What's Past is Prologue

Generative AI's Dream is a Thatcherite Nightmare

Credit: University of Salford

Note from Matt: Apologies for the radio silence the past couple of weeks. This piece, which I spent a lot of time on, should make up for it. I have some new ones in the pipeline, too.

If you like this and want to support my work, consider signing up for a paid membership. You get regular free posts, as well as some warm fuzzy feelings.

One last thing: This post is long and won’t show up properly in your inbox. To read the whole thing, open it in your browser or the Substack app.

---

I’m a bit behind on writing this newsletter — for which I can only apologize. The truth is that I have three stories, all partially complete, ready for me to cross the finish line. My inability to do so has been, in part, because I’ve felt extremely down.

I’ve been unable to shake the feeling that, no matter how soon the generative AI bubble bursts, so much damage has been done, and will continue to be done. Perhaps the first big blow to my mood was the reports that Amazon is preparing for a labor-free workforce — with the company believing that automation and AI will replace 600,000 workers by 2033.

To be clear, Amazon’s a horrific employer — particularly for those working in delivery and fulfilment — and I don’t believe that automation will be able to deliver even close to the kinds of job cuts it’s aiming for, at least by those deadlines. And yet, this news served as (if we needed one) yet another reminder of the fact that big tech doesn’t care about people.

Then, Amazon announced swingeing 30,000 cuts to its existing workforce — which it said absolutely wasn’t down to AI, or the insane capex commitments it’s made to support generative AI, but because of “culture.”

To be clear, if you decide to fire 30,000 people in one of the worst job markets in recent memory for a reason as nebulous as “culture,” you’re a scumbag. But also, it’s absolutely a lie.

For the past week-and-a-half, I’ve been reading through the past few years of quarterly and annual financial reports of the five companies most acutely exposed to the AI bubble — namely Oracle, Microsoft, Meta, Google, and Amazon. The research was for Ed Zitron’s most recent premium article, where he calculated how much revenue these companies will need to generate exclusively from AI in order to actually make their investments worthwhile.

These reports made for sobering reading. Since 2022 (but, really, since 2023), these companies have racked up hundreds of billions of dollars worth of debt and long-term lease agreements. Whether their bets on AI fail or succeed, they’ll have to service these obligations — with some of them ending when I’m (at least, in theory) supposed to be retired.

By the end of last year, Microsoft had nearly $400bn in long-term payment obligations. That’s roughly the annual GDP of Denmark. And almost all of those obligations are a consequence of its exposure to the generative AI boom.

It’s insane. Maddening. And there’s no way that this shit isn’t going to end up hurting people. Eventually, the bill comes due — and there’s no way to reconcile these long-term obligations with, say, the need for companies like Microsoft and Google to pay shareholders dividends, or run share buybacks.

And so, we’ll see cuts and cuts.

But here’s the thing: it’s not like Jassy, Nadella, Pichai, Zuckerberg, Altman, or any of these reprehensible toads feel even the slightest bit of emotional turmoil about what they’ve done, or what they will do.

I believe that these people are, at heart, bad people. I also believe that they are profoundly stupid, and through their wealth, they have alienated themselves from the human experience — meaning they are unable to understand why someone might be a bit miffed about getting fired at a time when tech hiring is non-existent, and job security means “as long as it takes for an Infosys consultant to learn how to do your job.”

I don’t say this because I’m horribly prejudiced against these fabulously wealthy scumbags — though I am, and you should be too — but because their own statements reveal their motivations.

Like when Sam Altman says that some of the jobs he hopes to eliminate through his technology were not, in fact, “real work.”

It’s when Dario Amodei gloats about how half of all entry-level jobs will disappear in the near future to a credulous Axios reporter, who fails to ask whether this is true (it isn’t), or even the simply question of “wouldn’t, uh, this be really bad?”

It’s when any tech CEO talks in a hand-waving way about AI eliminating jobs, but just shrugs and says “we’ll figure it out” — given that, even now, without the specter of the tech-driven annihilation of employment, we’re utterly failing to provide a robust safety net.

