A decade ago, the working assumption in venture capital was simple: software eats the world. Pick a slow, paper-heavy industry, build a tool that helps the humans inside it work faster, sell it as a subscription, repeat. The product assisted. The human still did the job.
That assumption is breaking down in real time. Across 2025 and into 2026, AI has captured somewhere between half and three-fifths of all global venture capital, and the money isn’t simply chasing “AI features” bolted onto existing software. It’s rewarding something more specific: companies that don’t assist a workflow but perform it end-to-end, priced not against a software budget but against the labour or services budget they replace.
That’s a different business model wearing the same “tech startup” clothing, and it explains a lot of what looks, on the surface, like an indiscriminate AI gold rush.
From software-as-assistant to software-as-worker
The clearest evidence sits in legal AI, customer service, and coding. Harvey raised $200 million in March 2026 at an $11 billion valuation to run AI agents across due diligence, compliance and litigation support inside law firms — work that used to mean associates and billable hours. Cognition, maker of the autonomous coding agent Devin, raised $1 billion in May 2026 at a $26 billion valuation, with customers ranging from Mercedes-Benz to Goldman Sachs to NASA. Sierra and Decagon have done the same in customer service, both crossing or approaching unicorn-plus valuations on the strength of agents that resolve tickets rather than merely suggesting replies to a human agent.
None of this is proven at the level of durable economics yet — retention and pricing power over multiple years is still being tested. But the direction is unambiguous: capital is being priced against headcount, not against seats.
The capital cycle underneath the capital cycle
AI has not simply created a new venture theme. It has created a new capital cycle.
The first mistake is to ask whether AI is real or a bubble. It is both. Real technology waves attract speculative capital because the prize is large enough to justify imagination. The internet was real, but many internet investments failed. AI can transform the economy and still produce poor returns for investors who fund the wrong layer, at the wrong price, with the wrong assumptions about supply.
The capital-cycle question that has shaped every prior boom remains unchanged here: what capacity is being built, how fast is it arriving, and who loses pricing power when it does? In AI, capacity no longer means only factories, ships or mines. It means GPUs, data centres, power, frontier-model capability, cost per token, open-source substitutes, funded competitors, enterprise pilots, workflow data, distribution, talent density and governance permissions. Each layer has its own cycle, and they are not moving in sync. Frontier models are moving toward oligopoly formation. Compute, chips, data centres and power are closer to a classic capex cycle. Generic copilots and thin wrappers already risk oversupply. Vertical workflow AI remains more attractive where it owns process, data and distribution. AI governance and control may become a durable infrastructure layer because regulated enterprises cannot let autonomous systems act without permissioning, audit and enforcement.
The reframed question
The old question — does this company use AI? — has stopped being useful. The more useful question now is whether a company has become the place where the work actually gets done, where AI activity is governed, where an industry’s workflow is owned end-to-end, where strategic capacity is being built, or where capital itself is being structured differently. Companies that hold one of those positions — a model, a workflow, a governance layer, a distribution channel, or physical infrastructure like power and compute — sit in a genuinely defensible spot. Companies that occupy the thin layer between someone else’s platform and someone else’s customer do not, no matter how fluent their AI feature is.
That’s the lens worth applying to the next eighteen months of AI venture activity — not whether a company has adopted AI, but where it sits in this new map of control points.
The quiet category nobody coordinated on
Here’s the more interesting signal, and the one closest to home for anyone advising regulated industries: at least seven unrelated venture funds, spanning early-stage and growth investors, have independently backed AI governance and security companies in the same narrow window, without any sign of coordinating with each other. Noma Security, WitnessAI, Geordie AI and others have all raised meaningful rounds built around one underwriting logic: AI cannot be allowed to act freely inside a bank, an insurer or a hospital system unless it can be governed, audited and permissioned.
This category is still small in absolute dollar terms next to coding agents or foundation models. But independent, uncoordinated conviction from seven-plus investors is a stronger signal than any single fund’s thesis — it suggests the market has quietly decided that AI governance is its own permanent layer of infrastructure, the way cybersecurity became permanent infrastructure around the last cloud computing wave. Anyone advising regulated-enterprise platforms is watching this category for exactly that reason.
Owning the workflow, not renting it
Healthcare offers the cleanest proof point for a related idea: vertical depth beats horizontal breadth. Ambience Healthcare raised $243 million in mid-2025 to own clinical documentation across health systems, and roughly 54% of the $14.2 billion raised by digital health startups in 2025 went to AI-led companies — a sharp pivot away from the old consumer-wellness story toward enterprise and clinical workflow ownership. The same pattern is showing up in legal, fintech onboarding, asset management and logistics: platforms that own a specific regulated industry’s workflow and its underlying data are outcompeting generic productivity tools that any incumbent could absorb as a feature next quarter.
