Copyright PCPRO October 2025
Everyone on the internet appears to be outraged about something. You can’t move for social media postings lambasting some new cause or public issue. And yet many of these postings have little or no reference sources, and if they do, it’s to a site whose information integrity is often, shall we say, somewhat lacking.
So I was intrigued by a particular thread on the hellscape that is X showing a photo of a poster from the University of East London (tinyurl.com/373wifitrees). It said: “Wi-Fi doesn’t grow on trees. Your screen time is damaging the climate.” Duly outraged, I went to read more at the University of East London’s website (tinyurl. com/373eastlondon).
This makes the claim that the “growing demand for digital services has created a new challenge for the environment. Sending emails, texts, browsing the internet, uploading videos and more all come with a cost – a few grams of carbon dioxide are emitted due to the energy needed to run your devices and power the wireless networks you access.
“With over four billion people across the world using the internet, it is estimated that three per cent of world electricity is consumed by data centres – accounting for more greenhouse gas emissions than the entire aviation industry.”
All of these are reasonable statements. However, a typical Wi-Fi access point is running at around 5W, and that is a really quite insignificant amount of power consumption in the broader context of a house’s electrical requirements.
Another claim I have read is that you should delete unnecessary photos on your mobile device, because these are being backed up to one of many cloud storage facilities.
And all of those pictures require storage, and that consumes power. Again, this is entirely true, but it must be put into context. Just what proportion of the energy consumption is related to those baby photos you took when your offspring first started to walk?
A bigger question arises about how much processing power is required to support the huge data centres that are being built by the big AI players. Some are claiming that all sorts of exotic power generating systems can be used, from vast solar panel arrays to small-scale nuclear reactors.
It’s far from clear that this enormous rush to build ever-larger AI farms will be met with the appropriate power facilities to feed them, or whether the local networks will be brought down. Or whether the prices of wholesale electricity will go up to match the demand. It could be a requirement that such facilities should be entirely self-powered, but only a naive fool would assume that such an outcome is likely.
But will it all be necessary anyway? Is this a vast bubble that is about to burst? I read this post on X last week (tinyurl.com/373cursor): “Why AI is a house of cards: You pay $200 a year for an AI app (like Cursor). Cursor pays OpenAI $500 for API tokens ($300 of which is VC funding). OpenAI pays AWS $1,000 for compute ($500 of which is VC funding). $AWS pays $10k for $nvda GPUs. See the problem?”
Now I am not claiming that the figures are accurate, but there is an underlying truth here. None of this infrastructure can make money at the moment. It isn’t an issue of “we will break even when we can scale out enough”. The costs just continue to go through the roof. And adding in newer and faster generations of technology from Nvidia might seem like a good solution, but not if the workload is increasing at a rate that can consume every clock cycle that you throw at it.
Is this a race to the bottom, where only the companies backed by the deepest pockets can hang on long enough to pay the power bills, build the data centres and fill them with extraordinarily expensive hardware?
It feels like a game of Russian roulette played out by the biggest companies on the planet. When the music stops and the money runs out, the fall-out will be quite devastating. Or have we got to a point where these companies are simply too big to fail?
It feels like a game of Russian roulette played out by the biggest companies on the planet. When the money runs out, the fall-out will be quite devastating
At some point, these enormous AI factories have to come up with a business model that actually works. In 2024, OpenAI lost $5 billion on revenue of just under $4 billion. But this is small fry: Google, as Alphabet, is spending $85 billion in 2025, up from initial estimate of $75 billion, on AI and other related technical infrastructure work. I have read that Microsoft is spending $80 billion in 2025, too.
At some point, this all adds up to an awful lot of money. Will the music stop soon?
No comments:
Post a Comment