolutions never proceed along neat and predictable lines. One big reason is economics: even if the technology works brilliantly, that doesn’t mean it will be affordable or profitable at scale.
This has become clearer with time. I recently read an article that drew a parallel with the late 1990s. Back then, companies like Global Crossing laid tens of billions of dollars’ worth of undersea fibre cables. The infrastructure was essential; the internet could not have grown without it, and much of it remains in use today. Yet the companies went bankrupt because the business case didn’t exist. Customers weren’t willing to pay the price, and investors were wiped out.
This is not a new story, and even fibre-optic cables were not the first time; railways were much like this. The technology was world-changing and laid the foundations of the modern economy. Yet the first wave of investors, who piled into the thousands of new railway companies during the British “railway mania” of the 1840s, mostly did badly. A handful of big networks eventually flourished, but by then the early shareholders had long since been wiped out or diluted. Even when a technology is destined to succeed, the first generation of investors often fails to capitalise on it.
Today’s AI looks similar. In 2025 alone, global spending on AI data centres is expected to reach $400 billion. These facilities depreciate rapidly, with chips lasting three to five years, and other equipment lasting about a decade, incurring $40 billion in annual depreciation. Against this, revenues are only around $ 15–20 billion. To break even, revenues would need to grow ten times as much. To generate even modest returns, it would need to hit $480 billion — five times Microsoft Office’s and twelve times Netflix’s. It’s a number that borders on fantasy.
The difficulty is not technical but economic. Users expect AI to be priced similarly to consumer apps, but the underlying costs are of a different order. This is the heart of the problem. Users often think of AI as another software tool, similar to email or a streaming service, which should cost a few dollars a month or even be free. However, the costs of delivering AI are on a completely different scale altogether. The capital intensity is enormous, and for the companies leading this race, the stakes appear to be existential. No one can afford to fall behind, so each continues to invest billions in data centres and chips, even as returns appear increasingly negative. The race is driven by fear rather than a clear understanding of sustainable profits. Of course, some of the infrastructure being built today will turn out to be indispensable, but that won’t be much of a consolation for those paying the bills now.
In my earlier column, I argued that AI is a revolution, but not one that will unfold along the straight lines the hype suggests. That seems even truer today. Either prices will rise sharply, slowing adoption, or spending will have to slow, limiting progress.
And this brings us back to the topic of investing. Right now, investors--whether small retail shareholders in tech stocks or the biggest venture capitalists in Silicon Valley--seem dazzled by any mention of AI in a company’s plans. However, being dazzled is not a sound investment strategy. We could well be in the 1998 of AI, with the equivalent of the dotcom crash just beyond the horizon. The technology will endure, but that does not mean today’s investors will prosper.
@nikhilkamathcio There are plenty of makhana in stores in North America but the sales are very low.
It has no taste and costs a bomb compared to any snack here.
It is like Indian version of Avocado.
@nkumar1212@rpopatna We sent an email seeking confirmation on 10.07.2024 and, as a reminder, again on 09.10.2024. We are awaiting confirmation from RPO Patna.
@rpobengaluru . we are getting the earliest appointment with enquiry officer for my passport renewal issue in July. Can you prepone it for May as it is a bit urgent
@rpobengaluru . we are getting the earliest appointment with enquiry officer for my passport renewal issue in July. Can you prepone it for May as it is a bit urgent
@rpobengaluru . we are getting the earliest appointment with enquiry officer for my passport renewal issue in July. Can you prepone it for May as it is a bit urgent
Fast, lift, sprint, stretch, and meditate.
Build, sell, write, create, invest, and own.
Read, reflect, love, seek truth, and ignore society.
Make these habits. Say no to everything else.
Avoid debt, jail, addiction, disgrace, shortcuts, and media.
Relax.Victory is assured.
@naval