A few weeks ago, one of my founder friends sent me a screenshot of his monthly token bill. £18,000. For a team of three.

He wasn't complaining. He was bragging. That £18,000 had replaced what would have been at least two senior engineers, six months of work, and a product launch that probably wouldn't have happened at all without AI doing the heavy lifting. Every founder reading this knows exactly what that feels like - the slightly dizzying realisation that your tiny team is now operating at a scale that would have been impossible eighteen months ago.

I've been keeping a close eye on what's actually happening in the industry over the past few weeks - the deals, the funding rounds, the regulatory shifts - and pulling out the things that matter, so you don't have to spend your evenings doing it. What struck me wasn't any one story. It was the sheer scale of what's being committed, consolidated, and rewritten - and how little of it is being driven by anyone in Europe.

Let me walk you through what I found, because I think it changes how we should be thinking about the next twelve months.

See below to continue…

The founders who close rounds fastest aren't just the ones with the best pitch; they're the ones who show up prepared.

Carta Launch gives early-stage founders a clean, investor-ready cap table for free, so when the conversation gets serious, you're already serious.

Manage equity, issue shares, and track dilution, then instantly share it with a built-in data room that gives investors exactly what they need, exactly when they ask. 

The best time to get your cap table right is before you need to.

Access for FREE here.

Continue…

For the last two years, the working theory for most B2B SaaS founders has been: build a thin layer of product on top of a frontier model, charge for the workflow, and ride the wave. That theory is now in serious trouble.

Anthropic has spent the past few months hiring senior product and engineering talent from across the enterprise software world - including Workday's CTO Peter Bailis, who left a C-suite role at one of the biggest HR software companies in the world to join Anthropic as a member of technical staff. The competitive irony, as TechCrunch's coverage points out, is that Workday's CEO had publicly said Anthropic was a Workday customer. The customer is now hiring the supplier's CTO to build competing products.

This isn't a one-off. Anthropic has launched a marketplace for Claude-powered enterprise software, committed $100 million to a Claude Partner Network with Accenture, Deloitte, Cognizant and Infosys, and is reportedly building HR-focused "people products". The model maker is now also the application maker, the marketplace, and the consulting partner.

Meanwhile, SpaceX has just struck a deal with Cursor giving it the right to acquire the AI coding company for $60 billion later this year - or pay $10 billion just for the privilege of working together. xAI needed a coding front door to distribute its model, and Cursor was it. OpenAI has Codex. Anthropic has Claude Code. xAI now has Cursor.

What this means for the rest of us: if you're building a product whose core value is "we make GPT/Claude easier to use for X", your moat is shrinking by the week. The labs have decided they don't want to be infrastructure. They want to be everything.

Anthropic is reportedly weighing offers that would value it at over $900 billion, according to Bloomberg - potentially overtaking OpenAI as the world's most valuable AI startup. This is just months after closing a $30 billion round at a $380 billion valuation in February. So the company has more than doubled in value in the space of a quarter, and its annual revenue run rate has gone from roughly $9 billion at the end of 2025 to over $30 billion by March.

Read those numbers again, because they're genuinely difficult to process. A 1,400% year-over-year revenue increase. A valuation that has more than doubled in ninety days. Investors so desperate to get in that, according to TechCrunch, one institutional investor prepared to commit $5 billion couldn't even get a meeting with Anthropic's CFO.

For founders, two things flow from this. First, there is more capital chasing AI right now than at any point in tech history - which means the bar for what counts as "fundable" in adjacent spaces is moving fast. Second, the gravitational pull of these giant rounds is sucking talent, attention, and partnerships toward a tiny number of US companies. If you're building in Europe, you're building inside that gravitational field whether you like it or not.

The number that genuinely stopped me was this: combined 2026 AI capital expenditure across Microsoft, Alphabet, Meta, and Amazon is now tracking somewhere between $650 and $700 billion. Six hundred and fifty billion dollars. In a single year. From four companies.

To put that in context, it's larger than the GDP of most European countries.

