AI Didn't Kill Venture. It Killed the Ten-Year Bet.

General Catalyst committed a $1 billion investment and received zero equity in return.

When I first saw the announcement I was confused, because I had always believed General Catalyst to be a T1 VC and I was shocked that they chose to not receive equity.

IM8, co-founded by David Beckham in 2024, operates under a Nasdaq listed company called Prenetics, and it passed $100M revenue in eleven months, so it is now operating at around $200M in revenue a year. It's clearly a successful business, but it sells powders in sachets and has nothing to do with tech, and one of the largest and most successful tech VCs in the world just handed it $1B without taking any equity.

I had to dive in deeper after seeing the headline, because none of this made sense to me. I thought I had a good grasp of venture, and of the many "games" played by VC firms to generate returns, even if I did not think all of the decisions were smart.

The first thing I uncovered was the structure of the deal. IM8 spends money on advertising to bring in new customers, and General Catalyst agreed to pay up to 70% of that advertising bill each month. In exchange, it takes a slice of what those specific customers go on to spend, and it keeps taking that slice until it has been paid back an agreed amount. Once it hits that number it stops, and every pound those customers spend afterwards belongs entirely to IM8.

Now I understood the mechanics, and I wanted to understand the decision. General Catalyst is a firm whose whole existence is built on investing in tech companies to own a slice of the pie and waiting for that slice to become worth more, but here it was choosing to own nothing at all, in a business that had nothing to do with tech beyond selling powder in sachets.

My curiosity was piqued, and I needed to know why a firm like that would walk away from the one thing it is supposed to want.

So what is venture normally, and why is this so strange?

Let me take a step back, because none of this seems weird unless you understand the venture space.

A venture fund is in essence a team with a pot of money. Someone goes to pension funds, universities and wealthy families, collects a few hundred million pounds, and promises to invest it in young companies. The people who put money in are called limited partners, the people managing the pot are called general partners, and the arrangement between them is where everything interesting hides.

Typically, the arrangement says the pot has a life of ten years, so at the end of that decade you count what the investments turned into, you hand the money back, and the managers keep a slice of the profit as their reward. That slice is the reason anyone does this job, and it only gets paid when the counting happens. On top of that, venture firms take a piece of the funds raised to run the fund, with the most common model being 2 and 20, meaning 2% of the fund is used for operations annually and 20% of the profits are kept by the venture firm.

Of course this is an oversimplification and more semantics are involved, but it paints the picture of the general structure.

I had to dive deeper, and my first question was why ten years rather than eight, or fifteen, or whenever the companies happen to be ready.

As with many things in life, I found the answer was history repeating rather than anything logically grounded. The first venture firm, founded in 1946, was set up under a law that stopped it from paying its own staff in shares of the companies they backed, and one of its people, Charles Waite, rescued a struggling business and took it public, for which he received a $2,000 pay rise while the company's boss walked away $10m richer. That was unsustainable, so the industry redesigned itself into a structure that let the managers take a share of the profits directly.

But if you are going to pay someone a share of the profits, you have to agree when you count them, and someone picked ten years, and everyone has copied it ever since without going back to check.

So the ten-year clock was never about how long it takes to build a company; it was a timeline set for when the manager gets paid, and every rule that flows from it, including the ones founders are lectured about in meetings, comes from a payroll problem someone solved in the 1950s.

What are the implications of the 10-year clock?

Once the date is fixed, every decision is based on arithmetic rather than judgement, and it is worth walking through because it explains behaviour that otherwise looks mad.

I had to dive deeper and started to look at the returns of venture firms, and I found a study of over 11,000 venture-backed companies which stated that around 60% never gave their LPs back all of the money they put in, and 27% went to zero completely. Of the remaining 13% of funds that did well, 90% had invested into a unicorn.

So picture what a VC is really buying, which is a bag of a hundred lottery tickets where ninety-nine of them are worth nothing much and one ticket generates the returns. That is not a risk VCs manage, it is the entire VC business model, and it means a fund manager cannot be interested in a company that might reasonably become nice rather than enormous.

This is why VCs ask the market size question in the first meeting. The investor is not being nosy about your ambition, they are checking whether you are mathematically capable of being their one ticket, because if you are not, then a good outcome for you is a rounding error for them. And the bigger the fund gets, the bigger that required outcome becomes, which quietly shrinks the number of companies they are allowed to find interesting.

Has venture stopped working?

My curiosity led me to this question next, and I was expecting a nuanced answer, but the reality was shocking even to me, an extreme sceptic.

Cambridge Associates tracks the industry's performance, and over a twenty-five year period from 2000 to 2025 American venture capital returned 6.9% a year, while over the same stretch simply buying the S&P 500 returned 9.1%, the Nasdaq returned 9.8%, and even small company stocks returned 8.0%.

