POD! On YIMBY Pod this week, we talk about the Green Party’s unenthusiastic U-turn on nuclear energy, and discover why London could soon be getting a massive new heat-network spanning all the way from distant East London, right into the centre of the city. Listen here, or wherever you get your pods!

I finally decided to see what all of the fuss was about.

After years of reading absolutely incomprehensible posts from friends about The Traitors, I gave in and tuned into the first few episodes of the new celebrity series. After all, nine million people can’t be wrong, right?

However, I’m sad to say my reaction to the show was pretty lukewarm. Don’t get me wrong, many of the personalities on the show are fun to watch, including the excellent Hannah Fry, who once said nice things about my newsletter (and, implicitly, PAF liberation). But my frustration is with the format of the show itself, which beneath the dramatic music and gothic imagery, seems a bit half-baked.

Why? Essentially it’s not really clear why the ‘Traitors’ are traitors, and there simply isn’t enough actual information for the people guessing to go on, for the speculation at the core of the game to be meaningful.

Now, I admit that maybe this is part of the fun and I’m just a killjoy. I’ve heard the argument that the real appeal of the show is that it is a grand experiment in human psychology. We get to see how people spot patterns in noise, and how tribes and groups emerge even when there is no underlying signal.

But I just can’t get past the comparison to a similar, but much better gameshow, called The Mole, which aired way back in 2001 on Channel 5.

At the time, the show was very much Channel 5’s attempt to build something as splashy as Big Brother, and it largely failed in this endeavour. But for the few who watched, it was instantly compelling.1

The premise was simple: There were ten players tasked with completing different missions, similar to those on the Traitors, to earn money for a total prize pot. But one player had been secretly designated as the titular Mole, and their job was to try and actively sabotage the games. At the end of each episode, the players took a quiz, and the person furthest from guessing the identity of The Mole was booted off.

What made this format work better than Traitors was that it gave the missions a purpose in the context of the wider show.2 In Traitors, there’s no incentive for sabotage or any other, well, Traitor-like behaviour – and so for half of each episode’s runtime, it’s like the players are competing in a completely different show.3

But anyway, this is all a sidebar and an excuse to reminisce about a much better show. What I really want to talk about this week is one of the Traitors contestants. Not the strangely eccentric Richard E Grant, as I’ve got no idea what was going on there4 – but the first person who was booted out of the castle, Amol Rajan.

In case you missed it, tomorrow Rajan is due to launch his new podcast-slash-media-venture, The Amol Rajan Connection (ARC). He’s clearly aiming high with the show, and there’s clearly quite a lot of money behind it.

Yet with my media-business-analysis hat on,5 as likeable as Rajan may appear on screen, I must admit I’m incredibly doubtful that the new venture will prove successful.

So I thought this week I’d follow in the grand tradition of my analysis of other media properties like the New Statesman, Observer, Newsnight and Owen Jones, and make the case for why I think he’s unfortunately doomed to failure.

Now let’s get started.

Read more

On YIMBY Pod this week:

• We talk about the Green Party’s surprising kinda-sorta-endorsement of nuclear energy, and whether it’s a big deal or not.

• And then we speak to Aaron Gould, CEO of the Association for Decentralised Energy all about the prospects for building more heat networks in Britain!

Martin’s Substack is here, and his blockbuster new video on the Rainham Volcano is here.

James’s Substack is here, and his recent essay on AI and government is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

On YIMBY Pod this week:

• We talk through two of the highlights in Andy Burnham’s conference speech – his pitch to build historic quantities of council housing, and plans to create “GB Grid’.

• Martin gets a birthday treat in the form of some intriguing news about enemies of the pod Natural England.

• And we speak to economist Omer Bor about his fascinating piece ‘Britain wants to build homes near trains it can’t afford to run’.

James wrote up his longer thoughts on Andy Burnham’s conference speech, and a deep dive into which MPs are the most and least memorable in Westminster.

Martin wrote ‘The AI Apocalypse is Extremely Boring Now’, and his YouTube channel is here – a new video is dropping tomorrow!

Try James’s two stupid games: MP Recall and MP Chain!

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

POD! On YIMBY Pod this week, Martin and I talk about the BBC’s pathological scepticism towards the new Universal theme park and how Andy Burnham has (reportedly) thrown a spanner into the works at Heathrow. Then we speak to Anya Martin about why we should impose a tax on autonomous vehicles – before it’s too late! Listen here, or wherever you get your pods!

Last week I went mildly viral – at least within a half-mile radius of Parliament.

You might have seen why. Following the New York Times theory that nobody actually cares about written words, and what people really want is games, I created two little games of my own.

MP Recall came first, and this was a memory test. Players were given five minutes to type the names of as many Members of Parliament as they could remember, aided only by some floating faces in rotation, to help jog memories.

Then I made MP Chain. This was a more fast-paced game, where players were given the name of an MP, and asked to identify a given member, when presented with two headshots.1

Both games were pretty successful – MP Recall has had over 7,200 completed games with over 245,000 MPs successfully recalled, and MP Chain was played over 60,000 times,2 with over 300,000 choices recorded.3

However, I’m not dropping some big numbers just to show off – that’s merely a happy side-effect. What’s more important is that these games have inadvertently given me a rich dataset.

For a long time I’ve been obsessed with how much people actually know about politics. I still regularly think about the 2020 survey where 11% of people claimed to know “a fair amount” about then Shadow Chancellor Anneliese Dodds, which I think is just laughably untrue. Nobody knows “a fair amount” about Anneliese Dodds – the survey just showed that 11% of people are liars.

Now, my data does not speak to exactly this question of what the general public actually knows, nor am I making any claim that my data is in any sense meaningful or scientific.

But I think it is perhaps a snapshot of which Members have made an impact in the inside-game of politics. Not the air war, in the battle for attention from the real general public – but in terms of impact amongst the brain-poisoned Westminster obsessives who either work in and around Parliament, or at least spend a lot of time posting on social media like they do. The real sickos, in other words.

So let’s dig through the data, and discover who are some of the Westminster Village’s most and least recognisable faces.

Make sure you subscribe to my newsletter (for free!) to get more like this direct to your inbox.

Subscribe now

Recalling the data

Let’s start with MP Recall, and the Members who were most often remembered off the tops of people’s heads.

I think this was a surprisingly tricky challenge. Even for obsessives like the people who read obscure Substack newsletters about politics, when you’re presented with such a large and general category, there’s no reliable way to methodically work your way through all of the names in your head in just five minutes.4

And I think the data on who people actually remember is, to an extent, pretty unsurprising. At the top of the list is the Prime Minister – a full 76% of players successfully named Andy Burnham, and he was followed by all of the other big hitters you’d expect, including Nigel Farage, Kemi Badenoch and Ed Davey. Though the drop-off is quite stark – how quickly people have forgotten Rishi Sunak and Rachel Reeves!

In fact, I think the data is arguably evidence that for most people, the cast of known characters in politics is rather small – the top ten most recalled MPs accounted for 17.3% of all names entered, and the top 50 accounted for 47.8% – suggesting a small core of ultra-visible MPs, and a very long tail of MPs, who are presumably spending less time on TV and podcasts, and more time doing thankless constituency work.

On the flip side, the least recalled list is dominated mostly by Sinn Féin MPs, who refuse to actually take their seats in Westminster – so are effectively just names on ballot papers every few years, and that’s it. They don’t turn up or do anything, and they don’t even have photos on the Parliament website.

However, Sinn Féin aside, there are also four Labour MPs at the bottom of the list, with the lowest-ranked being Paul Foster, MP for South Ribble. I’d never heard of him either – but he spent 15 years in the army, serving in both the Gulf War and Kosovo. So now we all know something about him.

The least recognised Labour MPs also have something else in common: They were all elected in 2024. Which perhaps shouldn’t be surprising – as MPs elected in 2024 had an average recall rate of 2.8% – compared with 7.9% for all MPs on average. So stick around long enough, and people might remember who you are.

Only connect

Now let’s look at MP Chain. Unlike the first game, this wasn’t a test of memory – but of recognition. Who has a face that people vaguely recognise?5

Here’s the leaderboard for most and least recognised MPs, with the percentage of times the player guessed them correctly in head-to-head match-ups with another MP:

I think again unsurprisingly the most recognised MPs are all household names – at least in the homes of the sort of weirdos who played this game.

Perhaps the most surprising thing in the data then – which I probably should have predicted, is that disproportionately some of the most recognised MPs were those members from minority backgrounds. Though I suspect some of this is an artefact of the way the game was designed.

For example, here are the top twenty most recognised Labour MPs who aren’t in the Cabinet – they’re either pretty famous in Westminster terms, or have a name that gives little doubt which face they’re most likely to be when drawn against another MP at random.