Sovereignty, power and the new defence economy
Defence and resilience technology has moved from the margins of venture capital to its centre. Shield AI raised $2 billion in March 2026 at a $12.7 billion valuation. Helsing raised €600 million toward European technological sovereignty. Founders Fund’s bet on Anduril is now the largest single cheque in that firm’s 21-year history. China is running a parallel, equally aggressive bet on domestic chip and compute sovereignty. Read together, these aren’t separate stories about defence-tech enthusiasm — they’re the same underlying theme expressed in different geographies: control over compute, power and strategic capacity is becoming as investable as any application built on top of it.
Even the funds are restructuring
It isn’t just the portfolio companies that are changing shape — the funds themselves are too. Founders Fund closed two growth funds, $4.6 billion and then $6 billion, within about twelve months of each other. Trian Fund Management and General Catalyst’s joint $7.4 billion take-private of Janus Henderson and Sequoia’s $7 billion late-stage fund both point toward permanent-capital, evergreen structures displacing the traditional ten-year venture model at the top of the market. Capital itself is being deployed differently, not just into different things.
What the money actually looks like, fund by fund
Step back from the headline narrative and a more disciplined picture emerges. Funds are barbelling: writing lots of small, exploratory first cheques on one end and a handful of enormous, concentrated follow-on rounds on the other, while the traditional “normal” growth round — somewhere in the $150–500 million range — has noticeably thinned out. A small clique of marquee funds — Sequoia, Founders Fund, General Catalyst, a16z, Lightspeed, Kleiner Perkins, Coatue — keep showing up together in the same handful of trophy companies, behaving less like venture investors and more like concentrated index holders of three or four AI winners everyone already agrees on. A separate set of funds shows almost no overlap with that group at all, doing genuinely uncorrelated, original discovery work instead. And in coding agents specifically, Founders Fund backed one rival (Cognition/Devin) while a16z backed another (Anysphere/Cursor) — a clean example of funds hedging across competing bets rather than picking a single winner, which is happening in foundation models too.
Encouragingly, the exit window also looks like it’s reopening — not in one sector, but across healthcare, insurtech, security and AI tooling — suggesting this build-out is starting to convert into realised outcomes rather than paper valuations.
The geography of conviction
The United States remains the deepest pool by far, with AI now representing close to 88% of all US venture capital so far in 2026. Europe shows genuine formation strength — Mistral’s €1.7 billion raise, Helsing’s sovereignty round — but a real capital-depth gap, with AI at just over half of European VC, the lowest share of any major region. India is showing disciplined, directional platform formation, helped along by roughly $48 billion in planned Amazon investment into India’s AI and cloud infrastructure. China has re-accelerated sharply — up sharply year-on-year into space, quantum, fusion and AI — though mainstream financial press has flagged real valuation-bubble risk alongside that acceleration, a caveat worth carrying into any read of the region.
What this doesn’t prove yet
It’s worth being honest about the edges of the evidence. Capital flows tell you where conviction is forming, not whether the underlying businesses will earn durable margins — legal AI, customer-support agents and AI governance all have strong funding momentum but thin, early evidence on retention and pricing power. Climate-as-a-category has effectively been repositioned into infrastructure and power economics rather than disappearing. And Europe and India remain comparatively under-documented in the public capital-flow data, more a gap in what’s visible than evidence against the thesis in either region.
Who actually keeps the returns
That capital-cycle logic is also why the AI shift is harder for existing software companies than investors think. Old SaaS companies are built to sell high-margin subscriptions. But AI agents may change the game from “pay for software” to “pay for work done.” If incumbents try too hard to protect their old margins, they may keep the financial model but lose control of the customer’s workflow. The winners will be companies willing to change the product, the pricing and the cost structure at the same time — even if that means lower margins for a while.
The best AI investments will sit where the product cycle is still early but the capital cycle is not yet euphoric. The danger zone is where both the story and the funding market are hot, but the company owns no durable control point. In the early phase, access feels like edge because the scarce asset is proximity to credible AI companies. Once capital floods in, access becomes commoditised. The real edge shifts to refusal: knowing which hot rounds not to chase, which technical claims are not yet moats, which companies are merely renting distribution from OpenAI, Nvidia or cloud platforms, and which businesses are actually converting capital into proprietary data, workflow control, switching costs or infrastructure advantage. The simplest test is whether each additional dollar of capital deepens the moat or merely keeps the company in the race. AI has eaten venture capital; whether venture capital keeps the returns now depends on getting that question right.