As a side note, any journalist who interviews a tech CEO who says that is obligated to ask: “well, does that mean your company is willing to pay more tax?” And if that CEO represents a company that aggressively avoids tax — which, in fairness, is most of them — ask whether it’s fair to expect society to pick up the tab, and not them.

And if they say “we comply with all the tax laws in all of the countries in which we operate,” it’s your job to say, “yeah, but you also use a bunch of complex tax schemes and trans-national corporate structures to limit your tax obligations, don’t you?”

Make the bastards squirm.

As I’ve said again, and again, and again, big tech hates you. They do not care if your living conditions crater. Even without AI, they were happy for your kids to attend crumbling schools, to drive on pothole-pockmarked roads, and to see the functions of the state slashed, eliminating basic social goods like early-years services and cultural spending. Your value to them extends as far as your ability to consume — and no further.

To be clear, I do not believe that generative AI will come close to delivering the crushing blow on employment that its boosters portend. It’s just not reliable enough. The economics don’t make sense. It kinda sucks.

But I do believe that the disdain that big tech holds for ordinary people — and their belief that said people are, essentially, interchangeable units of labor that can be replaced at a whim, whether that’s by an offshoring firm or by AI — is very real.

We need some clarity here. We need to be direct. These people are happy to bring about the conditions for mass, Great Depression-style unemployment, and the only thing that’s stopping them is the fact that their means of doing so isn’t particularly good, and it’s rather expensive to run.

If generative AI fails to deliver on these aspirations, those aspirations won’t go away. They’ll just morph. There’ll be a new technology, or a new strategy.

This is what keeps me up at night. This is what I’m afraid of. And it’s because I know precisely what it leads to.

And that’s what this newsletter is all about.

While I’ve repeatedly emphasized the fact that I do not believe the long-term goals of the generative AI industry are even remotely achievable (these goals, I add, continuously grow grander, with OpenAI signing more than $1tn dollars in compute deals, a truly absurd number that nobody should take seriously), I also believe that we need to discuss what a tech-driven annihilation of employment actually looks like.

As a Brit — and, in particular, one from Liverpool, a detail that’ll become pertinent later — I know what this means.

The Iron Liars

As a reminder, for this piece, we’re disposing our skepticism of generative AI and operating under the assumption that the predictions made by Amodei and Altman and others are, in fact, inevitable.

Mass unemployment will, inevitably, be a consequence of this — and we’ll talk about that later in the piece — but I feel like perhaps the biggest, and arguably most destructive, consequence will be a deepening of the stratification between the rich and poor, those with assets and those without, combined with an elimination in the ways to actually achieve economic mobility.

That’s the thing. Part of what I find so distasteful about the generative AI bubble is that the undercurrents driving it — the desire for the consolidation of wealth in the hands of a small few, although the current generation of shareholding-rich don’t wear suits but rather Patagonia gilets, and the disdain for the economic circumstances of the poorest in society — are so old, they’re boring.

The economic model of generative AI is Thatcherism by another name — a failed economic experiment that, because the Iron Lady “stuck it up [their] junta,” we’ve failed to acknowledge was, in fact, a spectacular catastrophe with dire generational consequences.

You might accuse me of reaching — trying to link my two personal bête noirs through a contrived thread — but, before you close the tab, I’d ask you to think about it.

Generative AI presupposes the consolidation of economic power into the hands of a small few, as wealth gravitates to those either building the models, or running the infrastructure that said models depend upon, or those that fund and own the companies that build the models.

OpenAI lost $11.5bn last quarter — a truly insane number, and roughly $2bn more than Merrill Lynch lost at the height of the global financial crisis, right before it had to be rescued by Bank of America. And that loss is not, by all accounts, an outlier — and it’s something that’ll almost certainly continue into the near future.

Anthropic spent a sum greater than the entirety of its year-to-date revenue on compute from a single provider — AWS. Add in the compute that it gets from Google Cloud, as well as its basic operational costs, and we’re looking at several billion dollars in losses. And, again, there’s no sign that’ll change in the near future — despite the genuinely laughable projections that Anthropic has made.