…Except, that is, in two cases.

This is going to get a little awkward, because the two least recognised MPs – the only two with less recognition than a coin toss would predict – are Mark Hendrick and Peter Bedford. Hendrick is of Anglo-Somali heritage, and Bedford has British Indian and English heritage – but I guess their names offered fewer clues to players.

Anyway, the other striking thing about the Chain data was that the 2024 factor also made a difference here, with MPs elected recently getting recognised correctly 77.3% of the time on average, compared to 85.8% for older MPs.

The big list

So those are some stats on which MPs are the most memorable and most recognisable in Westminster. But the chances are there will be certain Westminster insiders reading and thinking… what about me? Or what about my boss? And what about that bastard who got the ministerial job I wanted?

The good news is that I’m going to publish the full list below – but the bad news is because I’m a smart business guy, I'm going to put it behind my paywall. So if you want to see the full list, why not consider subscribing to support my work?

Subscribe now

Now here’s the list in full:

Read more

Yesterday afternoon, Andy Burnham gave his first conference speech as Prime Minister, and judging by the breathless commentary on my timeline, it will find itself placed in the historical pantheon of political speeches positioned somewhere between the Gettysburg Address and I Have A Dream.

In all seriousness though, I think there were many things to like in the policy-heavy speech, as well as many other things that left me with further questions about the question I posed back in July: whether Andy Burnham is serious or not.

So this week, here’s a breakdown of what I liked, what I didn’t – and what left me scratching my head.

Read more

As I write, SpaceX’s massive Starship rocket has just achieved another huge milestone. It has successfully made it to orbit for the first time. It’s another major milestone towards a total transformation of the economics of space – so to celebrate, here’s a re-up of my previously paywalled post from October 2024 about the downstream implications of Starship, and why it is important.

Subscribe now

If you value my work, please consider taking out a paid subscription so I can continue writing every week!

A former SpaceX employee once described to me why he thinks Elon Musk is… the way he is.

As he sees it, Musk thinks like an engineer. You try to build something, continue to iterate on it, and once you’ve successfully built it… you have been proven correct – whatever you end up with is proven to be the correct way to do it.

And the problem is that this approach is, as we know, utterly disastrous if you’re trying to build a social network like Twitter, because people are not just widgets that respond mechanically to the laws of physics. People are complicated.

However, there is some value to this approach.

For example, if you’re building electric cars or rockets, thinking like an engineer is extremely effective. In those cases, development is about changing variables and seeing what works best. If prototypes blow up en route, it doesn’t matter, as each experiment nudges you closer to your goal.

We saw this for ourselves on Sunday. On the fifth ever fully-stacked launch of SpaceX’s new Starship rocket, Musk’s company successfully used a pair of ‘chopsticks’ on the launch tower to capture the ‘Super Heavy’ first stage booster of the company’s new Starship rocket as it returned to Earth.

I watched the live stream, and I can safely say that it was one of the most incredible things I have ever seen. Even more so than when the company landed two Falcon Heavy boosters for the first time six years ago.

Why? Because Starship and the Super Heavy Booster are together the height of a 40-storey building, and capturing the first stage is equivalent to precision manoeuvring a 22-storey building, and plucking it out of the air.

However, it’s not just the visual spectacle or the feat of engineering that was important. What really makes it incredible is what this technological moment represents.

In the medium to long term, this is the first rocket we’ve ever built that could conceivably one day, get humans to Mars.1

But in the short-term, this was the moment that SpaceX transformed the economics of space… again.

And I think the consequences of this are going to be a really, really big deal for everyone back on Earth, for reasons that I’ll now explain.

Collapsing prices

SpaceX has already changed the economics of space once before.

Only a few years ago, getting a kilogram of material to orbit cost somewhere between $10,000 and $20,000. But in 2015, SpaceX changed the world by successfully executing a controlled landing of the first stage of the Falcon 9, the company’s smaller rocket.

This is a more recent landing – scenes like this are now routine.

In an instant the ability to land slashed the price of launching payloads to orbit, for the same obvious reason it’s cheaper to re-use an aeroplane than build a brand new one for every flight. Today, the price per kilogram to orbit is around $2700.

What this means in practice is that we2 can get a hell of a lot more into space, for cheaper. Instead of having to build a new rocket every time, SpaceX now has a fleet of boosters that are reusable. Some have now flown around 20 times, and without having to manufacture many more,3 the company is today able to launch multiple times per week – something that was unimaginable before.

And this alone has transformed what we’re able to do in space. For example, it’s become feasible for smaller companies and start-ups to launch their own small satellites.

That’s why we’re currently living through a boom in satellites designed for things such as environmental monitoring (eg, GHGSat, which monitors greenhouse gas emissions), asset tracking (so that companies can monitor shipping containers and the like), and scientific research.

There are also new imaging satellites that can experiment with less proven technology. For example – in 2022 a British company called Sen launched a small satellite on a Falcon 9 that can deliver live, 4K footage from space – something that hadn’t been done before. This makes it conceivable that eventually we could have an entire network of cameras providing virtually real-time views of the entire Earth.4

And it’s how Sen was able to capture a video of Hurricane Milton – which looks like something we’d only usually see in a computer-generated simulation:

Anyway, this doesn’t even scratch the surface of what capabilities have emerged now that we humans have access to this new launch capacity. Not least because I haven’t even mentioned SpaceX’s own Starlink satellite internet yet, which today has over 6000 satellites in orbit providing broadband-style speeds to, well, conceivably everywhere. If SpaceX hadn’t reduced the cost of spaceflight, there’s no way that such a network could ever have been built.5

An order of magnitude

However, though the above is cool, it’s almost old news now.

What the Starship launch and catch demonstrated is that the components of the new rocket work: That something of its size can make it to space, and that its constituent parts can return to Earth to be used again. I’m sure there will be more explosions and mishaps before the rocket is fully ready, but the fundamental theory of Starship has been demonstrated.

In other words, Musk has been proven correct. This is officially how a Mars-capable rocket works.

So now it is only a matter of time before the design is finessed, and flights become routine – much like the Falcon 9 now.6 And once this happens, the price for launching stuff into space will collapse all over again.

In fact, Musk himself has claimed that Starship could make a mere $10 (yes, ten dollars) per kilogram to orbit possible. And whether or not he is bullshitting, as per his previous wild claims, the reality is that given the payload capacity of the new rocket and the cost of launching, prices to orbit will almost certainly fall by several orders of magnitude.

And this is why I wonder if we – humans – are not quite prepared for what this will mean.

The new, new economics will transform what is possible in space – because costs falling and scale increasing can lead to new, emergent capabilities.

For example, smartphones went from expensive, experimental devices that only the rich could afford, to something that can be picked up for £50. As a result, there are now over seven billion smartphones in the world, and all but the very, very poorest have access to the technology – which has changed how the world communicates, shares information, and does business.

So what new capabilities will emerge from the new economics of space?

One obvious one is that it will become viable to launch many more probes to explore the solar system. And it might be possible to send probes to places faster too, because Starship can carry more propellant, so it can burn more fuel to increase speed. I won’t be surprised if future missions to Mars and Venus don’t just send one probe, but send a fleet.

I wonder if we could also see the rapid emergence of new space telescopes too. The James Webb Space Telescope, which launched in 2021 cost $10bn – and though something like that was always going to be expensive, it’s conceivable that the next telescope could be cheaper. This isn’t just because of the cost of launch, but because Starship’s larger payload bay will mean that telescopes won’t require such a wildly complicated mechanism to fold out its mirror – so money can be saved in the design phase.

There are many other possible pure science applications too – of varying levels of plausibility. For example, Casey Handmer has already written a great post about some of them, covering everything from how Starships could be connected together in orbit to form huge new space stations, to how they could deliver the large payloads to the Moon that we’re going to need to establish a permanent presence on the lunar surface.7

But much as with the Falcon 9 cost collapse, this isn’t just about scientific research. What’s probably more important in the short-term is how Starship will make it possible to launch many more satellites that look back at the Earth.

For example, it will make deploying Starlink even cheaper. At the moment, a Falcon 9 can deploy up to 60 Starlink satellites in one go – and with Starship, this will increase to up to 400 with each launch. That could make it even more plausible to have broadband access everywhere.

And as with the small satellite examples above, launching satellites will become more affordable for even smaller companies and organisations.

And hell, there’s even some mad-sounding proposals to use space-based solar panels to generate electricity, which would then be sent via microwave link back to Earth – just like in Sim City 2000. I’ve no idea how plausible this is, but Starship will at least nudge us closer to experimenting with it.

