“I think it’s the most valid criticism of AI right now, there is a lot of expense”

The artificial intelligence race is causing large technology companies to squander enormous sums of money to have the best AI, the one that the most people use and, above all, the one that generates the most money (three concepts that are far from the same thing and, rather, let them tell Google). According to Goldman Sachs databig tech and their infrastructure providers have on their roadmap to spend more than a trillion dollars on chips, data centers and software. The million dollar question is: is there a return after that investment?

The CEO of OpenAI, one of the companies in the fight and certainly one of the most interested (it does not have the muscle of veterans like Google, Meta or Microsoft), has already recognized it in an interview for CNBC: It is absolutely normal to worry about that spending on AI due to the waste and the uncertainty of when they will get their reward (if it ever arrives).

Sam Altman’s statements. Asked about the doubts generated by AI, he responded bluntly that it is “the fairest criticism that can be made of AI at this moment.” And he added: “I know big things are happening, but I know there is a lot of waste.” He also put on the table the two questions that companies that adopt AI in their processes ask: how long do they have to wait for this change to be noticed in their income and how long for costs to be under control. Spoiler: taking into account the latest movements from Uber and Microsofttwo completely successful companies, things are looking bad.

The most interesting thing about this display of honesty is where they come from. Altman is the person who has raised the most money for fund OpenAIone of the leading companies in the AI ​​sector but also one of the newcomers, a baby compared to mythical companies that have been dominating the technology for decades. That Altman talks about waste is a before and after in the industry’s discourse.

Why is it important. Until now, the ROI of AI has been a concept on the lips of skeptical analysts, economists and people who disbelieve this boom who point directly to a bubble about to burst. But Altman has integrated it into his corporate discourse and that represents a paradigm shift: it is no longer a critical position from outside the sector, it is that the most influential company verbalizes it to users and investors.

As we mentioned in the intro, Goldman Sachs already considered this same question back in 2024 with its “Gen AI: Too Much Spend, Too Little Benefit?”. Economist and 2024 Nobel Prize winner Daron Acemoglu of MIT published a study called “The Simple Macroeconomics of AI” where he estimated that the real impact of AI on economic productivity in the next decade would be a paltry (especially if we take into account the speeches and investments) of just 0.5%.

Context. That in this phase of expansion and training of AI It is not profitable it is no secretbut this is both an economic and a technical problem. With data in hand, there are reasons to worry. This recent Cast AI report It includes the analysis of 23,000 computing clusters, revealing that the average utilization of GPUs is only 5%. That is, 95% of the most expensive and advanced hardware on the market (those highly sought-after NVIDIA graphics cards) is operating well below its capacity.

Part of the explanation lies in FOMO: explains Venture Beat that there are many companies acquiring AI chips not because they need them right now, but out of fear of running out of them in the future. The phenomenon is not new, we already saw it during the pandemic with semiconductors (and on a domestic scale, with toilet paper).

There is someone who wins. In this story of companies investing to win the AI ​​race and other companies adopting it to modernize, there is someone who is winning from the first minute: NVIDIA bills the same whether its chips work at 5% or 100%. And he is breaking all his records. In 2024 record revenues of $60.9 billion (126% more than the previous year) thanks to this excessive demand for data centers.

The large cloud providers, the holy trinity composed of Amazon, Microsoft and Google (the three occupy 70% of the market, according to Synergy data) bill the same regardless of whether the client is achieving results. According to Synergy Research Groupthe global cloud infrastructure market will exceed $330 billion in 2024. The underlying problem is incentives: those who have the most weight in the pace of investment in AI are precisely those who lose the least, hence no one is taking measures against waste.

Yes, but. Making a catastrophic reading of Altman’s statements would be a mistake and in fact, the CEO of OpenAI himself expressed his confidence “the industry will solve it quickly.” After all, in that initial phase it is normal to incur losses and if not, tell Netflix with streaming. Current waste may simply be the cost of setting up infrastructure whose value will be realized later. Even the Goldman Sachs report acknowledges that bubbles take time to burst, meaning there is room for AI to deliver on its promises. Of course, a good part of the current artificial intelligence expenditure is linked to GPUs with specific architectures that could become obsolete in the face of more efficient models or specific architectures.

In Xataka | The problem is not spending a lot of tokens, it’s that most of them are being wasted

In Xataka | Anthropic has moved ahead of OpenAI in its race to go public. This is very bad news for Sam Altman

Cover | TechCrunch (CC BY 2.0) and Giorgio Trovato