Is VC subsidising your AI costs – and what happens when it stops?

“I’m back to the drawing board because the budget I thought I would need is blown away already.” Those were the words of Uber’s chief technology officer, Praveen Neppalli Naga, speaking earlier this year about the company’s pivot to AI coding tools. If the CTO of one of the world’s most successful technology businesses is caught off guard by the true cost of AI, where might it leave the thousands of early-stage founders who have built much of their cost base around today’s artificially low prices? And, most importantly, what can you do to prepare for when costs increase?

Nearly two-thirds of UK organisations now use AI – 64% according to a recent AWS report, up from 52% last year and well ahead of the European average of 54%. And with demand at those levels it comes as no surprise to learn that AI-native startups captured 51% of all European tech funding in Q1 2026, making it Europe’s biggest AI quarter on record. In London alone, 80% of a record €6bn in tech investment flowed into AI. Investors are all in and so are entrepreneurs. But both should be aware that a correction is on the horizon. 

The AI-fuelled entrepreneurship boom also extends well beyond the VC-backed startup world. The UK saw a record number of tech incorporations in 2025, with new firms founded soaring by almost 50% in the past five years, according to an RSM UK analysis. A new generation of founders is driving it – people who have never needed a technical co-founder, a developer, or an agency. Tools like Anthropic’s Claude and Lovable have made it possible to vibe code – to describe what you want in plain English and have a working app or product materialise in hours. 

So whether it’s the solopreneur building on TikTok, the first-time founder launching a SaaS product from their bedroom, the experienced professional turning redundancy into a business, or the VC backed startup – all of them are operating on the same assumption: that AI makes it possible to run a lean, scalable business at minimal cost. The barrier to starting has never been lower. 

But what was once a murmur at the back of the room is getting harder to ignore: the pricing everyone has been building on cannot stay the same for much longer. The AI tools that feel affordable today are being sold at a loss, propped up by venture capital that cannot sustain those losses indefinitely. And when the correction comes, it is those who have become completely reliant on AI to run their businesses that will be the most exposed. 

The problem is, many of those businesses have not only gone all in, but they’ve also burned the boats. The adoption wave has prompted mass redundancies that may prove to have been deeply premature. British companies reported that AI had resulted in net job losses over the past twelve months, down 8% – the highest rate among leading economies including the US, Japan, Germany and Australia, according to a Morgan Stanley study. Roles have been eliminated, hiring decisions reversed and entire functions restructured – all on the assumption that AI is cheaper and more reliable than the people it replaced. But it is starting to become clear that is simply not the case.

Okay, so how much more expensive is AI going to get?
I think it’s helpful to understand why we’ve ended up in this price mismatch in the first place. And that’s because AI companies have not been pricing their tools to generate profit – they have been pricing them to capture market share, funded by investors who believe that dominance today translates to monopoly rents tomorrow. It is a classic land-grab playbook, and it has worked spectacularly well at driving adoption. But it has created an illusion: that the cost of compute is what entrepreneurs are currently paying for it.

The real numbers are staggering. McKinsey research projects that by 2030, data centres will require $6.7 trillion worldwide to keep pace with AI demand. To generate enough revenue to cover the trillions being poured into AI data centres, Gartner senior director analyst Will Sommer has said AI companies would need to reach close to $2 trillion per year in revenue by 2029. Based on current economics and a ten percent profit margin per token, Gartner calculates that token consumption would need to grow anywhere from 50,000 to 100,000 times its current rate by 2030. Essentially, for the economics to work, almost every business in operation would need to be running AI continuously, across every function, at a scale that simply does not exist yet.

Signs of strain amongst the generative AI giants are beginning to show. Anthropic has recently blocked third-party AI agents from its subscriptions and shifted to pay-as-you-go pricing. Gartner also predicts enterprise AI costs will rise at least 40% by 2027 and analysts are warning of 30-50% API price increases within 18 months. But there is little need to read between the lines when OpenAI’s own head of ChatGPT has openly described their current pricing as “accidental.” 

It is also worth noting that the cost advantage of AI over human labour is far narrower than the narrative suggests – in many cases, it does not exist at all. Bryan Catanzaro, vice president of applied deep learning at Nvidia, recently acknowledged that for his team, the cost of compute far exceeds the cost of employees. An MIT study from 2024 analysed the technical requirements of AI models performing jobs at a human level and found that AI automation would be economically viable in only 23% of roles where vision is a primary part of the work. In the remaining 77%, it was cheaper for humans to continue. For founders who have already burned the boats, that is a significant finding to sit with.

