On Wednesday, September 9, ten startups announced funding rounds. Together they raised roughly $1.47 billion.
Six of those companies took about $1.35 billion of it. The other four split what was left.
That split says more about the current market than any single deal does.
The headline round
The biggest was Harvey, a San Francisco company that builds AI software for lawyers. It raised $550 million at a $15.5 billion valuation. Diffusion and Lightspeed Venture Partners co-led the round. Sequoia, Kleiner Perkins, Andreessen Horowitz, Coatue and Goldman Sachs Alternatives all came back in.
Harvey’s software handles contract review, due diligence, compliance checks and litigation work. Reuters has reported that 80% of the 100 highest-grossing US law firms already use it.
The valuation curve is the striking part. Harvey was worth $8 billion in December 2025. Then $11 billion in March. Now $15.5 billion. That is a jump of roughly 41% in under six months.
The company also announced it had bought Guardrails AI, a startup working on agent security. That was Harvey’s fourth acquisition this year.
What investors are actually buying
A few years ago, the biggest cheques went to companies building the models themselves. Chips, computing power, foundation models.
That has shifted. Look at what else got funded on the same day.
- Clay raised $115 million at a $7.1 billion valuation. It makes sales software driven by AI agents. Thirteen months earlier it was worth $3.1 billion.
- Lightfield raised $47 million to build a customer database designed for AI agents instead of for humans typing into fields.
- Savvy Wealth raised $100 million to give independent financial advisers one system instead of twenty separate tools.
- Perry Weather raised $110 million for weather and safety software used by organisations running outdoor sites and crews.
- CloudNC, based in London, raised $20 million for software that speeds up CNC machining. Lockheed Martin’s venture arm joined the round.
Not one of those companies builds AI models. Every one of them sits inside a job somebody has to do anyway.
That is the pattern. Investors are paying up for companies that own a workflow, not companies that own a model.
Why the shift makes sense
The reasoning is simple enough.
Good models are getting cheaper and easier to reach. If your only advantage is calling a strong model through an API, so is everybody else’s.
But if your software becomes the place where legal work actually happens, or where sales data actually lives, something different builds up. Customer data piles on. Teams learn your buttons. Leaving starts to cost real money and real time.
Harvey shows how that works. Legal work needs an audit trail. It needs to match how a particular firm does things. That knowledge builds over years, and a new competitor cannot copy it by renting the same model.
Euno, which raised $23 million the same day, is chasing a version of the same idea. Its pitch is that an AI agent cannot act safely inside a company until it knows which internal definition of a number is the correct one. Most companies are sitting on decades of messy, undocumented data. Somebody has to sort that out.
The part founders should read twice
Now the uncomfortable half.
That 91% concentration is not a rounding error. It is the shape of the market. The very large cheques are going to a small number of companies investors already believe will win.
If you are raising a first or second round, the bar has moved. “We use AI” is not a pitch anymore. Investors have seen the demo.
What seems to work instead is being able to answer one question clearly: why does your position get stronger as models get better?
Harvey’s answer is distribution plus legal-specific knowledge. Lightfield’s answer is customer context that builds up over time. CloudNC’s answer is that machine shops are hard to reach and harder to switch.
If a better model would erase your product rather than improve it, that is worth knowing before you start pitching, not after.
The geography problem
One more thing worth naming, especially if you are building outside the US.
Eight of those ten companies are headquartered in the United States. Euno splits its operations between San Francisco and Tel Aviv. CloudNC is in London.
Nearly all the large money went to US-based companies. That is not because the engineering elsewhere is weaker. It is because that is where the enterprise budgets and the big funds sit.
So if you are building outside the US, selling into American enterprise accounts early is not a nice-to-have. Judging by where this money went, it is closer to a requirement.
What one day tells you
A single day of funding is not a trend. But it is a useful photograph.
The photograph shows legal work, cybersecurity, cancer drugs, sales, wealth management, weather, water systems and factory machining all raising money inside the same twelve hours.
The common thread is not the industry. It is that each company sits close to a decision that costs real money to get wrong.
Access to good AI is becoming ordinary. Owning a problem people cannot afford to leave broken is not.
