AI will revolutionise the world and, in doing so, displace vast swathes of the economy and create some of the most profitable and valuable companies ever. Or at least, that's the line we have been sold. Over the past few years, this narrative has only grown stronger, and investors, particularly institutional investors, venture capitalists (VC) and tech giants, have poured hundreds of billions of dollars into the AI space to secure their slice of this revolution. However, upon reflection, many of them are expressing deep buyer's remorse. Why? Well, when you look at the numbers, AI can simply never become properly profitable.

The first hint of anything being financially wrong came last year from the Analytics India Magazine. They noticed that at the time, OpenAI was spending $700,000 per day, or $255 million per year, just on maintaining the underlying infrastructure of ChatGPT. At the time, OpenAI was only predicted to have an annual revenue of $200 million. In other words, despite the vast success of ChatGPT and the switch to a for-profit model, it seems OpenAI would continue to run at a loss of hundreds of millions of dollars per year, as they couldn't even cover their basic costs. At the time, OpenAI was predicted to have a 2024 revenue of $1 billion, but as operational costs were set to grow dramatically, Analytics India Magazine predicted OpenAI would go bankrupt by the end of 2024.

As it turns out, these predictions were a little off. Thanks to major deals with the likes of Apple and Microsoft, OpenAI recently reached an annualised revenue of $3.4 billion. However, it has also dramatically expanded its infrastructure since last year, meaning its operational costs have also skyrocketed. Indeed, despite this huge up tick in revenue, there is no sign of OpenAI reaching profitability, and in fact, all signs point towards the company burning through funds.

But, surely, this will get better. Right? These AIs will improve, costs will be reduced, and more customers will come. Indeed, this is the sentiment that many AI investors went by, with many seemingly repeating the adage, "Build it, and they will come."

However, even the leaders of these investment institutions who are heavily invested in AI are questioning the business fundamentals of the AI industry. 60% of Sequoia Capital's new investments are in AI firms. Yet, Sequoia Capital Partner David Cahn has found some profoundly worrying facts. You see, by the end of this year, AI companies, such as Google, Meta, Microsoft, AWS and OpenAI, are set to spend $150 billion on GPUs, the specialist type of computer processor needed to make them work. Cahn found that running these GPUs and other data centre costs will surmount to $150 billion a year. Cahn also found that profit margins need to hover around 50%, as this generates enough funds to pay for AI development, new infrastructure, hiring better talent, and so on. As such, to see a return on their massive GPU expenditure, the major AI companies must generate an additional $600 billion of revenue!

To put that into perspective, Google, Meta, AWS OpenAI, Microsoft and Tesla, who are responsible for the vast majority of this GPU expenditure, have a combined annual revenue of just over $840 billion. In other words, these companies need to grow their revenue by over 70% to see a reasonable return on their investment!

With OpenAI's monumental increase in revenue, you might think that they could reach this figure, but don't be so fast. While AI products can be useful, they are far from the impeccable productivity machines their creators claim them to be. Indeed, MIT's Daron Acemoglu found AI will only improve US productivity by 0.5% over the next decade and will only increase US GDP by 0.9%. This predicted GDP increase amounts to only $2 billion, making that $600 billion revenue increase impossible.

Investors have also noticed this revenue issue. Wall Street expected these companies to spend, on average, $60 billion year-over-year on chips and data centres. Yet, they expect them to only generate an additional $20 billion in revenue.

These investors have even dismantled the idea that these infrastructure and running costs will decrease. Goldman Sachs' Jim Covello has raised serious questions over whether AI can be profitable or influential. However, one of his most insightful findings is that because Nvidia has a near monopoly on AI chip manufacturing, costs won't go down. There is no market force pushing for cheaper hardware. Moreover, from an engineering standpoint, GPUs can't get cheaper to manufacture or operate, as they are already incredibly optimised.

In fact, costs are only set to rise. AI is predicted to have a huge impact because it is expected to improve exponentially with time. As such, it will find more use cases and more customers and reach a tipping point of wild profitability. However, this ignores the reality of building AIs. As I covered in my previous article, "AI Is Hitting A Ceiling It Cannot Pass", AI is experiencing diminishing returns. For current AIs to get even marginally better, they require exponentially more training data and exponentially more energy and infrastructure to train them. As such, the cost of operating cutting-edge AIs is set to only grow exponentially, not reduce, all while the AIs' rate of development slows to a crawl.

As such, the gapping gap between AI expenditure and AI revenue is only set to grow.

This is what could kill the AI industry. AI companies promising to revolutionise the future, while failing to balance the books. It is all hype, no substance. Which is a shame, as AI technology itself is not broken. It can be incredibly powerful when used appropriately. But, don't go thinking this issue is an accident. AI leaders like Altman and Musk know precisely what they are doing. They are gaining the stock markets, making their companies worth way more than they should be, and attracting vast swathes of investment, all so they can cash out before the House of Cards collapses. These investment banks and VCs know this, too, and are betting they can sell off at the right moment. The examples I have used in this article of them sounding the alarm bell aren't really to warn the world not to invest in AI, but more to ensure they cash out at the peak. But, when this bubble eventually pops, it won't hurt them. Instead, it will tank an industry, ruining the careers of thousands, and it will be the everyday investors who lose their hard-earned cash.

Thanks for reading. Content like this doesn't happen without your support. If you want to support content like this, or read articles early, go and follow me and my project Planet Earth & Beyond.

(Originally published on PlanetEarthAndBeyond.co)

Sources: Pymnts, Morning Star, Technext, Forbes, Yahoo Finance, Will Lockett, Cruchbase