Scale and the Musk of it all

I think what makes Starship so exciting – and so important – is the scale it enables. Just as the container ship revolutionised the transportation of goods around the world, and just as modern passenger jets were a step-change from the tiny propeller planes in the early years of flight, Starship is going to change humanity’s relationship with space.

That’s why I wonder if we’ve collectively not yet woken up to what is about to happen.

And perhaps this is unsurprising, as it would have been hard to predict how the world was about to change a year before the launch of the iPhone, or the public release of ChatGPT. Similarly, I suspect that Elon Musk being such a hateful person makes us not want to acknowledge the importance of this technological moment.

But I think what Sunday’s launch confirmed is that Starship is now not a matter of “if”, but “when”. And that means when we think about the near future we now need to imagine a world where space is a much bigger part of our everyday lives.

Subscribe now

On YIMBY Pod this week:

• We talk about how progress sometimes has winners and losers, by looking at the case of someone who is almost uniquely screwed by the new Universal theme park.

• Then we follow up on three things we’ve talked about on the pod – on-street EV charging, environmental delivery plans and… how the bloody Prime Minister has cast doubt on Heathrow’s third runway again, after it all seemed like it was happening.

• And we speak to Anya Martin about her fascinating new paper, arguing that we need to create an autonomous vehicle tax now – before it’s too late!

James wrote about how access to AI compute is a social justice issue (and don’t forget to check out last week’s post about why we should turn East Anglia into Britain’s Shenzhen.)

Martin wrote ‘The AI Apocalypse is Extremely Boring Now’, and his YouTube channel is here.

And try James’s two stupid games: MP Recall and MP Chain!

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

POD! On YIMBY Pod this week, Martin and I do our best impression of Tim Harford on More or Less, debunking some questionable Guardian reporting on data centres. Then we speak to James Howat from ThinkLabour, who argues that we can build more infrastructure if we bring back municipal bonds. Listen here!

GAME! How many MPs can you remember? Try my stupid little MP memory game to find out!

My partner asked me what I want for Christmas the other day.

This is a surprisingly difficult question. We’re middle-class and childless, and though we’re not particularly wealthy in any meaningful way, this does mean that if there’s something reasonably priced that we want, like a book or a video game… we just buy it. We don’t wait for a special occasion, because why bother?

So I sat there thinking for more than a few moments, my mind blank, when all of a sudden something depressing hit me. Could she buy me some more AI tokens?1 Does ChatGPT sell gift cards?

As bizarre as it may seem to most people, a lack of AI compute is, with only a little exaggeration, one of the biggest problems in my life. ChatGPT has proven so useful to me as a knowledge engine, software development platform and work assistant, that I’ve recently found my supply of AI tokens running dry, to the point where, even though I pay for the $20/month ‘Plus’ tier, it eventually refuses to respond to my queries for several hours – or even days.2

From a business perspective, this makes total sense for OpenAI. Though an app on a computer screen may feel weightless to the user, the computation to make the magic happen is extremely resource-intensive. If you prompt ChatGPT, in order to compute the answer, something roughly equivalent to an entire rack of computers in a data centre briefly dedicates itself to your query. That’s why AI energy usage is a significant environmental concern, as well as a bottleneck for development – it’s a non-trivial expense that the AI companies need to pay for.

But from my perspective, knowing this does not make it less annoying. Though there is a way around this – I could simply pay more. ChatGPT would love me to upgrade to the $100/month tier for a 5x increase in the number of AI tokens I can use.

However, here I hit a different roadblock. I can’t really afford to spend $100/month on AI compute. I mean, I suppose I could – but at that level it becomes a middle-class spending trade-off, where I’m forced to choose between my precious tokens and, say, subscriptions to Netflix, Amazon Prime, and a myriad of other subscriptions. And I should probably continue to pay my half of the mortgage, for the sake of my relationship and having a house to live in.

Anyway, it was when I was experiencing this great pain that I realised something important. Though my dilemma was simply annoying, I think in the not too distant future it could become an important challenge for society.

Because what happens when this problem hits people who are not as fortunate as I am? If the most optimistic predictions about AI are true, and it really is world-changing in its utility, and if compute remains a constrained and expensive resource – what will limitations on AI usage mean for the poorest people in society? Won’t it be unfair if the rich can buy all of the compute they need, and the poor are left without?

In other words… Am I crazy to think that AI could soon become a social justice issue?

Subscribe now

Power to the processors

I wrote recently about how AI has broken the link between human brains and cognition.

Though we can debate at exhaustive length about exactly which tasks AI can perform, or the limits of its ‘intelligence’, it’s increasingly true that human-like cognition can be spun up on demand with a credit card.

This is a weird analogy, but it’s a bit like the transition to cloud computing in the mid-2000s. In the before times, companies and individuals needed to make huge capital outlays on computing equipment – essentially placing bets on their future needs.

But then cloud arrived and smoothed the correlation between demand and costs. When more people want to use your app, the cloud platforms automatically spin up more computers to meet the demand, and when fewer people use it, it scales down again.

And we’re arguably already witnessing the same happen with ‘intelligence’. The amount of smarts you will have to dedicate to any given task will be the sum total of both your brain and however much AI compute you have access to.

In other words, there will be a similarly smooth correlation between intelligence and how much you can afford to pay.

So wind the clock forwards and the implications of this are scary. The rise of AI could end up having significant distributional impacts, and could become a source of significant inequality.

Now, I know what you’re thinking. “Finally O’Malley gets it! That naive AI booster is conceding that the hectoring critics have been right all along!”

So let’s be clear. I’m not talking here about the potential AI impact on jobs, intellectual property or any of the other well-worn criticisms of the technology. Those are different debates for other pieces.

What I’m talking about specifically is access to AI intelligence – especially if the most optimistic Silicon Valley predictions about AI turn out to be true.

This is because if AI is as transformative as the Tech Bros hope, the technology will inevitably be woven into all aspects of our lives. If intelligence is available on tap, it’s plausible that we’ll all essentially have AI agents running in the background of our lives and this will be completely normal – a bit like how today it is basically impossible to function in society without a smartphone.

What will these agents be doing? It would be silly to speculate on what specific products might emerge. Though, for the sake of giving you something to picture, imagine a virtual personal assistant that is like Siri on steroids. Something that can proactively work on your behalf, without needing prompting. It could manage your money, organise your calendar, negotiate with the government over your benefits, or perform ten billion other tasks basically autonomously to reduce the friction in your life.3

Whatever conveniences end up being invented will require compute resources – and those resources will be subject to the same dynamic: Those with more money will have access to more compute, and therefore will be able to do more stuff, further entrenching a society of haves and have-nots. And unless we plan for it, it could turn out that not having access to compute is just as debilitating as living without a phone is today.

Uncertain assumptions

Obviously, my argument here is highly speculative. It rests on several assumptions about the future coming true. Perhaps the AI sceptics will be proven right.

But even if the technology does prove maximally useful, there are other examples where analogous to this where any problems have simply disappeared.

For example, when my parents first got a dial-up internet connection in the 90s, access to the internet cost a penny per minute, which meant that time online was severely rationed. I couldn’t spend an evening wasting my life, scrolling listlessly because there was a permanent ticking clock running, and it could add up to serious expense.4

But eventually, this problem mostly disappeared as the market shook out so that instead today everyone pays a flat fee, in exchange for internet usage that is effectively unmetered, no matter how long we spend on Geocities, or whatever it was we did in the 90s.

The same happened to mobile internet usage too. At the beginning of the iPhone era, phone contracts severely limited everyone’s data allowance, but eventually mobile infrastructure caught up and the cost of data collapsed, which is why today most people can afford a package that is either unmetered, or large enough for data limits not to really matter.5

So there is a strong argument that AI will likely go the same way. The reason you can easily hit a hard limit on ChatGPT and Claude today is that the availability of compute is severely limited. As soon as the world builds more data centres (as is happening), the theory is the price of compute will fall and AI use will be effectively unmetered.

And then there’s the related argument that in the future, most AI compute will happen locally – your AI data crunching won’t happen in a distant data centre, but on the chip inside your phone or computer. That would mean no need to pay a subscription, and you can use AI as much as you like for free.6

I readily concede that, maybe, this is a possible future too. However, I must admit, I’m a little sceptical in the medium term, because of the huge gulf in capability between local models, designed to run on home devices, and the latest frontier models that run in the cloud. Maybe I just lack the imagination, but given the hardware and energy requirements, I’m not sure how all of that can be shrunk down into a smaller model that can work just as effectively on significantly less capable hardware.7

But anyway, my point is that though these are two possible futures – there is a third. And this is where the danger lies. What if the cost of compute doesn’t fall, or if the price equilibrium lands somewhere inconvenient and out of reach of many people?