What founders need to face
For early-stage entrepreneurs specifically, the risk is acutely pronounced. Businesses have been priced, modelled, and pitched to investors on the basis of AI economics that, as we’ve discussed, will not hold. Product margins have been calculated against today’s API costs. Growth strategies have been built on the assumption that what you pay for compute this quarter is roughly what you will pay next year. If prices rise 40% by 2027 – let alone the 30-50% analysts are projecting in API costs within 18 months – those models collapse.

The most important question any founder can ask themselves right now is this: if the cost of my core AI tools doubled or tripled overnight, what would my business look like? If the answer is that it would fundamentally break – if margins disappear, pricing becomes uncompetitive, or the operational model stops working – then you are carrying a risk you have not yet accounted for. And you are not alone, most aren’t asking themselves this vital question either. Start with a simple audit, find out how exposed you are and adjust accordingly. 

This will also affect where capital is heading
Many of the founders we work with have also felt huge pressure to include an AI focus in their pitch deck, or even base their whole business model around it. But there are signs that the rotation away from overweighted AI exposure is already underway at the largest end of the market. EQT – one of Europe’s biggest private equity firms – has spent recent months systematically acquiring businesses with predictable, asset-heavy revenue streams including Yorkshire Water, Nord Anglia Education and Coller Capital. Its latest move is a third takeover bid for Intertek, the FTSE 100 testing and certification giant.

The pattern reflects what is becoming known as the HALO thesis – heavy assets, low obsolescence – a deliberate shift toward businesses that generate reliable cash without depending on a strong exit environment to deliver returns. According to Bain & Company, the PE industry is currently holding roughly 32,000 unsold companies worth $3.8 trillion, with average hold periods stretching to seven years. In that context, a business solving a real, durable problem starts to look a lot more attractive than one built on today’s AI pricing. The smart money is already moving.

For founders still in the ideation stage, this is worth taking note of – and it seems many already have. The headlines of today would have you believe that almost every company being founded is AI-first. But of the 485 businesses that recently applied to our Impact Investor Office Hours programme, just 5% fell into that category. The remaining 95% are building for problems that existed long before AI arrived. Health and wellbeing accounts for 30% of those applications, climate and sustainability 27%, and financial resilience 20%.

The common thread is that they are practical responses to systems that are visibly failing – fertility treatment that takes an average of 3.5 years to access on the NHS, food systems where more than half the calories in the average British diet now come from ultra-processed products, and financial services that still exclude millions. The founders building in these spaces are filling a gap that existed long before AI arrived and will exist long after the pricing correction comes.

This is not to say that you shouldn’t leverage AI at all if you are a founder, or those that stand to win are those who use it least. But it is worth interrogating whether you are using AI as a tool in service of something bigger, rather than the foundation which the entire business model rests on. Starting with a genuine problem, a real customer, and a durable need – and then asking how AI can help you solve it faster or cheaper – is a fundamentally different proposition to starting with AI capability and working backwards to find a use case. The former survives a repricing. The latter may not.

At Unrest, our focus has always been on impact above all else – we have long believed that the businesses creating genuine, durable value for society are the ones that withstand value shocks over time and prove most resilient when markets recalibrate. And we are starting to see that view reflected back to us in the conversations we are having with other investors.

This is not the end of AI – but there will be an in-between
So far this has read like a damning verdict for the future of enterprise AI. But I want to be clear that in no way do I believe the end is nigh. Innovation has a long history of righting economic imbalances over time, and there are reasons to believe it will do so here too. Gartner projects that the cost of inference – how AI analyses data – for a large language model with one trillion parameters will fall by more than 90% over the next four years. AI infrastructure will improve. Model design and hardware supply will follow. The economics of AI will eventually settle into something sustainable.

But between now and then, there is likely to be a period of real discomfort – a gap where the cost of compute and the drive for efficiency are in a constant battle with each other, where prices rise before they fall, and where businesses built on today’s subsidised economics face the consequences of not having planned for it. That period may also force a reckoning for those companies that moved fastest to replace human workers with AI tools: a situation where the cost advantage they banked on has eroded, forcing them into the costly and time-consuming process of rebuilding the human capability they previously cut.

The founders who come through it best will be the ones who treated AI as the brilliant tool that it is, but treated its rollout as a marathon to be paced rather than a sprint. At Unrest, we are encouraging our portfolio companies to pressure-test their cost model against scenarios where their AI bill is materially higher. To think carefully before eliminating human capacity that took years to build. And to ask themselves, with real honesty, what their business looks like if the tools they depend on most become significantly more expensive – or significantly less accessible – in the next eighteen months.

This is one of the most unpredictable technological advances in a generation. The entrepreneurs who thrive will be the ones who built for the uncertainty of it – not just the opportunity.