And the Q1 earnings just confirmed that the spending isn't slowing down - it's accelerating. Google Cloud grew 63% year-over-year. AWS grew 28%, its fastest pace in fifteen quarters. Microsoft's AI business is now running at a $37 billion annualised revenue rate. Sundar Pichai literally told analysts on the earnings call that "we are compute constrained in the near term" and that cloud revenue would have been higher if Google could meet demand.

Meta got punished by the market - its shares dropped about 6% - for raising its 2026 capex guide to $125-$145 billion. Not because investors think the spending is wrong, but because Meta's AI investment is internal (powering its own ads and recommendation systems) rather than producing visible third-party cloud revenue. The market is now sophisticated enough to distinguish between capex that generates external receipts and capex that doesn't.

For founders, the practical implication is uncomfortable but useful: compute is the constraint. Memory chips are sold out. GPU spot prices have shot up. If your roadmap depends on cheap, abundant inference, you need to model the scenario where that doesn't materialise on the timeline you assumed.

This is the part that genuinely unsettled me.

While the US labs are spending hundreds of billions building the future stack, the EU's attempt to soften its own AI Act collapsed last week. Twelve hours of trilogue negotiations on the Digital Omnibus ended without agreement on 28 April. That means, unless the next round on 13 May produces a deal, the original AI Act high-risk obligations come into force on 2 August 2026 - exactly as written, with none of the postponements industry had been planning around.

Many companies have been quietly recalibrating their compliance roadmaps to assume the December 2027 extension would land. That assumption is now exposed. If you're operating an AI product in the EU that touches anything in Annex III - recruitment, credit decisions, education, essential services - you need to be planning against the August deadline, not the one you'd hoped for.

On the more positive side, the UK launched a £500 million Sovereign AI Fund on 16 April, with Technology Secretary Liz Kendall pitching it as a way to keep British AI companies "starting up, scaling up and staying" in Britain. The first equity stake went to Callosum, an AI infrastructure startup, with six other companies - including Prima Mente, Cosine, Cursive, Doubleword, Twig Bio and Odyssey - getting access to the UK's AI Research Resource supercomputer network.

It's something. But £500 million sits next to France's €109 billion AI package and the US's $650 billion in private hyperscaler capex, and the proportions speak for themselves.

If you're a founder reading this trying to work out what to actually do with this information, here's what I'm taking away.

The infrastructure layer is closed. You are not going to outspend Microsoft on data centres. You are not going to out-research Anthropic. The frontier model game is over for everyone outside a handful of labs, and pretending otherwise is just expensive cosplay.

The application layer is open but narrowing. The labs are coming for the obvious horizontal categories - coding, design, HR, productivity. Building "AI for generic enterprise function" is increasingly building directly into their roadmap. The defensible plays are vertical specialisation, deep workflow integration, regulated industries, and the kinds of unsexy domain knowledge that frontier labs don't have. Mistral has quietly been doing exactly this in Europe - positioning itself as the AI integrator for European enterprise rather than chasing a head-to-head model race it can't win.

Compliance is a moat for those who take it seriously. The August AI Act deadline is now real. Most founders are wishing it away. The ones who actually build for it will have something to point to when their bigger US competitors arrive in Europe and realise they have six months of work to do.

And the £18,000 token bill on my friend's screen? That's not a sign of overspending. That's a sign that the cost of building has shifted from labour to compute, and that the founders who figure out how to deploy that compute against real, narrow, unglamorous problems are the ones who'll still be standing when the dust settles.

The trillion-dollar trade happening above our heads isn't really about AI. It's about who controls the next economic stack. We don't get to play in that game. But we do get to use the rails they're building - and that's still the best opportunity any of us are likely to see in our careers.

✅ Know a founder trying to make sense of where AI is actually heading? Forward this their way. The clearer the picture they have of what's happening above the application layer, the better the bets they'll place this year.

(👉 Vote now — we’ll share the results in next week’s issue. All votes are anonymous.)

(Last week’s poll results)

All the UK’s best startup events in one place.

Access here and please share it with others.