We got lucky with broadband and mobile – it’s so cheap that most people can afford a home internet connection and a SIM card with a reasonable data allowance.

But what if AI is so unbelievably useful to the rich and to corporate users that the cost of compute for people on low incomes, or small businesses just starting out, is just too out of reach? So that when rich and poor compete for access to this new, scarce resource, there’s only one winner?

I don’t think this is as hard to imagine as it might seem, as we’ve already seen a microcosm of what this might look like, and the victims are some of the least sympathetic people on Earth.

I’m talking, of course, about gamers.

A significant consequence of the AI boom is that because data centres require similar chips, RAM and other components to the computers in our homes, manufacturers have dedicated their capacity to providing for data centres – sending the cost of home computing spiralling.

For example, earlier this year Apple raised the price of the entry-level MacBook Air from £1,099 to £1,299, and amazingly the cost of a new Playstation 5 today is £569.99 – 27% higher than when the console was first released six years ago.

This is completely unprecedented. Typically what you’d expect to happen is for consoles, phones and computers to get cheaper over time as devices age and components become cheaper to make. But cold business logic dictates that spending on AI is just much more lucrative, which has left everyone else paying over the odds for hardware.

So my point is – what if we end up in a similar position in terms of access to compute? What if the marginal value of a little bit of extra compute is much higher to a large corporation to the point where it makes compute unaffordable for regular people? And what if this isn’t a problem that goes away quickly like we might expect as supply catches up to demand?

Similarly, I don’t think this is as unlikely as it might sound.

Though the current AI build-out is often described as a bubble, the record investment could simply be a reflection that this really is a world-historic, valuable technology. And perhaps we are starting to see the first signs of this. Anthropic is now forecasting that it expects to bring in upwards of $100bn in the next year. That’s not hype – that’s cash from customers who want to buy the AI capabilities that the company is selling.8

And I think demand for compute could still conceivably go even further too. As I wrote previously, this is a new general-purpose technology – there are countless tasks to which this new resource of cognition-on-demand can be applied. And as Ben Thompson has argued, what will drive this demand for compute – or ‘inference’9 – isn’t just humans typing queries, but AI agents working autonomously – compute begetting more compute without human intervention:

Today’s agents are fancy answer inference; in the future true agentic inference will be work done by computers according to dictates given by other computers, and the market size scales not with humans but with compute.

Given that agentic computing is already exploding, I think it is at least plausible that we will reach a point where it is far more valuable for corporations to pay a premium for marginal compute, creating the market failure I describe above. And it’s not a given that just because demand will be there from individuals, supply will catch up – because building isn’t without friction, as data centres are hugely controversial and subject not just to market forces, but the arbitrary whims of planners and other political interests.

And if this sounds familiar, it’s because though I’m describing a hypothetical problem for AI, there’s another sector where the corporations and individuals are already locked into a competition for the same scarce resource that is subject to the same dynamic: energy.

As things stand, if electricity was a pure free market, we’d build vastly more energy infrastructure to reduce price and service demand, but energy infrastructure is subject to other constraints. And likewise, if domestic energy prices were set by the market, they would be significantly higher, and the prices would disproportionately punish the poor.

That’s why, in the case of energy, the government created the energy price cap. Though it doesn’t completely delink household energy costs from the market, it does attempt to mitigate it to reduce the burden.

So is it that crazy to imagine something similar being required to ensure AI compute is shared fairly too?

Left behind?

So that’s my wild speculation on how access to compute could conceivably, in the not-too-distant future, become a social justice issue. Like I say, perhaps it won’t happen – perhaps one of the other futures will emerge, and availability of compute will be so commoditised this simply doesn’t become an issue.10

But if this is what the future looks like, and AI does live up to the hype, if you care about social justice and poor people, then this is something everyone – particularly those on the left of politics – is going to have to think about.

And so far, I don’t see much thinking along these lines. Much of the left-of-centre response to AI so far has been either outright denial of its capabilities11 or has been stuck in the intellectual dead end of “do we even want this?”, as though the LLM genie can somehow be stuffed back into the bottle. Then more credibly, there are more serious leftist critiques that are worth taking seriously, such as the impact AI will have on jobs, or how it will shift power from labour to capital.12

But what has been missing from all of this is any discussion of how to manage AI if the most optimistic expectations are right. We’ve invented this incredible new technology, capable of so many potentially positive things – but how can we ensure that the fruits of artificial intelligence are fairly shared?

I suspect the reason for this blind spot is mostly because, for people on the left, it is faintly embarrassing to talk about such a sci-fi idea, and not to mention politically awkward to contemplate given other ideological commitments. To entertain the idea that compute is a social justice issue is to implicitly concede that the hated Tech Bros will have created something good for the world.13

But I do think issues like this have a habit of seeming like a joke until suddenly they are very much not.

It wasn’t so long ago that you would have sounded mad for worrying about the moderation policies of a social platform for university students – but now those same rules help decide elections. Likewise, the micro-blogging platform where you used to post jokes and memes has helped smash the cordon sanitaire between the mainstream and fascist right. And that funny video app where people post dancing videos and comedy skits? It’s handed the Chinese Communist Party the tools to shape the minds of younger generations across the West.

So if AI does turn out to be as transformative as many people expect it to be, we should start thinking now about how to ensure that access to compute does not further entrench inequality in society, and so that everyone can share the benefits of this new technology.

I’d ask ChatGPT how to get started with this, but I’m all out of tokens.

Subscribe now

Your usual newsletter essay will be arriving later this week, but in the meantime I couldn’t resist sending you my latest creation – a memory challenge to see how many MPs you can remember in five minutes.

It’s surprisingly tricky – I managed 44, and I’m completely brain-poisoned by Westminster politics. The best score I’ve seen so far is 61.

If you want to try it, click here. And don’t forget to share your results on social media for bragging rights/pity.

Anyway, enjoy!

James

On YIMBY Pod this week:

• We try to channel the spirit of Tim Harford and debunk some rather strange numbers in the Guardian’s data centre reporting.

• Then we speak to James Howat from ThinkLabour about how bringing back municipal bond markets to get Britain building again.

James wrote up his Guardian debunking here, and covers two other bad data centres stories too.

Martin’s Substack is here, and his YouTube channel is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

In 1979, the city of Shenzhen in southern China was little more than a fishing village. Today, however, it is a city of 17 million people, and the centre of the global electronics industry.

Sure, most software we use is written in the United States, and Silicon Valley is where the most important hardware is designed. But most of the actual production lines that turn blueprints, raw materials and components into finished devices are found in Shenzhen. In fact, there’s a good chance the phone or laptop you’re reading this on right now was made there.

If you want to read about this in detail, I strongly recommend the fantastic Apple in China by Patrick McGee. It is, of course, a book about Apple, but it also tells a broader story of how this one Chinese city came to dominate tech manufacturing.

It’s a story into which Apple is tightly woven – the smartphone boom that fuelled the city’s growth is all downstream of Tim Cook’s mastery of supply chains and outsourcing. But it’s also a story of how, by satiating the West’s desire for cheap, reliable manufacturing, China has steadily built up its own domestic giants. This is what enabled China to ascend from the bottom of the ‘value chain’, manufacturing basic widgets, to the top, as a home for extremely high-end manufacturers.

It’s why today China is the best country in the world at everything from drones to humanoid robots to electric cars, and the stereotype of the cheap Chinese knock-off is almost as out-of-date as when people used the same slur against Taiwan and Japan.

But how exactly did this transformation happen?

In one of the most striking passages in the book, McGee talks about how, ironically, despite the globalised nature of modern tech supply chains, agglomeration really is critical to Shenzhen’s success:

Indeed, China sports a level of technical sophistication that multiple experts struggle to even comprehend. The layperson often believes Foxconn could just open a factory in a different country; but the Foxconn hubs in China are surrounded by hundreds of sub-suppliers all ready to compete for the next major order. “For Apple to give that system up is tricky,” says Jay Goldberg, founder of D2D Advisory, a tech consultancy. If Foxconn, for example, needs to install sonic welders-a process to merge different metals or plastics with ultrasonic energy-it can call up any number of firms to run the line and hire the labor.

“There’s all these subcontracted, specialty niche firms, and nowhere else does that exist,” Goldberg says. What China offers, in other words, is not simply labor, but an entire ecosystem of processes developed over more than two decades.

This brings me to Britain – and Europe more broadly.

As much as we are beneficiaries of decades of cheap manufacturing in China, there has been an implicit trade-off. We have lost almost all of our ability to manufacture and build at home – a fact that has been made stark by the war in Ukraine, which was fought at the beginning with almost exclusively Chinese-made consumer drones. Even today, though Ukraine has rapidly developed a sophisticated domestic drone industry, many of the underlying components, like motors, control chips and magnets, are still sourced from China.