GPT-5 reportedly cost $500m per training run. Last year, Anthropic CEO Dario Amodei predicted that by 2025, models will cost as much as $1bn to train.

Despite the veneer of Anthropic and OpenAI as scrapy, disruptive startups, the reality is that they’re a facade for the ambitions of their backers — which are dominated by the biggest of the big tech companies.

Microsoft owns a 27% stake in the new for-profit OpenAI entity, with the possibility to increase that percentage over time, and has a revenue share agreement that sees it take 20% of everything it brings in. Anthropic’s biggest stakeholders are Amazon and Google, with other big tech investors including Qualcomm and Intuit.

Generative AI is legacy business, draped in the startup aesthetic. And, honestly, it reminded me of Thatcher’s economic legacy.

Thatcher famously privatized state-owned assets (including those that, by their very nature, were natural monopolies, like the provision of water or energy, and thus, didn’t really make sense as private businesses), ostensibly to turn the British public into shareholders of the utilities that were previously owned by all — only for said utilities to eventually gravitate into the claws of larger corporations and investment funds.

It was a bait-and-switch.

The largest shareholder in Thames Water — the company that provides water and sewerage for much of the South-East of England, or, at least, tries to — is the Ontario Municipal Employees Retirement System. Anyway, here’s an ad pitching shares in the various state-owned British water authorities.

The largest shareholder in BT is the Indian conglomerate Bharti Airtel, followed by Deutsche Telecom. Anyway, here’s Thatcher-era advert for the initial BT share sale, aimed at normal people

Blackrock owns the largest stake in Centrica — the company that operates British Gas, and which made insane profits in 2022, when prices spiked forcing consumers to choose between heating their homes or eating. Anyway, here’s an advert telling people to buy shares in British Gas from 1986.

Council houses — which kept a lid on housing prices in the UK, until the Thatcher-era government sold them off and stopped building more — are increasingly in the hands of private developers. Anyway, here’s an advert telling people to buy their council houses.

All of these programs, though draped in the cloak of mass economic empowerment, were scams, each designed to benefit those already wealthy, and I see similarities in the generative AI bubble.

Should generative AI become what its boosters promise, we’ll see a similar gravitation of wealth.

Sure, we’ll be told about the “new opportunities.” How anyone with an idea can vibe-code an app, allowing Brenda and Dave from next door to become a new Mark Zuckerberg, and how it’ll allow people to work faster and make more money as a result. But it’s all bullshit.

With models costing billions of dollars to train — and that’s before we mention the cost of inference — companies like Anthropic and OpenAI will essentially be the necessary feudal kings of the new economy.

I say “necessary feudal kings” because there’s, given the operational and development costs of generative AI, the pool of companies who can possibly replace them will be vanishingly small, and they’ll have to be backed by a bigger, much larger benefactor.

The direction in which wealth travels will change — albeit only superficially.

Whenever a Brit pays a utility bill, a chunk inevitably travels to the shareholders who now own those formerly state-owned enterprises through dividends. Brits have subsidized nearly £85bn in dividends to water companies since privatization, all while the quality and cost of service has declined, and the infrastructure has crumbled to the point where human waste routinely flows into rivers, lakes, and coastal waters.

Last year, the water companies discharged raw sewage into these waters 450,000 times.

Altman and Amodei dream of a world where, in every economic transaction, a chunk ultimately lines their pockets, as well as the big tech companies, venture capitalists, and sovereign wealth funds that back them.

Generative AI is an exercise in centralization, with the generative AI companies and cloud hyperscalers acting as the nucleus of this new economy. Every transaction, every payment, every interaction resulting in cash gravitating towards them.

When Amodei talked about his vision of the first billion-dollar company with zero employees, save for the founder, what he was really describing was a world in which wealth inequality is even more stark.

