4 February 2026: Water | AI
We’ve moved from ‘water crisis’ to ‘Water bankruptcy’ // What if AI isn’t causing graduate job losses? [J2T #658]
Welcome to Just Two Things, which I try to publish once or twice a week. Some of these Things may also appear on my blog from time to time. Links to the main articles are in cross-heads as well as the story. A reminder that if you don’t see Just Two Things in your inbox, it might have been routed to your spam filter. Comments are open.
1: We’ve moved from ‘water crisis’ to ‘water bankruptcy’
The latest United Nations water report uses increasingly stark language to describe and redefine a burgeoning global water crisis. The report is headlined, “Global Water Bankruptcy: Living Beyond Our Hydrological Means in the Post-Crisis Era”, and they mean the metaphor to be taken seriously.
I’ll come back to the metaphor, but — as if there’s not enough to worry about right now — the story it tells about the state of global water supplies approaches the cataclysmic.
From the summary:
Nearly three-quarters of the world’s population lives in countries classified as water-insecure or critically water-insecure. Around 2.2 billion people still lack safely managed drinking water, 3.5 billion lack safely managed sanitation, and about 4 billion experience severe water scarcity for at least one month a year.
Surface water supplies are shrinking, and an increasing number of rivers simply no longer reach the sea (including, for example, the Colorado River). Half of the world’s major lakes have shrunk in size since the early 1990s. Huge areas of wetlands have been “liquidated”, if that is the right word, over the last five decades. Glacier loss exacerbates all of these trends.
(NASA has been tracking the steady shrinking of Lake Powell. Images: NASA/ United States Government.)
As a result groundwater supplies now supply only 50% of drinking water and 40% of agricultural water, and we are drawing heavily on water supplies from aquifers, much of which is not replaceable:
Around 70% of the world’s major aquifers show long-term declining trends. Excessive groundwater extraction has already contributed to significant land subsidence over more than 6 million square kilometers—almost 5% of the global land area—including over 200,000 square kilometers of urban and densely populated zones where close to 2 billion people live.
If that is the snapshot, the systemic effects that they create in combination is something of a “black elephant”: one of those sets of predetermined drivers that are individually problematic and collectively and systemically disastrous.
70% of global freshwater use is for agriculture, and 3 billion people, and half the world’s agricultural production, is in areas where total water storage is already in decline or unstable. This is worsened by soil degradation, because poor quality soil is less able to hold moisture. More than half of global agricultural land is now moderately or severely degraded, increasing the risks of desertification.
The combination of soil degradation, climate change and groundwater depletion also increases the likelihood of drought, which the report describes as “increasingly anthropogenic”, rather than it just being caused by rainfall deficits.
In addition, the water reserves that exist are also being polluted by agricultural run-offs, industrial or mining effluents, and salinisation, further reducing usable freshwater supplies.
And in sum, the planetary “freshwater boundary” has now been breached:
two important elements of the freshwater cycle—“blue water” (surface and groundwater) and “green water” (soil moisture)— have been pushed beyond a safe operating space.
(Source: UNU, “Global Water Bankruptcy, 2026)
The usual way we talk about global water is in terms of “crisis” or “stress”. The report has opted to talk about “bankruptcy” instead to underline the extend to which, as a species, we are running down our reserves of water, and in a way that may not be reversible. We are, if you like, spending our savings.
The language of “water bankruptcy“ comes from the hydrologist Kaveh Madani, and it is outlined in an accompanying (open access) paper to the report.
It starts with the way that the word “crisis” is typically understood:
a temporary departure from normal conditions, triggered by an acute shock... and followed by some form of resolution, either a return to a prior equilibrium or a transition to a new, more stable state... Extraordinary measures are mobilized for a limited period with the goal of “getting through” the shock and restoring functionality.
There are events in the world of water that fit this description. But increasingly, they do not:
[T]he similar patterns of chronic overuse and degradation across the world are not temporary deviations caused solely by climatic anomalies; they are the cumulative result of decades of systematic overspending of surface and groundwater, pushing systems toward their boundaries and into a failure mode.
(Source: Madani, Water Bankruptcy: The Formal Definition).
Madani draws on our understanding of systems to make this point clearer. The language of “crisis” speaks to the idea of shocks causing oscillations around a “stable baseline”. But this is no longer an accurate description of large parts of the global water system.
Instead, what we are seeing is a “long-term path-dependent transformation of the coupled human–water system”. The baseline itself has shifted.
The result is a twin crisis: “The bitter reality for many water systems worldwide is that they are facing both insolvency and irreversibility.” (Emphasis in original). And hence the language of bankruptcy.
Of course, the language resonates because it draws on ideas of financial bankruptcy, and in a New Scientist article, Madani makes the most of this, talking about “surface water” as being like a “checking account” (or current account), and that this is now empty. Groundwater, glaciers, and so on represent “the savings account that we inherited from our ancestors... they’re also drained now.”