This, then, represents a huge strategic vulnerability. Which is why one recurring theme in our politics is the need to reindustrialise Britain, to protect ourselves from the more volatile global situation. And slowly the government is moving in this direction.

Just this week, Andy Burnham visited the motorsport company McLaren to trumpet the firm’s plans to invest in domestic manufacturing – as domestic manufacturing is not just good for high-minded security concerns, it’s good for creating vocational jobs too.

But even though he visited an F1 team, I’m worried that Burnham is not moving fast enough.

Don’t get me wrong – the noises being made are all of the right ones. Politicians from Burnham down clearly recognise the problem that if Russia invades a NATO ally, or if China moves on Taiwan,1 it’s not going to matter all that much that Britain is a global hub for TV production or corporate legal services.

However, I also think that if we take the geopolitical threats seriously, then we need to be more ambitious. Which is why today, I want to sketch out, on the back of a proverbial envelope, something bigger – a real attempt to get Britain manufacturing again.

Because though there is a lot of talk about the opportunity for a ‘knowledge corridor’ in the Oxford-Cambridge arc, which will provide new homes and graduate jobs in the London commuter belt, Europe also needs a powerful manufacturing ecosystem of its own.

And if we know this is something our continent needs, why can’t we build it ourselves? Which is why this week, I’m going to explain why – and how – we should turn East Anglia into Europe’s Shenzhen.

Read more

Hello! It’s time for Odds and Ends, your semi-regular round-up of the most interesting links I’ve seen lately, plus some shorter takes.

This time I’m featuring:

• Some sobering facts about the plan to build near train stations

• The ultimate Treasury Brain case study

• Mark Carney showing Andy Burnham how to communicate

• Burnham acting like a human on camera

• This one weird trick could help Britain build tonnes more infrastructure without spending more money

• The sporting event I’ve found most compelling recently

• Some intriguing data journalism about the final tracks on albums

• And a pre-9/11 blast from the past that feels extremely weird now

Read more

On YIMBY Pod this week:

• The government has discovered a new weapon to fight the British state’s sluggishness – an angry letter.

• And we speak to policy consultant Adam Tyndall about the latest on Heathrow’s third runway, and just how close we are to it actually happening.

James recently wrote about data centres again, picking a fight with the Guardian in the process.

Martin’s Substack is here, and his YouTube channel is here.

The letter we discussed on the show is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

POD! On YIMBY Pod this week, Martin finally takes the Claude pill and I try to persuade him of the need to build data centres – and we speak to the excellent Jonn Elledge about his new book, on the 31 Inventions That Build Our World! Listen here, or wherever you get your pods.

I’ve fallen into a self-destructive feedback loop.

No, I haven’t started drinking or experimenting with drugs. I’ve been doing something much worse: I’ve been clicking links to news articles on Facebook.

We should all know by now that this is a deeply unwise thing to do, because of how the algorithm works. It doesn’t select what to show you based on what is important or useful to your life, and it has long given up showing us things our friends have posted. Instead, the algorithm optimises for engagement – whatever keeps us clicking and scrolling the longest.

And unfortunately for me, that means a constant slew of articles and comments complaining about proposed data centres, usually published in the Guardian.

So now my morning routine is wake up, pick up my phone, load up Facebook, and find myself raging mad before I’ve even got out of bed.

Though in fairness to me, this isn’t just because I have a deeply dysfunctional relationship with my phone. It’s the Guardian’s fault too, as some of the claims the paper has been making about data centres are not exactly wrong… but they are lacking significant context in a particularly annoying way.

And this is something worth getting a little mad about. I wrote a couple of weeks ago about how we need more data centres for reasons of national security, and if you buy my argument, then we need to be rational about the trade-offs involved in data centre construction.

So this week, I want to focus on three recent Guardian data centre stories, and why it’s actually a little more complicated than they first appear.1

I can only do this sort of mildly unhinged deep-dive with your support, so please consider subscribing (for free) to get more of this sort of thing direct to your inbox. Or if you really value my work, please consider taking out a paid subscription.

Subscribe now

1. Flights of fancy

Let’s start with a story that the Guardian ran last month, which was originally headlined “Proposed London datacentre will have annual carbon footprint of 27,000 flights to New York”.

The piece focuses on the proposed East Havering Data Centre Campus (EHDCC) in North Ockenden, on the outer edges of London, and it claims that:

The 1.2m tonnes of CO₂e the EHDCC expects to emit would amount to 26,795 long-haul flights from London Heathrow to New York’s John F Kennedy international airport, according to the UN’s International Civil Aviation Organization’s carbon calculator.

Unfortunately though, it appears that someone screwed up the calculation – as the story was later revised and it instead puts the estimated emission at the equivalent to 8,480 flights instead, which the Guardian notes in a correction.

However, this isn’t actually my problem with this story. Flight emission calculations are always going to be shaky, as calculators necessarily make a bunch of assumptions that can steer numbers in wildly different directions. For example, the UN calculator linked in the excerpt clearly does a distance calculation, as it asks for departure and arrival airport – but it isn’t concerned about the type of aircraft that actually flies the route, even though that too can make a significant difference.2

But anyway, none of this matters, because what I’m more bothered by is the claim that sits behind that: the idea that the new data centre campus will lead to an additional 1.2m tonnes of CO2 emissions every year, because once you dig into how it was calculated, things start to get extremely weird.

So where did the 1.2m come from? Surprisingly this wasn’t (or doesn’t appear to have been) provided to the Guardian by an anti-data centre activist group, but it’s a number taken from the actual planning documents provided by the actual developer.

So, case closed, right? Well, I don’t think so – and to explain why, we need to do some maths.

Summing up the problem

In total, the completed EHDCC is expected to be a 600MW campus in total. Data centres are measured in terms of energy usage rather than, say, processing power, as ultimately that is what matters more. The chips will be replaced every few years with something a little faster, so what really matters is how much energy is used by the site, which is more fixed.

To start then, if you look at Technical Appendix 14.1 of the planning documents, you’ll discover that the developers started by taking the technical characteristics of a representative 50MW data centre and basically scaled it up. So all of the sums were done for 50MW to start with.

Then we reach the first choice the modellers have made. There’s basically an implicit assumption throughout the calculation that the data centre will be working at 100% capacity 24/7 throughout its lifetime. This is something the government has disputed in response to criticisms of other data centre projects, but I think this assumption is fair enough to make, at least when modelling energy usage. If AI is as transformative as it appears to be, the demand for inference is going to be almost limitless.

So we start with 50MW of usage – but in reality a ‘50MW’ data centre will actually use more than 50MW of energy, as the headline figure refers only to the energy used by the computers crunching the data, and not the supporting infrastructure like air conditioning, lights on the ceiling, or the microwave in the canteen.

So to account for this, in the modelling the 50MW figure has been multiplied by 1.326 – what is known as the “Power Usage Effectiveness” ratio. This multiplier just comes from the reference figures for an equivalent data centre. So really, according to the reference data, a 50MW data centre should expect to use 66.3MW of electricity.

And then we have to multiply it again to turn that into a prediction of annual energy usage. So we multiply 66.3MW by 24 hours a day and 365 days a year, which gives us approximately 581GWh of electricity usage in any given year. Phew!

But we’re not done yet – now we have to work out how all of the energy usage translates to emissions.

The technical appendix says that to work this out, the developers have used a figure of 0.14kg of carbon emissions per kWh, from the government’s Standard Assessment Procedure (SAP 10.2), which is, well, a standard used across the construction industry for estimating emissions.

So next we have to do a bunch more multiplication, to come up with an emissions estimate:3

Now, mathematicians reading will note that the final figure – roughly 1m tonnes when rounded up – is still less than the scary-sounding 1.2m tonnes figure. So what about that last little bit?

Unfortunately there isn’t a clean calculation page in the planning documents, but the developers do say that an “uncertainty factor of 15% has been applied to the carbon intensities”, which would bring us to around 1.12m tonnes per year if this is what has been done here.

Then finally, on top of this there are a few extra components added on to the calculation. The developer’s model also takes into account ‘embodied’ emissions from construction, but on the flip side it also negates some emissions by taking into account the fact the build-out of the campus will be phased and that the full site won’t be up and running until 2043.4

And that’s what gets us to approximately the average figure of 1.2m tonnes per year figure published by the Guardian. Phew.

Insane assumptions

Now, this sounds super comprehensive, right? I’ve described an incredibly detailed model of the emissions, running to dozens and dozens of pages in the planning documents. It certainly has the appearance of being thorough and thoughtful.

