AI took three quarters of UK venture funding this year. Its margin problem is still unsolved.
Tech Funding News Gavin Lester ● Covered by 6 sources
UK AI startups grabbed $12.6bn in H1 2026, three-quarters of all British VC money. But their margins still trail software by 25+ points, and the compute bill keeps growing, not shrinking, as they scale.
Based on reporting by Tech Funding News, Gavin Lester — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Britain's AI companies just pulled off the best six months anyone can remember. HSBC Innovation Banking and Dealroom put UK AI funding at $12.6bn for the first half of 2026, more than four times last year's figure and roughly 73% of all UK venture capital. Total UK venture hit $17bn, the strongest start to a year since 2022. Impressive numbers. But they're being poured into businesses whose cost structure looks nothing like the SaaS companies that trained investors' instincts for the last two decades.
The uncomfortable number comes from ICONIQ's survey of around 300 AI executives: average gross margins for AI products sit near 52% in 2026, up from 41% in 2024. Progress, sure. Still 25 to 30 points below the 75-85% software has delivered for years. And the distribution is worse than the average suggests — Bessemer found the fastest-scaling companies, the ones hitting $100m ARR in eighteen months, running at just 25% gross margins. They're effectively buying growth with GPU spend.
Here's the part that should unsettle anyone valuing these companies like software firms. ICONIQ's data shows inference costs rising from 20% to 23% of total spend as products mature, while talent costs shrink. That's backwards from how SaaS worked. In software, cost of goods sold was basically fixed, so margins expanded with scale. In AI, the dominant cost is per-query, and it doesn't amortise away. A company losing money at 100 customers doesn't fix itself at 1,000 — it just loses more, faster.
What's changing is the calculus around who should own their own hardware. Most companies, most of the time, should keep renting compute — paying for elasticity while the product is still shifting weekly is the right call. But once usage is sustained and predictable, running your own GPUs starts to make financial sense, and a growing number of UK AI firms are crossing that line. The one company with genuinely public numbers on self-hosting, 37signals, isn't even an AI business — it spent $3.2m on cloud in 2022, moved to owned hardware for about $700,000, and expects to save over $10m across five years. The principle transfers even if the workload doesn't: once the bill gets large, you owe it to yourself to actually run the maths.
Britain adds a wrinkle nobody in the US has to deal with. The grid connection queue for new capacity sits at 125GW against 45GW of peak national demand, transformer lead times run two to four years, and data centre vacancy has collapsed from 27% in 2016 to 8% now. So the decision to self-host can't be made when the API bill finally hurts — by then, the power and space aren't there. The analysis and the procurement have to happen at the same time, or not at all.
My take — AI-written commentary, not fact-checked reporting
Nobody wants to hear that three-quarters of a country's venture capital is chasing a business model that still doesn't have working unit economics, but that's exactly what's happening, and pretending otherwise just delays the reckoning. Falling inference costs might bail out plenty of these companies before anyone notices, which is the best argument for staying calm — but betting a portfolio on a price curve continuing forever is not a strategy, it's a hope. The investors still applying 2021 SaaS multiples to 2026 compute-heavy businesses are the ones who'll be surprised twice: once when growth slows, and again when they realize margin never was optional.
Read more about this at: Tech Funding News