But in practice water bankruptcy is worse than financial bankruptcy. Financial bankruptcy only involves insolvency, whereas water bankruptcy, as discussed above, also involves irreversibility.
The New Scientist also covers some of the results of looming water bankruptcy. In Madani’s home country of Iran, water shortages have been a significant factor in some of the country’s recent protests. Globally, it threatens the agricultural livelihoods of something like a billion people, mostly in lower income countries. In Bangladesh, the water supplies in Dhaka are being polluted by the chemicals generated by the ‘fast fashion’ industry, but effective regulation would damage business and employment.
The biggest implications here are that the world will have to learn how to live with less water, and manage water in different ways.
This is because “bankruptcy” is a structural condition, and needs a structural response. Water bankruptcy
occurs when natural income and liquid assets, even if fully mobilized, can no longer cover existing claims without unacceptable sacrifice of essential functions and damaging the natural capital. Water bankruptcy management... focuses on acknowledging both insolvency and irreversibility, recognizing claims, reallocating burdens, and designing a sustainable path forward under tighter constraints in a new reality.
In the ‘Water Bankruptcy’ paper, Madani suggests some structural steps that might help:
Admit defeat—honestly and early, because “Public trust is better served by transparent acknowledgment of limits and losses.”
Shift from supply expansion to demand reduction and reallocation, because “redesigning allocation through pricing, regulation, and negotiated settlements to prioritize essential uses and protect critical ecosystems” becomes a central policy task.
Protect remaining natural capital as the core asset, because “[e]ach additional centimeter of subsidence or square meter of dried wetland represents a loss of resilience that no engineering project can fully replace.”
Re-imagine development goals, because “[a] shift to resiliencemode accepts hydrological constraints and designs economies around them, through strategic trade, diversification, and managed retreat from unsustainable activities.”
Make bankruptcy management and reallocation just and inclusive, because “participatory processes and strong social safety nets” are essential “to prevent bankruptcy management from becoming a technocratic justification for dispossession.”
And of course, what’s true of water is true of quite a lot of other human-natural systems at the moment. This might also be a model for approaching some of these other systems. Because it’s not just water that’s gone down the drain.
2: What if AI isn’t causing graduate job losses?
I’d hoped to get to the recent report by Oxford Economics on AI and jobs a bit sooner. It is headed, “Evidence of an AI-driven shakeup of job markets is patchy”, and suggests that the story that AI is leading to lay-offs especially in entry level graduate positions could be wrong.
I’ll work through their evidence in a moment, but obviously when someone proposes that evidence that suggests that a widely held belief about the jobs market might be wrong, it should lead us to [1] question the belief, and [2] ask questions about what it is about the belief that has helped it to circulate.
The Oxford Economics Briefing is only a few pages, and comes with charts. It suggests four reasons why the ‘AI is replacing graduate jobs’ story might be wrong:
Globally, the markets where we’ve seen graduate employment increase are those where there is also evidence of economic slowdown;
There’s been an increase in the number of new graduates, for example in the United States, so some of this might be a supply side issue;
There’s no evidence of a productivity increase in companies reporting lay-offs, which you’d expect if it was a structural labour market change, not a cyclical phenomenon;
There’s not evidence of an acceleration in AI adoption among firms—if anything, it’s the opposite.
They also suggest that some of this story may be about causation, not correlation. So it’s worth going through some of the Oxford Economics charts in a little more detail.
Economic slowdown
(The rise in new graduate unemployment isn’t out of line with other graduate unemployment.)
This is US data, but it broadly shows that the pattern of unemployment among recent graduates, compared to all graduates, is similar to the historic data, going back to the turn of the century. In addition:
Economies, such as Japan and South Korea, where broad labour market conditions have held steady haven’t recorded declines in graduate unemployment rates. On balance, the cross-country data also seem to point to normal cyclical factors rather than structural AI-related developments driving changes in graduate unemployment rates.
Larger numbers of recent graduates
In case you’re looking at the chart above and thinking that the gap between the red and blue lines in the recent uptick is higher than previously, this could be because the supply of recent graduates has gone up, in both the US and Europe. In the US, the proportion of of 22-27-year-olds with a university education has risen from 32% to 35% since 2019. The Eurozone data is for 25-29 year olds, so does not match exactly, but the proportion with a university education has climbed from 39% in 2019 to 45% by 2024.
Against this backdrop, even a modest softening in hiring can generate a greater rise in graduate unemployment, without requiring a distinct AI-driven shock. Larger cohorts entering the market when vacancies are easing make graduate unemployment inherently more sensitive to cyclical conditions.
No sign of a tech-driven productivity increase
The underlying labour economics when technology replaces people shouldmean that we see productivity increases in the companies that make this change. But, as the Briefing puts it,
If jobs are being replaced, where’s the productivity surge?