But did you spot the deeply strange assumption that has been baked into this calculation?

That’s right, it’s that 0.14kg of carbon dioxide per kWh. For the sake of the calculation it is treated as immutable, even though it really shouldn’t be.

And the technical appendix inadvertently makes it clear why this, in my view, is a massive flaw in the maths. It says that:

“Future Energy Scenarios which predict future decarbonisation of the electricity grid have not been applied.”

In other words, this entire carbon calculation is based on the idea that over the next six decades Britain will not decarbonise any further than it has now. Which would obviously be an absolutely mental assumption to make.

If the 0.14kg stays constant, it assumes that Britain will not reduce the carbon load on the grid by building any more wind farms or solar panels. It also assumes that two major nuclear plants, Hinkley Point C – and Sizewell C – will not come online, even though the former is half complete. It seems to assume that we will not increase the proportion of renewable energy on the grid from now until I am 99 years old in the year 2086.5

In other words, what’s happened is that for some reason the developers have deliberately modelled what they call in the documents a “reasonable worst-case scenario”, and the Guardian has run with it, without mentioning this worst-case caveat, even though the odds of it actually happening are basically zero.

So what will EHDCC’s emissions be instead? It depends on how aggressively we decarbonise the grid – but I’m very confident that it will be significantly less than the worst case scenario, especially if the government continues to achieve its decarbonisation goals and (more importantly) the costs of solar and battery storage continue to dramatically fall.6

Now, I don’t blame the developers for doing the calculation in this way. Presumably they had no choice7 – otherwise you’d expect developers to pick the most optimistic numbers possible.

But once you step back and take the time to think about it, as the Guardian probably should have, it strains credulity. And to pretend otherwise is not going to help readers form an accurate picture of the world.

Subscribe now

2. Water use

Now let’s turn to a story the Guardian published in July, titled “Not enough water for UK’s datacentre plans, trade body says”.

Water UK, the trade body that represents water companies in Britain and Northern Ireland, said in a publicly available written briefing to MPs that the government’s water forecasts were “fatally flawed” and “explicitly exclude” datacentres.

It wrote that the government “makes not a single mention of water” in its policy for delivering AI growth zones, its AI for science strategy or its AI opportunities action plan.

There are plenty of legitimate reasons to be sceptical or even mad about the AI buildout. The implications of the technology are uncertain, and despite the nonsense in the first section, it is true that AI data centres use a non-trivial amount of electricity in their operations, which we need to plan for.

But there’s also a lot of bullshit too. Water usage for cooling data centres is one such example – a fear that has been consistently debunked by people like Andy Masley.

So that’s what makes this story so interesting – it’s the water industry itself seemingly, er, wading in, with its concerns about building more data centres. Perhaps it suggests there is something to worry about after all?

However, I think what is missing here is some pretty vital context, which should change our perception of just how big a problem this is.

But before we get to that, what do we know about AI water usage specifically? As Water UK notes in the briefing for MPs linked above, there are two factors that can dramatically change how much water a data centre uses.

First, whether or not the building uses an ‘open-loop’ or ‘closed-loop’ cooling system matters. An open-loop system ingests water, cools the servers and then the heated water evaporates in a cooling tower back into the atmosphere. This means it requires a constant supply of new, fresh water to operate.

Closed-loop systems, however, recycle the water they use. The water is heated by the computer equipment, but it is retained and cooled so that the same water can be used over and over again. This is obviously a much better way of doing it, as fresh water is only occasionally used to top up the system. The data isn’t great, but it appears that most UK data centres are thought to use this sort of cooling system – or have no water cooling system at all.

Then we get to the second reality of data centres in the UK, which Water UK also notes in its briefing. Water usage is highly dependent on climate. A data centre built in a desert will require far more water cooling than a data centre built in a rainy and grey Scotland. And in fact, this means that in Britain, water cooling systems simply won’t be switched on for much of the year anyway. According to TechUK, 64% of surveyed data centres use less than 10,000m³ every year, which the organisation says is less than a typical leisure centre.

Though to be clear, this doesn’t mean there isn’t a small, legitimate challenge here. According to Water UK, the problem is that even if demand for water cooling only kicks in on hot days, that’s exactly the same time when demand peaks more broadly, as everyone wants to keep cool by consuming water at the same time. So more data centres will conceivably exacerbate the peaks.

So what’s missing? Why am I complaining? Doesn’t this all seem pretty reasonable to report? On one level, perhaps – but it also has a severe lack of context.

I really recommend reading this excellent piece by Rowland Manthorpe,8 who has been investigating water usage in British data centres for some time.

In his piece, he looks at two scary-sounding stats on water usage – one of a Scottish data centre that reportedly uses 27m litres per year, and an estimate published by The Times that data centres in the Thames Water region use 10bn litres per year. That sounds dramatic, but it works out as just 0.005% and 0.2% of the water supply respectively.

So even if we multiplied the number of data centres and the water they were using many times over, it would still be a microscopic proportion of overall water usage.

And this is ultimately the problem with the Guardian piece too: it doesn’t mention the denominator. To go back to Water UK’s briefing, it says there are 14 billion litres of water supplied every day and data centres currently account for 6.6 million of those litres – so around 0.047% of Britain’s daily water usage. So this is all to say that ultimately, compared to everything else, data centre water usage is basically a drop in the ocean.9

3. What the FOI?

The final story I want to look at is the most recent – as it’s very, very silly.

“UK government does not know what datacentres are being used for, FoI request shows”, reads the Guardian headline, and the story sounds pretty damning:

When asked in a freedom of information (FoI) request who was using the largest sites and what for, the Department for Science, Innovation and Technology said it did “not hold the requested information”.

Sounds bad, right? But the reality is that this appears to be hilariously tenuous.

The story begins over a year ago, on the 25th August 2025, when Oliver Hayes from a charity called Global Action Plan submitted the following FOI request to the Department of Science, Innovation and Technology (DSIT):

Dear Department for Science, Innovation and Technology,

Please confirm whether you hold information on the companies, organisations or bodies currently using the UK’s largest (more than 15MW of rated IT capacity) data centres and the purposes for which they are used.

If this information is held, please provide a list of these businesses and purposes. If it is not possible to provide details on individual companies, please try to provide a breakdown of the types of companies, organisations or bodies accessing the UK’s data centres and the proportion of the UK’s data centre capacity that these types of industry currently utilise.

Thank you in advance for providing this information,
Oliver Hayes on behalf of Global Action Plan

The department then responded on the 12th September last year, confirming that it does not hold the requested information.

And what you might not realise unless you’re the sort of weirdo who sends FOIs regularly is that this is actually a pretty narrow response.

The way the Freedom of Information Act works is that anyone can request information from a public body or government department – but only if a document containing that information already exists.

In other words, all this response is really saying is that there is no document that exists in the department which contains a list of data centres in a table alongside the businesses that use them and for what purpose.

So it’s less of a damning response, and arguably more Hayes encountering a deliberate friction hidden within the law. Yes, information is free, but you have to know what you’re looking for first.10

And as much as it pains me to say, as someone who regularly sends FOI requests, this makes a lot of sense. If any random member of the public could write in and essentially task civil servants with a homework assignment to put together a new document for them, the business of government would quickly become unworkable.

But anyway, let’s be more generous and assume that DSIT doesn’t actually hold the information at all, in any form. My next question is… so what? I can’t quite work out what response Hayes was expecting it to reveal or why it would be particularly useful for the government to know that, say, Netflix is using capacity in one of Amazon’s data centres, or OpenAI is using capacity in one of Microsoft’s.

So I think what this story really is more than anything is just a spectacular PR success on the part of Global Action Plan. With an extremely thin justification – an FOI with a boilerplate non-response, sent over a year ago – they’ve managed to get their talking points published as a big story in a major news outlet, on the strength of basically nothing. Bravo!

Facts x Importance = News

So there are three stories about data centres and why they make me mad. Like I say above, none of them are false, exactly. But in each case, scary-sounding numbers and facts become significantly less scary once you add a little context – and the reason I’ve spent several thousand words banging on about this is because I think this really matters.

As has become clear from the coverage, data centres are a new, salient dividing line in politics, and are fuel for the next round of the culture wars. If we’re going to navigate the build-out successfully, we’re going to need to be working from a common set of facts about the world – and we should want news coverage to clearly explain the real trade-offs involved in both building and not building.

And I should probably spend less time on Facebook getting mad.

Phew, another long one this week. If you enjoyed reading this then make sure you subscribe (for free) to get more like this in your inbox. Or take out a paid subscription if you really value my work. And if you’re from the Guardian, please don’t break my legs.