Yes, you read that right, and VC was outperformed by the most basic of indices. You could have bought an index fund on your phone, paid almost nothing in fees, been able to sell any day you liked, and done better than the people locking your money away for a decade and charging you handsomely for it, because venture lost to large companies, small companies and technology companies since the data has been tracked.

The reason this data shocked me was because I had seen statistics before and they were not this alarming, and what I uncovered is that the venture industry had an answer to this narrative problem, because without funding it's an industry that doesn't exist. Venture firms decided to use data from 2003 instead, after the dot-com crash, which is a bit like reporting your gambling record from the moment you started winning.

There is a second number in the same report that almost nobody talks about, and I think it is the more revealing one. In that half-year, buyout funds, the ones that buy established companies, gave their investors more money back than they asked for, whereas venture funds did the opposite, asking for $26.9bn and returning $16.1bn, and since 2022 venture has called in roughly 1.6 times more money than it has handed back.

So when you see articles about private markets having a liquidity problem, there's more than meets the eye, because private markets are mostly fine and venture specifically is the part that has stopped giving money back, and the two get bundled together in a way that flatters one of them.

Was AI the killer, or did it revive venture?

What fascinated me was the 2022 date, because it's no secret that in 2022 money stopped being free, with macroeconomic factors, mainly interest rates, hitting global liquidity, but 2022 is also around the time the rise of AI commenced with ChatGPT launching late that year.

Most people say AI changed what companies need and made it cheaper to run a business, and whilst this may be true, I think it did something narrower and more damaging than that, because it broke the ability to see ten years ahead, and seeing ten years ahead is the only thing venture was ever selling.

In the past I could easily spew ideas about where I saw the world in ten years, and now I find it almost impossible, which I realised whilst having a conversation with my co-founder Alex, who asked me where I see the world in ten years.

Universal Basic Income? Sure, but this is basically communism, and communism doesn't work.

Elysium? Maybe, but will there then be an uprising, and if said uprising occurs, will the robots and drones mean that for once the people with power will win and maintain it?

Will AI die because it's too expensive? Possibly, but has there ever been such a disruptive technology which didn't garner mass adoption? Not that I'm aware of.

Taking a step back once again, ask yourself an easy question, which is what you will eat tomorrow, and you can probably answer that with reasonable confidence. Now ask yourself what your favourite piece of software will be in ten years, and you have no idea, and neither does anybody else.

Strip everything else away and venture rests on one promise, and that promise rests on the second question: that a small number of people can look at a young company and correctly guess what it will be worth in a decade. Not that they can pick good companies, because plenty of people can do that, but that they can pick them a decade early, hold on through everything in between, and be right about a world that has not happened yet.

It's rarely a bet on the company itself when you look closely, but a bet on the whole category the company sits in and on the founders behind it, and they are asking one question: will this company still exist and still be worth something in a decade?

Three things happened at once to make that bet harder.

1. The ground moves between funding rounds now. A company can be built on something the machines cannot do yet, and then a new model arrives that does it, and the company's reason to exist quietly evaporates, so it was not beaten by a competitor anyone had on a list, it was beaten by a software update.

2. Cheap building cuts both ways. If you can build your product in three weeks, so can the four people who saw your launch, and the same tools that got you started are handing everyone else the same head start.

3. The numbers show it, which is what makes it hard to argue with. Look at how much money a company keeps from its existing customers a year later, and for traditional business software that figure sits around 82%, whereas for AI-native companies it runs closer to 48%, so half the revenue walks out of the door, and that small, efficient team everyone admires is partly a sign that the product is easy to leave.

Meanwhile the fund still has to hold its investments for ten years, and the window for predicting an AI company's position is closer to eighteen months.

There is a fair objection here, because it is the first thing a good investor would say back to me. If the future is less predictable, surely the lottery ticket is worth more rather than less, since a divisive future produces bigger winners? That is true about the companies and false about the funds, because a lottery ticket only pays out if you are still holding it when the draw happens, and a ten-year fund cannot hold a fifteen-year ticket.

And if I was to be facetious I would argue that IPOs, dividends and OTC sales are now more possible than ever, which does add flexibility to the exit, but are these really a solution, or do we need a few more terrible IPOs like SpaceX and major firms like Anthropic to warn against secondary trading, before the house of cards falls?

Either way, more ways out does not fix the thing I keep circling back to, because an exit only helps if somebody can put a price on what is being sold, and pricing is precisely the bit that is broken.

Venture is turning into private equity

Which brings me back to where the money has gone, and once you see the pattern it is difficult to unsee.