Forget about the lip-service that Silicon Valley pays to economic mobility — the idea that if you’re an early employee, you can see your entire life transformed as your shares morph into gold — or even the RSU programs that many tech companies offer their staffers.

And then we get to employment.

Managed Decline

We’re often told that, for every job that AI displaces, another will take its place — something that begs the question “where” and “how” and “what will these jobs involve” and “how much will they pay?”

If the whole point of AI is to, in fact, save labor costs, the idea that this technology will create equal or similar jobs is, in fact, laughable.

Are we supposed to believe that those jobs will spawn from the hellmouthes of OpenAI and Anthropic? Although these companies have, admittedly, ramped up hiring over the past few years, they’re hardly labor-intensive companies. Anthropic, according to Wired, had 1,300 employees in May. In August, OpenAI had (according to The Verge) “roughly 3,000 employees.”

It’s, incidentally, eyeing up an IPO that would see the company valued at $1tn — a market cap that Microsoft only reached after nearly 45 years of existence — and assuming its headcount crosses (let’s be generous) 5,000, we’ll have an employee to market-cap ratio of 1:200,000,000, a truly insane figure.

The reason why I feel so strongly about the parallels between Thatcherism and the generative AI dream is because, even if the details differ slightly, I believe that the end outcome will ultimately be the same.

Thatcher entered office in 1979 and immediately began privatizing or liquidating state-owned entities. Unemployment doubled from 1.5m when she entered office, to 2.5m in 1981, to 3m in 1982, and it remained that high until early 1987, with a peak unemployment rate of 11.8% in 1984.

The young were, of course, the most disproportionately affected — with young people accounting for roughly one-third of those unemployed.

As I’ve written previously, generative AI’s most immediate victims in the labor force are those on the bottom rungs of the employment ladder.

Admittedly, an unemployment rate of 11.8% doesn’t sound too bad — it’s a couple of percentage points off the US peak during the global financial crisis, and a couple of percentage points less than the Covid peak — but it’s worth noting that these job losses weren’t distributed equally across the country.

In Northern Ireland — a nation already suffering from a prolonged sectarian civil war — one-in-five adults were unemployed. Traditional mining towns — like those in Yorkshire and Wales — faced the biggest brunt of the collapse in mining and heavy industries. In Norris Green, one of the poorest parts of Liverpool, which until recently, was the poorest city in the UK, youth unemployment was the highest in Europe.

Unemployment was, essentially, centralized and concentrated into specific locations.

Now, if we believe the most bullish projections for generative AI, the job losses from automation will be far worse than even the worst years of the Thatcher era.

And yet, Thatcher provides a useful case study of what happens when places are subjected to long-term periods of economic decline, deprivation, and joblessness. And what’s scary is that, even after a period of relative economic revival (which isn’t even guaranteed, assuming the mass-automation of labor), the scars still linger.

Bridgend, a mining town in Wales, suffered tremendously during the early 1980s. In 2009, it achieved global infamy for a cluster of suicides, mostly affecting young men. In 2013, it had some of the highest rates of antidepressant prescriptions in the entire UK.

According to UK government statistics, some of the cities most keenly affected by Thatcher-era de-industrialization and joblessness (Port Talbot and Burnley, to give two examples), had the among highest rates of opioid deaths in 2024 — with economic deprivation strongly correlating with the rate of overdose fatalities.

And then we come to my hometown. Liverpool.

In the 1970s, Liverpool was a hub of manufacturing and trade. Its docks connected the UK with the Americas, and the city’s factories dotted the skyline.

And it was devastated by the Thatcher premiership. While this was, in part, because the UK’s primary trade partners shifted after its accession to the EEC in 1973, making its docks, which faced away from Europe, less relevant, government policy played a major role.

As a convinced monetarist, Thatcher raised interest rates to battle inflation — which inevitably resulted in job losses — and reduced the amounts that central government paid to local councils, forcing councils to either issue redundancies or raise local tax rates. The city haemorrhaged jobs, both in the private and public sector. By one contemporaneous analysis, Liverpool was the UK city most affected by Thatcherite economic policy.

