Originally published in Svenska Dagbladet by Björn Jeffery, August 21, 2025
Meta is posting record profits and investing more than ever in AI. But look closely at the accounting, and there’s a potential cost bomb hiding in the numbers.
Some phrases become immediately memorable. In 2020, investor Puru Saxena posted the following brief comment on Twitter: “Do you like accounting numbers or do you like making money?”
Saxena was pointing to something most investors already know: there can be a vast difference between the figures a company reports and how things actually look in reality.
When Meta reported its quarterly figures this summer, they showed enormous profit — around 180 billion Swedish kronor for the quarter. Despite that, money was pouring out of the company, and Meta doubled its total debt. The reason: AI — and a gamble in the accounting.
The central question for Meta — and for every other company investing in AI hardware — is how long that hardware remains useful. This determines how long it gets depreciated over. Meta calls this category “network equipment,” and in practice it means chips and servers. In principle it’s no different from when a company buys a property: you make a judgment about how long it will remain usable, and try to follow the standards in the industry.
The trouble with AI chips and large data centres is that there’s no real standard yet. The question is less about whether the hardware can still be used, and more about whether Meta will actually want to use it. The moment chipmaker Nvidia releases new products with higher performance, pressure builds on buyers not to fall behind.
What happens if competitors buy the new chips and Meta gets left behind? In what has become a race on the AI market, caution is in short supply. None of the major AI players can afford to risk falling behind.
At present, Meta’s chips and servers are treated — for accounting purposes — as having a useful life of five to six years. In most contexts that sounds reasonable. But the accounting doesn’t quite match the ambitions of Meta’s CEO Mark Zuckerberg. He has made it the company’s new goal to create so-called “superintelligence,” and has lured in talent with billion-kronor salary packages. That the company would hold back on investment at such a moment seems unlikely.
Look at Meta’s cash flow and you already see evidence of this. In the first half of the year alone they spent around 300 billion kronor. In early August they also took on debt to build a new data centre in the American state of Louisiana. Bringing in partners to finance large investments is not in itself unusual. But the scale for Meta stands out — and it signals a trend break. In total they are borrowing around 290 billion kronor, which is double what they had borrowed previously.
At the same time as these massive investments continue, a creeping scepticism is emerging both on the world’s stock markets and among potential customers. A new report from MIT shows that 95 percent of those who have run AI pilot projects have been unable to demonstrate any measurable benefit from them. Even the leading figure in AI development, OpenAI CEO Sam Altman, expressed concern about expectations for what the technology can deliver in the near term. Last week he said: “Are investors too excited? My view is yes.”
Despite the new scepticism, much suggests that the pace of AI investment for Meta and its competitors will continue to accelerate for some time yet. And with that comes the risk that a hidden bomb has been placed in the accounting, waiting. How much of the hardware will be considered good and usable just one year from now? Nobody knows.
But if it turns out that chip development demands new and faster investment, billions could go up in smoke along the way. And with that — a serious write-down for Meta and the other companies that have bet heavily. It is not easy to sit in the finance department in times like these.