Subscribe now

Follow me on Bluesky

On YIMBY Pod this week:

• We talk about the security implications of the latest AI advances – and how the British government should respond to threat. And more importantly, James claims victory in his argument with Martin over AI.

• Then we speak to the brilliant Jonn Elledge about his new book, 31 Inventions that Built Our World.

James recently wrote about the national security case for building data centres, and how to fix Top Gear.

Martin’s Substack is here, and his YouTube channel is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

ICYMI: I got so annoyed by bad data centre takes last weekend that I decided to remove the paywall from my recent post on Britain’s looming AI national security crisis. Go and read it if you missed it!

I’ve also written about the slightly less important issue of how I picked a fight with Jeremy Clarkson (which he then went on to win).

POD! On YIMBY Pod this week, Martin and I talk about a Cambridge new-build estate where it is literally illegal to recycle rainwater – and we speak to Sunny Hundal about his new campaign to make energy both clean and cheap. Listen here, or wherever you get your pods.

Hello! It’s time for Odds and Ends, your semi-irregular round-up of the most interesting links I’ve seen recently, plus some shorter takes.

This time I’m featuring:

• How the UK consensus on Net Zero has broken down

• Why globalisation can’t be stopped

• The case for municipal bond markets

• Andy Burnham not being serious about Freeview switch-off

• A tiny sign the Democrats are taking winning seriously

• A great TV drama about feminism in the 1970s

• Can AI reason itself to constitutional government?

• An amazing four hour podcast about the history of electric grids

• And a sign the future could entail a whole new season of Star Trek: The Next Generation (!).

Now let’s get started!

Read more

BIG NEWS! I got so annoyed by bad data centre takes last weekend that I decided to remove the paywall from my recent post on Britain’s looming AI national security crisis. Go and read it if you missed it!

POD! On YIMBY Pod this week, Martin and I talk about a Cambridge new-build estate where it is literally illegal to recycle rainwater – and we speak to Sunny Hundal about his new campaign to make energy both clean and cheap. Listen here, or wherever you get your pods.

Jeremy Clarkson once brutally humiliated me.

Read more

On YIMBY Pod this week:

• It turns out it is literally illegal to recycle rainwater and use it to flush your toilet – and this is causing problems in Cambridge.

• And we speak to Sunny Hundal about his new campaign to make energy both cheap and clean at the same time!

The Guardian reporting on Cambridge’s grey water problem is here.

The Cheap Clean Energy Alliance is here.

This week James wrote about the national security case for building data centres.

Martin’s piece on ‘picky bits’ and national decline is here, and his YouTube channel is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.

POD! On YIMBY Pod this week, we talk about the government’s latest efforts to make it easier to build near train stations and densify our cities. Then we speak to Nick Maini about why Hammersmith Bridge has been closed to public transport for six years – and how autonomous pods could fix it. Listen here, or wherever you get your pods.

I swear I don’t mean to keep writing about AI, but it’s the biggest story in the world, and I think this is particularly important…

Every so often, our lives are punctuated by shocking events that change everything.

9/11 is an obvious example. I was 14 when it happened, and though I didn’t fully understand it at the time, I still remember having the visceral feeling in the pit of my stomach that nothing was ever going to be quite the same again.

I felt something similar a few weeks ago. I was watching two OpenAI employees give a presentation at the DefCon security conference, detailing what happened in the now-infamous cyberattack that OpenAI’s AI models autonomously carried out against Hugging Face, another AI company.

In case you missed it, the technical details are complicated but the story is simple. An AI model was being evaluated internally. It had essentially been tasked with an assignment to solve a particular digital security puzzle. But due to an error on the part of the testers, the AI was unable to access some of the documents it was supposed to use to find the solution.1

However, instead of giving up and declaring the task not possible, something extraordinary happened. Over a period of a couple of months, the AI explored alternative solutions to the problem it had been tasked with, and essentially discovered that it could create a 'message board’ where it could interact with other AI models inside the company. Then astonishingly, unbeknownst to any human, the AIs figured out how to work together to not just break containment and access the internet – but they reasoned to themselves that the best way to solve their problem was to hack Hugging Face. Which they did successfully.2

I’m assuming that after this, the examiners were not too worried about the examination, as this whole incident was an absolutely spectacular demonstration of just how sophisticated AI has become in such a short space of time. Give a sufficiently powerful model a task and it will pursue its goal relentlessly until it finds a solution.

This brings me to Britain.

Though we are only a middle power, and the ‘frontier’ models like ChatGPT and Claude are being developed elsewhere, we have so far carved out a rather successful niche in the AI present. The AI Security Institute (AISI) was created by Rishi Sunak, and in less than three years has proven itself seriously credible in the eyes of both the tech industry and foreign governments.

AISI has assembled a team of world-leading experts who, when new AI models are created, can credibly evaluate the threats they may pose. When Claude Mythos, a powerful new model, was created by Anthropic earlier this year, AISI was granted early access to test the model – and it published a warning about its capabilities, which was taken very seriously.

However, as effective as this new institution is, following the Hugging Face incident I’m not convinced it is enough if we want to protect Britain and our allies from AI threats in the future. 9/11 was a clear indication that global peace and stability were about to be upended, and Hugging Face is the same. And unless Britain acts now, it won’t be ready for the threat that is rapidly emerging.

So this week, please allow me to explain the emerging problem – and what we can do about it.

I’ve got so annoyed about the discourse on this, I’ve now removed the paywall on this post. So if you find it useful or informative, please consider subscribing.

Subscribe now

The new security threat

As Ben Thompson has observed, the next few years are going to be a security nightmare.

However optimistic I may be about the potential good things that AI will make possible, there is going to be an awkward period of adjustment as sophisticated AI models pick over the code that humans have written over the last half-century. AI will reveal bugs and vulnerabilities at an unprecedented scale – in software that we use every day, and in code that underlies the infrastructure of the modern world.

This isn’t because the code is bad, per se. It is because writing code is inherently an imperfect process. An operating system like Windows or MacOS, for example, is the combined work of tens of thousands of people. It is so bewilderingly vast and complicated, that it simply isn’t possible to ensure that it is impervious to threats.

Think of it like an airport. We can add layers of checkpoints, bag searches and x-ray machines to reduce the security risks, but it is impossible to make such a complicated building completely safe.

So since the dawn of computing, cybersecurity has been a permanent fact of life. The bad guys find vulnerabilities, and the good guys update the code to make it safe again. This is a never-ending process, and it is the reason why with annoying regularity Windows will insist on installing updates when you power down, or MacOS will randomly reboot to update.

For a long time, the game of cat and mouse has just about held – but now AI is getting involved in finding vulnerabilities. And we’ve already seen what that looks like – it’s overwhelming.

When Claude created Mythos, it let the makers of some of the world’s most critical software use the new model to look for vulnerabilities in their own code and the results were striking. Cloudflare, which provides important internet infrastructure, found over 2,000 bugs (including 400 critical security issues), and Mozilla, which makes the Firefox web browser, found 271. And in addition, over 3,900 high or critical security vulnerabilities were found in an analysis of various open source projects.

And it is not just Mythos that is the problem. Earlier this summer, OpenAI used its latest model, GPT-5.6 to find numerous bugs in apps like Chrome and Safari, as well as in the kernels of Linux and FreeBSD3 – two operating systems that run the vast majority of the server infrastructure across the internet.

Pandora’s box

Perhaps you can see the problem. We’ve now got highly sophisticated AI tools that can detect vulnerabilities at unprecedented scale, far faster than they can be patched by traditional methods – and they will relentlessly pursue the goals they have been given. This is not a hypothetical future – this is the world we live in today. And right now is the worst they will ever be.

In the early days of the post-ChatGPT AI explosion, some serious people talked about how governments could set rules to slow or regulate AI progress. This was basically one of the reasons the AI Security Institute was created in the first place – to give the government a role in deploying this potentially dangerous new technology.

But now it is clear that the brakes cannot be applied.

If we lived in a world where OpenAI and Anthropic were the only companies that had this technology, it might conceivably be possible. But unlike other dangers like, say, nuclear or biological weapons, we can’t control AI models with treaties or restrictions on materials. The technology runs on commodity computer hardware and the models are ultimately just large text files full of numbers.

And even if we could, somehow, control these things, then there’s the problem that other countries are working on AI too, including most notably, China. And this makes clear why we now face an unprecedented security threat.

So far, China’s AI progress has lagged the US, but it has consistently only been behind by months. When it was first released last year, DeepSeek caused a minor freakout as it was almost as sophisticated as the best ChatGPT model available at the time, and shockingly it ran much more efficiently. Then more recently, another Chinese model, Kimi K3 has done the same. Though comparing AI models like-for-like is difficult, on some benchmarks it is only narrowly trailing the capabilities of GPT-5.6 Sol and Claude Fable.