I’m trying to paint a picture here. Liverpool was a city that suffered a tremendous amount, and in a relatively short amount of time. And it was very nearly destroyed.

Liverpool experienced high rates of emigration, as people left to find work elsewhere in the country, or abroad. According to the 2021 census, its population was three percent lower than in 1980. Urban blight proliferated as entire streets were abandoned.

In 1981, Toxteth — then the poorest ward of the UK’s poorest major city, and the place I’m proud to call home — was engulfed in riots, partly due to heavy-handed and racist police practices against the ward’s black population, and partly due to the economic deprivation that had swallowed the city, which affected black residents hardest.

Similar large-scale riots would occur in Toxteth again in 1985.

Liverpool had become a headache for the Thatcher government, not least because the local government had embraced a policy of open defiance, and it was left searching for a solution. Behind the scenes, Thatcher was encouraged to simply… let the city die.

Ministers close to Thatcher proposed a policy of “managed decline,” allowing the city to simply bleed jobs and residents, becoming Britain’s own Detroit — a city where deindustrialization has led to entire ghost neighborhoods.

“It would be even more regrettable if some of the brighter ideas for renewing economic activity were to be sown only on relatively stony ground on the banks of the Mersey,” wrote Chancellor Sir Geoffrey Howe at the time.

“I cannot help feeling that the option of managed decline is one which we should not forget altogether. We must not expend all our limited resources in trying to make water flow uphill.”

Fortunately, this was advice that wasn’t taken — although it was only after the mass inflows of EU development money, and its city’s recognition as the European Capital of Culture in 2008, that Liverpool’s fortunes revived.

Still, the name Thatcher remains somewhat of a dirty word on Merseyside. The anger and resentment hasn’t gone away, and in fact, it’s passed down through the generations. Outside the ornate St George’s Hall, champagne corks popped when she did, as people raised a glass to the demise of someone who caused so much harm.

Similar scenes would be repeated elsewhere in the country, from Brixton and Glasgow, to the mining towns of Yorkshire, where a Thatcher effigy burned alongside a floral arrangement that read “SCAB.”

The point I wish to stress is that you can’t just destroy employment at a large scale without experiencing long-term second-order effects.

Thatcher privatized or closed state-owned entities, and raised interest rates to cut inflation, resulting in a half-century of national employment rates that exceeded 10%. And what happened?

Poverty. Joblessness. Mental illness. Crime. Violence. Deprivation. Urban blight. Addiction and drug-related death.

And those ills didn’t end when Thatcher was booted out of office, or when Tony Blair ended nearly two decades of constant Conservative rule in 1997. They still — still! — linger.

Are we so stupid to think that the same won’t happen if generative AI destroys employment? Do we really think that, if we bring about the kinds of unemployment we saw during the Great Depression, we won’t also see a spike in poverty?

In mental ill-health? In crime? In violence? In deprivation. In urban blight? In addiction and drug-releated death?

Most of my readers are American. This stuff probably sounds familiar to you, as it’s exactly what happened to the Rust Belt and the Appalachia. It’s no surprise that the former coal towns of West Virginia and Pennsylvania were Ground Zero for the opioid epidemic. We’ve all seen pictures of empty streets and factories in Detroit, where weeds poke through the asphalt of roads that nobody ever drives down.

Every country has a similar story. Belgium has Charleroi. Germany has the Ruhr. France has Marseille and its North coast.

We all know what happens when economic opportunity goes away, and how even a short-term blip can have long-term, even generational consequences.

As a reminder, I don’t believe that generative AI will be the thing that brings about the employment apocalypse. But I do believe that Altman and Amodei are comfortable with that happening — which says everything you need to know about their characters.

A Warning from History

Although Thatcher’s policies led to a prolonged period of mass-unemployment, the core functioning of the welfare state existed — although in a somewhat diminished form. The NHS, which provides healthcare that’s free at the point of delivery, wasn’t changed to a US-style system. Unemployment payments, however meager, continued to flow to those left standing on the dole queue.