VCs have not stopped investing, they have changed what they are willing to be right about, because guessing what a company will be worth in ten years is near impossible, so they have moved towards things where the answer already exists: customers who have already bought something, contracts that have already been signed, businesses that already make a profit you can measure this quarter.

That is not venture capital.

That is what private equity firms and lenders have always done, which is why it feels so strange watching venture firms do it, and why my first reaction to the IM8 headline was that somebody had made a mistake writing it up.

There is a second reason too, and it is duller, but I suspect it matters more. Insurance companies and pension schemes sit on enormous amounts of money and have strict rules about matching that money against payments they owe in the future, so they cannot buy into something that might pay out in year nine, or year fourteen, or never, whereas if you give them a loan that pays a set amount over a set period they can buy it comfortably. So part of what looks like a change of investment philosophy is really a change of customer, because the pool of money available to sell to has moved.

They have swapped a question nobody can answer for one they can, and whether the new answer is any good is a separate matter, but it turns out there is a market that prices exactly that, every day, in public.

What the credit market thinks of the AI story

If AI has genuinely made these businesses better, there is one place to check without listening to anybody's opinion, and that is the price lenders charge, because lenders have their own money at risk, no shares to talk up, and no reason whatsoever to be generous.

On 9 July 2026, five days before the Beckham announcement, S&P cut Oracle's credit rating to BBB-, which is the lowest rung of the respectable ladder, one step above the category politely called speculative and impolitely called junk.

The reason was that S&P raised its estimate of Oracle's spending for the coming year from $60bn to somewhere between $90bn and $95bn, and now expects the company to burn through $42bn more than it takes in, but the larger worry was concentration, because roughly half of the future revenue Oracle has booked depends on a single customer, OpenAI, and Moody's has Oracle on a negative outlook as well.

So look at the shape of that, because Oracle is building enormous facilities, borrowing heavily to do it, and the demand holding the whole thing up traces back to one buyer who is themselves losing money at a remarkable rate, and the people whose job is to assess whether they will be repaid looked at that arrangement and marked the company down.

You see the same thing in what lenders charge against AI hardware, because the same chips, financed by the same company, have been priced across a range of roughly 1,275 basis points in under three years, and the thing moving that price was never the technology, it was who had signed the contract to use them.

So the verdict is already in, and it is blunt, because nobody is lending against the technology, they are lending against whoever promised to pay, and the AI story itself is worth precisely nothing to them.

Which means a good deal of what looks like an AI boom is an arrangement of debts, where money is borrowed to build the buildings, contracts are signed to justify the borrowing, and investments flow back and forth between the same handful of companies in a way that makes everyone's numbers look busier than the underlying activity warrants. That is financial engineering, and financial engineering works beautifully right up until the moment somebody near the middle of it cannot pay.

Is AI the symptom or the cause?

Both, and separating the two is the reason I bothered to write this article.

Venture was already ill, not quite terminally but on the verge, because the returns had not justified the lock-up for years, the money had stopped coming back well before anyone was talking about large language models, and the ten-year clock had been quietly failing since long before ChatGPT.

What AI did was two separate things that keep getting muddled together, in that it made the ten-year guess genuinely impossible, which is a real cause rather than a narrative, and it arrived with such force that every retreat from that guess could be dressed up as a bold move into the future, which is the part I would be sceptical about. Buying an accounting firm becomes an AI transformation, lending against an advertising budget becomes a new asset class, and refusing to make ten-year bets becomes leaning into a technological shift.

I should be fair to the other side, because this argument has been made before and it was wrong. In 2006 cloud computing collapsed the cost of starting a software company, in much the same way people now say AI has, and plenty of people predicted it would kill venture capital, but it did not, and instead it produced more funds writing smaller cheques.

What differs now is the direction of travel, because the cost of building has collapsed again, but the number of funds has shrunk to its lowest level since 2017 and a single firm took close to a fifth of all American venture money, so where cheap building used to open the door to more people, this time it is closing it.

So no, venture is not dying exactly, because the fee has come apart from the job and the firms have followed the fee, and what is dying is the specific promise that you can buy a claim on an unknowable future and be paid for the waiting.

Which brings me back to the sachets. General Catalyst did not lose its mind, and it did not suddenly develop a passion for nutritional supplements, it found a business where the thing it needed to know was knowable, where customers had already bought, where the money came back on a schedule rather than on hope, and where nobody had to be right about 2036. The billion dollars was not a bet on IM8 becoming enormous, it was a bet that people who bought powder last month will buy powder again next month, which is the sort of question you can answer.

Once I understood that, the headline stopped being strange, and what worries me is that it is not strange, because a firm that could no longer price a decade went and found something it could price to the month, and that says more about the decade than it does about the month.