All the same, it’s worth treading with some caution here, for a couple of reasons: productivity is notoriously difficult to measure, there are often lags before we see productivity growth in response to investment, and there seems to be some evidence elsewhere of a link between some sector level AI adoption and increases in unemployment.
If you are a company in one of thee sectors, you might release staff to free wage costs to fund AI-based innovation.
All the same, when you look at actual productivity growth figures, they have been sluggish for the past few years, with no real sign of improvement—as you’d expect in generally weak economies. (Again, Japan is a bit of an outlier).
No real sign of AI investment either
If we were going to believe the hypothesis that we were going to see AI-driven increases in productivity, but hadn’t seen it yet, we’d expect to see an increase in levels of investment in AI by firms at the moment, to create a platform for changes in the structuring of the business, new workflows, and so on. As it happens, there’s a rolling survey of US businesses that tracks this, which Oxford Economics includes in its Briefing.
The recent plateau – and in some cases reduction – in reported AI use could be consistent with initial experimentation by firms who then find AI doesn’t deliver promised gains. Although there’s plenty of anecdotal evidence of AI automating specific tasks and yielding large time savings, it’s also widely acknowledged that individuals typically overestimate the time savings from using AI. Some studies even find it can slow tasks down.
So this slowing could be the result of companies trying out AI and finding that it doesn’t produce expected efficiencies at the organisational level. In some cases, the’s system-level effect might be to wash out any gains. AI can help to process job applications, but it also makes writing compliant and competent applications easier, so it takes much longer to assess applications, for example.
And the kindest thing you can say about the corporate AI market right now is that it is sticky. The sceptical tech writer Will Lockett pointed out recently that Microsoft was having problems persuading clients to use Copilot; back in September Forbes reported that AI adoption rates at large companies were falling. These reports aren’t outliers.
So why might we believe that AI is causing employment among recent graduates? One reason might be correlation:
Since ChatGPT was publicly launched in November 2022, the US recent graduate unemployment rate has increased from 3.9%, close to its post-pandemic low, to a peak of 5.5% in March 2025.
But a stronger reason is because it is what companies have said as they have cut their hiring or made people redundant. Given the endless promotion of AI as being a marker of innovation and modernity by America’s big tech companies and their venture capitalist outriders, it’s not surprising that they say this.
The alternative—in which the CEO says, “our sales are flat and costs are going up, so we’re going to have to let some people go”—might even reflect poorly on the way the business is being run, and raise questions in the minds of investors.
But there’s a bigger question in this, I think. I suggested in a piece here last July (drawing on a couple of pieces by Nicolas Colin, and following the tech/finance innovation model of Carlota Perez) that the AI wave might be a marker of the end of the 50-60 year information and communications technology surge, not the beginning of a new wave of innovation.
The argument was that AI was about maximising the efficiency of the digital systems that business and government were now embedded in, in a similar way to how, in the ‘70s, at the end of the auto/oil boom, the invention of “logistics” rationalised and embedded vehicle systems at the heart of our economies.
Obviously having written that earlier piece I’m now primed for versions of this argument, but it was still interesting last week to notice that someone else was looking at AI through a pair of Carlota Perez glasses and coming to the same conclusion. Sameer Singh’s piece is headed, “AI: The Wrong Kind of Bubble”.
(“Both the Frenzy and Maturity stage often cause irrational, speculative behaviour — but for different reasons and with different results.” Source: Sameer Singh: breadcrumb.vc.)
Late-stage bubbles, in contrast to mid-cycle bubbles,
trigger overconsumption of the input that created the surge in the first place — leading to severe shortages, skyrocketing prices and geopolitical friction. This supply shock inevitably leads to abandoned infrastructure projects, sparking a collapse (or a series of collapses). The root of the collapse is a breakdown in the core assumption driving the technology surge — unlimited cheap inputs.
On skyrocketing prices, he has a telling chart looking at the recent surge in memory prices with the surge in oil prices in the 1970s.
(Source: Sameer Singh: breadcrumb.vc.)
There’s a lot more here, but for present purposes, the critical point is that AI use models as being constructed by Big Tech don’t align well with likely use cases. Singh has a view on this which suggests that successful AI models, when they arrive, will tend towards being more efficient (and cheaper), more open source, and designed for specific use cases.1
For the moment, in other words, there’s a big gap between the AI models that are good for Big Tech and its business interests, and the models that are good for their users. That gap will need to close, a lot, before we start to see meaningful productivity gains from AI. And so, for the moment, Oxford Economics’ take on graduate job losses seems to fit better with what’s going on.
j2t#658
If you are enjoying Just Two Things, please do send it on to a friend or colleague.
This sounds at least a bit like the Chinese approach to AI models, which also run on a lot less compute power.










Brilliant piece about water – and alarming.