Can the same be said for today?

The welfare state in the US is inconsistent and threadbare, and often time-limited. And even those programs that aren’t time limited are vulnerable to cuts and government shutdowns — as the current disruption to SNAP payments shows.

In the UK, we’ve seen the emergence of a grotesque “strivers versus skivers” narrative, with benefits claimants often presumed to be lazy, with their unemployment treated as a lifestyle choice rather than a product of economic circumstance.

Changes made during the David Cameron years means that people can have their benefit claims suspended — or sanctioned — for prolonged periods of time, often for the most trivial of reasons.

Fall ill and can’t attend a meeting? Fail to pick up the phone? Late for an appointment? Too bad. You’re not going to eat for the next month. If you think I’m exaggerating, go on the Reddit page for welfare claimants in the UK and look up “sanction.”

Over the past couple of decades, the culture in the UK has become far less generous to those least among us. In the US, that shift probably happened earlier, and something I imagine really starting with Reagan’s racist Welfare Queen trope. And so, I don’t think, as a society, we’re ready for mass AI-driven unemployment.

And that’s without saying anything about the economics of supporting a population for whom unemployment is now an inevitable and perpetual part of life. Where’s the money coming from?

Seriously, if unemployment spikes to, say, 40%, and remains that high for the indefinite future, how on earth are we going to afford to support them? How on earth will we pay for their housing, or their basic costs of living?

I believe that universal basic income (UBI) is a good, moral idea — and the only real solution to a world in which human labor no longer carries the value it once did — but the money for it has to come from somewhere. And while I’d love to see the rich taxed until their eyes pop, I also recognize that big tech is very good at minimizing their tax liabilities.

So, I ask, what happens next?

I’ve thought about this, and the only answer I can give is to advise you, the reader, to not be unlucky. That’s the best I’ve got. There but for the grace of God go I. And, for that matter, all of us.

And this is not, itself, a particularly original perspective. It’s essentially what Neil Kinnock, the former leader of the Labour party, and Thatcher’s one-time political rival, said in 1983 on the eve of a general election that would, regrettably, see her returned to office.

If Margaret Thatcher is re-elected as prime minister on Thursday, I warn you.

I warn you that you will have pain–when healing and relief depend upon payment.

I warn you that you will have ignorance–when talents are untended and wits are wasted, when learning is a privilege and not a right.

I warn you that you will have poverty–when pensions slip and benefits are whittled away by a government that won’t pay in an economy that can’t pay.

I warn you that you will be cold–when fuel charges are used as a tax system that the rich don’t notice and the poor can’t afford.

I warn you that you must not expect work–when many cannot spend, more will not be able to earn. When they don’t earn, they don’t spend. When they don’t spend, work dies.

I warn you not to go into the streets alone after dark or into the streets in large crowds of protest in the light.

I warn you that you will be quiet–when the curfew of fear and the gibbet of unemployment make you obedient.

I warn you that you will have defence of a sort–with a risk and at a price that passes all understanding.

I warn you that you will be home-bound–when fares and transport bills kill leisure and lock you up.

I warn you that you will borrow less–when credit, loans, mortgages and easy payments are refused to people on your melting income.

If Margaret Thatcher wins on Thursday—

I warn you not to be ordinary

I warn you not to be young

I warn you not to fall ill

I warn you not to get old.

I don’t believe that generative AI will result in the kind of layoffs and redundancies and firings that Amodei and Altman have foretold — at least, not directly. Like I’ve written in the past, AI provides excellent cover for outsourcing and offshoring, even if it isn’t itself doing any of the work that a human once did.

But what if I’m wrong and they’re right?

Then I warn you not to be ordinary.

I warn you not to be young.

I warn you not to fall ill.

I warn you not to get old.

God help us all.

---

Footnotes

• As always, you can reach out to me via email (me@matthewhughes.co.uk) or Bluesky.

• If you want to support this newsletter, consider signing up for a paid subscription. You get extra free posts and my eternal gratitude.