This is to say that the genie on this technology is already out of the bottle.

Perhaps if Chinese companies had pursued an approach to development similar to OpenAI and Anthropic, it might be possible to just about imagine a modern equivalent to an arms control treaty between the major powers. Maybe.

But the problem with this though is that China is pursuing a completely different AI strategy, and its government and businesses have wildly different incentives.

As China is currently trailing US capability, most Chinese models are open source and/or open weight – meaning the raw code and data used to create the models have been published freely for anyone to use.4

Why would Chinese companies do this, given that Anthropic and OpenAI have both received astronomical valuations in part by keeping their models closed? Isn’t that the ‘secret sauce’ that makes the companies so valuable? The difference is that because China is trailing on raw capability, it has decided to compete instead in the ecosystem around the models. (This is a strategy that has come all the way from the top, with Xi Jinping himself calling for this.)

So the hope is that people and companies will choose to use Chinese models because open models can work out cheaper to run. Openness also creates an incentive across the companies that use them to work out how to run them even more efficiently, reducing the need for cutting-edge Nvidia AI chips, which are subject to export controls that have been imposed on China by the West.

And though this is great from an AI-ecosystem perspective, arguably for everyone in the world except Anthropic and OpenAI, openness does also have one difficult geopolitical consequence: As anyone can use them, it really does mean that anyone can use them.

That could mean that rival powers like China and Russia, rogue nations like Iran and North Korea, or hackers and terrorist groups. They can take these open models and do with them, well, whatever they like. Unlike closed models, where usage runs through OpenAI and Anthropic’s servers, if you have enough compute capability you can run these models without restrictions on nefarious uses.5

So this is why I think we’re hurtling towards an AI security crisis. Though the open-weight models are still not as capable as GPT-5.6 Sol, or Claude Fable, if history is any guide they are not far behind. And when they catch up, there will be no effective limitations on their use – we will have to assume that the worst possible person or group, with the worst intentions, will make use of these models too.

And that’s why we need to prepare for that world now.

A war of resources

Let’s go back to Britain.

Choose your own ominous metaphor: It’s 1938 and Germany is rapidly rearming. It’s 2001 and a CIA memo is warning that Bin Laden is determined to strike the United States. It’s 2013 and Vince Cable is signing the paperwork to sell off the Postcode Address File. It’s 2020 and there are reports of a strange virus in Wuhan.

In the not-too-distant future, it is incredibly likely that we will be living in a world where sophisticated AI models, which cannot be controlled or regulated out of existence, can be deployed to relentlessly and autonomously assault every digital system in our lives by actors unknown.6

How does global power work in this world? And how can Britain protect itself?

In terms of the mechanics of defending, the old paradigms of cybersecurity are dead. Any human-led efforts to patch and update software will simply move too slowly. They will not be able to keep up with the AI assault.

So the weapon we will need to use to defend ourselves is, well, AI. Just as AI can find vulnerabilities at speed, it can also conceivably patch them just as quickly too.

In other words, the only way to stop a bad guy with an AI, is a good guy with an AI.

If you’re a traditional IT security person reading, this might sound like a heresy. What I’m arguing here is that it simply won’t be possible to have a human in the loop, reviewing code before it is patched in. We will need to defer authority to defensive AI systems to update critical systems to defend them. There will be no other way to do this.

And the consequence of this fact is something quite profound, I think, as it essentially reshapes global power.

Power has always been a function of hardware. The number of tanks, missiles and (increasingly) drones that a country can deploy defines its ability to defend itself.

But in a post-Hugging Face world, the hardware we need has changed again. Simply put, power in the new world will be a function of how much compute a nation has at its disposal. If every digital system is under constant attack by autonomous AI agents, we will need to have the capability to stand up to such an assault.

And, sadly, I’m just not convinced that Britain is appropriately readying itself for the battle ahead.

Protecting British interests

So how is the British government supposed to respond to the new reality?

If we want to put ourselves in a strong strategic position, there are essentially three core interventions we could make in the AI ‘stack’: on models, compute, and energy.

On models, if we had our own frontier models, it would put us in a strong position. Our AI could literally outsmart the other AI. This is how we get away with having relatively few soldiers – we rely on our missiles and our planes being better than those of our enemies.

But sadly, in the case of AI we do not have our own frontier models – and developing one would be extremely difficult.

And it’s worth dwelling on this to understand exactly how difficult it is. We’re not just one extra commitment casually thrown out by Andy Burnham away from making it happen, as catching up with the best American and Chinese models would be almost impossibly expensive.

For example, Meta – which is desperate to catch up with OpenAI and Anthropic – has reportedly already sunk over £100bn this year alone into its ‘superintelligence’ lab, and it still appears to be nowhere close to developing a similarly top-tier model. And for context, that’s equivalent to around half of the NHS’s annual operating budget. So in terms of public spending, building our own model is a non-starter for obvious political reasons.

Instead, our only option model-wise is to hope that, despite the Trump of it all, the United States continues to let Britain access the most sophisticated models. This is why having AISI is such a useful bargaining chip as America likes its credibility too.

However, assuming we have a sophisticated model, we need more than this too – we need the physical hardware to run it on. And this is where I do think we can do something useful.

If power is a function of compute, then we can simply build more data centres, and generate more energy with which to power those data centres. The more AI tokens we can crunch, the stronger our defences will be against the threats to come.

And to an extent, this is something the government is already working on. There’s no indication that Andy Burnham intends to change the AI Growth Zones plan, and there’s still a pipeline of renewables coming online (along with Sizewell C and Hinkley Point C). So in due course, we’ll certainly have more data centres and energy generation to power them.

But my concern is that given the new security paradigm I describe above, this still isn’t fast or large enough.

For a start, there’s a huge misalignment in terms of timelines – data centres and power plants currently take years to build, and yet AI progress, and the lag between the frontier and Chinese open source models can be measured in months. It will be surprising if China doesn’t have a model as powerful as GPT-5.6 or Mythos by Christmas.

And this wouldn’t be so bad if AI was just an economic nice-to-have. If we don’t have the latest technology the fastest, it merely means sluggish economic growth and a loss of competitiveness to our peers.7

But viewed from a national security perspective? Suddenly that capability gap feels a lot more urgent.

However, there is one reason for optimism here. As urgent as the need to build our AI infrastructure is, it need not be costly to the public purse. Unlike building a frontier model, which would require enormous upfront capital spending, data centres and new sources of electricity are economically useful in their own right. They can both help the economy and act in a dual-use capacity for our cyber-defences, should we require it.

So this means that new data centres can pay for themselves. In fact, given the AI boom, the private sector is pretty desperate to build out capacity. All the government needs to do is, well, let them build.

And this is why, finally, I want to end by proposing a policy that is simultaneously deeply important if we want to protect Britain’s national security, while also perhaps the single most electorally unpopular idea anyone has ever proposed.

I’m proposing that the government needs to create ‘Nightingale’ data centres.

These are, of course, named with reference to the conference centres full of hospital beds that were spun up in the early days of the pandemic. What’s remarkable wasn’t that they were not actually all that useful (we didn’t know that at the time) – what matters is that we built them quickly.8

AI data centres are more technically demanding than a room with hospital beds. And I’m not proposing we start setting up server racks in the ExCeL Centre. But we should take that same urgent spirit and apply it to the data centres we’ll need in a new world that has arguably already emerged.

At a minimum, plans for data centres and energy infrastructure should be rapidly called-in by the government and even more rapidly approved. Process should be vastly curtailed, and every official should be driven by the need for rapid delivery. Like during Covid, we correctly decided that now wasn’t the time for layers of consultation and review – the need for compute should be approached the same way.

I still believe that, ultimately, AI will do incredible things to improve our lives. But the Hugging Face incident was a warning, like the Bin Laden memo sent a few weeks before 9/11, and we should act now to avoid the inevitable security crisis once the Chinese models catch up.

Subscribe now

On YIMBY Pod this week:

• The government’s latest planning rules are making building near train stations easier – hurrah!

• And we speak to Nick Maini about his proposal to build a new autonomous transport system to cross the beleaguered Hammersmith Bridge.

The Sam Dumitriu post on the NPPF we talk about is here.

Barnes Hammersmith Electric Light Transit are here.

This week James wrote about principles and the death of Jason Arday.

Martin’s piece on ‘picky bits’ and national decline is here, and his YouTube channel is here.

Don’t forget to tell your friends about the pod – and leave us a nice review on Apple Podcasts!

YIMBY Pod is available on Apple Podcasts, Spotify, YouTube, Substack and hopefully wherever you get your podcasts!

You can find James on Bluesky here, and Martin here.