"The numbers sing to me, Phil" said my boss of the time.
We were in a pizza joint in an everyday suburb of Warrington, the only place our diaries allowed us to meet so he could look at the budget I was about to submit to our corporate centre for my business unit. (Not sure I can even remember now which unit that would have been, my job changed every couple of years at that time). I passed over a single sheet of A4 populated with a table of numbers with the only words being the column and row headings for sales income, numbers of people, direct costs, overhead costs and contribution, the measure of gross proft we used before corporate overheads. There would hve been no footnotes or graphs as it would have been in a proforma required by our corporate finance team so I started to give him the one minute overview. He held his hand up to stop me, made his pronouncement and glanced over the page for about a minute taking in the patterns of the current year forecast, budget year proposal and two further years of probably rose-tinted projections for the business plan. The latter might not have been entirely for the birds but would certainly have contained some flights of fancy.
"Right" he said and asked a couple of fairly straight forward questions which I answered. Then our pizzas arrived, he nodded, said "ok" and put the page away in his briefcase. An item which alone marked us out as unusual that evening in a Stockton Heath pizzeria. We shot the breeze while we ate and then went on our separate ways.
When I got home Mrs H, already a bit peeved at me having to go out on a rare midweek evening when I hadn't been due to be travelling, asked how the meeting went. "Oh he only looked at it for a minute and said ok". I'm not sure I should have said that.... "What was the point of that?" "I guess he wanted to look me in the eyes, love".
So it isn't ever just about the numbers, though on occasions they do leap out from the page or, indeed, sing. Here are some singing numbers, though some of them shout rather than sing...
Subsidising rail fares is currently costing the taxpayer £8,000 a minute, which is £4.2bn annually.
None of that cost goes towards enhancing the railways, HS2, maintaining or upgrading tracks, tunnels or stations, which is all extra. It's just the difference between what's taken through the fare box and the cost of running the services. And yet, we have some of the highest rail fares in Europe. Also the subsidy is currently on an upward trend despite passenger numbers going up.
Rail fares have needed subsidy most of the time in the last three decades and I remember it was as high as £5bn 20 years ago when I first went into the rail sector. But in the three years up to 2018 net passenger operator support, which is what the subsidy is called, was negative: in that period £2.2bn went back to the exchequer, offsetting rail capital investment. Reasons (or excuses)? We have an old infrastructure, which had inadequate investment for decades when the railway was entirely in public ownership, which is what it's returning to. The power of the trade unions? That's always been a factor and, as usual Labour has bought off the train drivers, though pay has always been high in the rail sector, job for job compared with other transport sectors. And railways are just expensive, all nations subsidise them.
But here's another set of numbers to compare and contrast. Only 9% of journeys are by rail, 89% are by road. 11% commute to work by rail, about the same proportion as walking. 67% drive to work. The total taxpayer spend on our roads is only £14bn annually. The equivalent figure for rail is £11.8bn when grants to Notwork Rail are added to the fare subsidy. (I suspect the latter figure must exclude HS2 construction).
Oliver Gill asked if it was right that we send so much subsidising the railways. He noted that the less well off tend to use public transport and that some level of subsidy is essential to avoid pricing people off the railways (followed by demands for line closures, I imagine). Railways are a public "good". But just how much should be allocated? He concluded that a sensible debate was needed, starting with not renewing Labour's "cost of living" rail fare freeze in March 2027. Don't hold your breath.
AI investment funded by debt is worryingly huge with, it seems to me, some guarantee that useful and marketable functionality is coming from it but no guarantee that all the hype will be deliverable. Morgan Stanley estimate that global AI-related debt issuance this year totals about $450 bn (£340 bn), already more than double the total for all of 2025. They also estimate global debt-financed capital expenditure will total over $4 trillion between 2026 and 2030. That's a rather colossal write off if it proves never to deliver the expected returns. The UK exposure isn't tiny either: AI has accounted for nearly half (47%) of corporate bond issuance this year.
David Smith correctly concludes that the financial risk is much greater than than the bust associated with the the dotcom bubble in 2000 which hit markets significantly, wiping off $5 trillion in market value and causing a multi-year economic downturn. An AI crash would wipe off more than the investments made as values of companies like Nvidia (which makes the best chips and is offering AI start ups buy now, pay later deals), Microsoft, Alphabet (Google) and Meta (Facebook) are pumped up by them all predicting future growth in profits from the success of AI.
In the case of the 2008 financial crisis Smith points out that in 2007 there was a crazy amount of leverage on financial derivatives but it remained below the surface, hidden in instruments such as collateralised debt obligations, which most senior and central bankers had no idea about. But the AI danger is in plain sight.
Hmm. To put the numbers in some context the US economy is worth over $32 trillion annually and it represents 26% of the global economy. But share portfolios (and my pension) could take a very large multi-year hit if there's a "correction" as the shares listed above figure prominently in so many of the most popular investment funds. There are also companies in Asia and banks exposed if AI doesn't deliver the financial returns that have been built in to valuations.
I've also been wondering how we can supply all the proposed new AI data centres with electricity and water. Affinity Water supplies water to around 4 million homes in the home counties and northwest London. There are 88 new data centres planned for its area of which 40 have planning permission (maybe more now). Two are so large that, together, they will require the same amount of water as 49,000 houses.
Most of our data centres serving AI applications are operating in inference mode, which is what is happening when you run a Google query and it offers its AI mode response. The other key mode, learning, where the machine is fed massive datasets in order to recognise patterns, requires much more computational power so will have much higher demand for electricity and cooling water. Some people suggest that AI doesn't use the amount of power and water that's been reported which may be because they're unaware of the two functions. Unless we have data centres running in learning mode we aren't really a player in AI, we're just a "taker" as it were.
The much lauded tech deals Starmer signed with Trump in September 2025 were already being put on ice by December. It was reported that the £31bn had been put on ice due to our digital services tax and food safety rules. Eh - not to do with tech the second of those but that's a Trump negotiation for you. Indeed the Brits said the Yanks were just playing hardball. But it wouldn't surprise me if the costs of operating in the UK had also made the likes of Google and Microsoft get cold feet. Which reflects a classic structural issue in the UK energy market. After all, we have lots of windmills which, much of the time, sit idle or idling. Site the data centres in the right place to reduce transmisson costs (or avoid the need to install so much equipment), equip them with some huge batteries and (theoretically) bingo - couldn't you escape from charging them a rate based on the price of gas? I suspect this isn't just because Ed Miliband, our SoS for Energy Security and Net Zero until recently, wasn't up to the job. We could site the learning data centres anywhere where they can get the limited number of people they need, but the inference data centres need to be near to centres of population. At least if you want the AI to give you the answer to your query instantly, which we seem to.
Maybe we should ask AI how to run our railways efficiently and whether the colossal debts being run up producing it will ever be repaid. But it would probably "hallucinate" (i.e. make up) the answers. Still we could ask it to make up a catchy tune and sing them to us!
Sorry but rail fares must rise to avoid taxpayers spending billions more. Oliver Gill, Sunday Times, 11 October 2026
Unlike the last crisis, the danger is hiding in plain sight. David Smith, Sunday Times, 11 October 2026
The water demand for the largest data centres came from a similar source a while ago, but I don't have it to hand.
No comments:
Post a Comment