Where the AI Money Is Really Going (And Why It's Not ChatGPT)

If you have been watching AI headlines, you would be forgiven for thinking all the money flows to ChatGPT and Claude. But the week of 10 July 2026 told a different story. The huge cheques went not to the chatbots you already know, but to the startups building on top of them. Think of the big AI models as electricity: powerful, but not much use until someone builds a fridge, a lamp, or a laptop that plugs into it. This week, investors bet billions on the fridge-makers.

Four names led the week, and each one takes a general AI model and points it at a very specific job. Legora builds legal AI — software that helps lawyers read, draft, and check contracts far faster than doing it by hand. It raised a reported $550M at a $5.55B valuation, with what backers called nearly every top-tier venture firm on its cap table. Prime Intellect makes a platform for building AI 'agents' — programs that don't just answer questions but carry out multi-step tasks for you. It closed $130M at a $1B valuation and says it already earns over $100M a year.

The other two are about robots — but in a roundabout way. Rhoda AI stepped out of 18 months of secrecy with a reported $450M round and a system it calls FutureVision, which teaches machines to act by predicting what happens next in video (imagine a robot that has 'watched' enough footage to guess how to pick up a cup). Mecka AI, which raised $60M, does not build robots at all. It gathers and cleans up recordings of everyday human actions — chopping vegetables, folding laundry — so that robotics companies have good examples to learn from. It is the least glamorous job in the room, and possibly the most useful.

Here is the week at a glance. All figures are as reported by the companies and press, so treat valuations and revenue as their claims rather than audited facts:

The plain-English reason is that the base models are getting crowded and expensive. Only a handful of players — OpenAI, Anthropic, and xAI among them — can afford to train the giant models everyone else builds on, and their cheques dwarf everything else. So newer investors look one floor up, at companies that take those models and turn them into something a business will actually pay for. A law firm does not want 'an AI' — it wants a tool that reviews a lease in ten minutes. That gap, between raw capability and a finished, trustworthy tool, is exactly where these startups live.

There is a geography angle too, and it is worth knowing. Crunchbase reported that roughly 88% of 2026 AI-startup funding went to US-headquartered companies, and more than 80% of all global startup funding landed in the US. In practical terms, that means the tools reaching your desk over the next year will keep being built mostly by American firms, backed by American money — something to keep in mind if you care about where your data lives or which regulations apply.

OpenAI itself is making a related bet. Its new Deployment Company agreed to buy a firm called Northslope — its second applied-AI purchase since May — as part of a reported $4B push into 'forward-deployed engineering.' In plain English: OpenAI thinks that helping companies install and actually use AI is worth just as much as making the AI smarter. So expect the next wave of products you meet at work to feel less like a bare chatbot and more like a finished tool aimed at one job — with a team behind it to set it up. The takeaway for you: the AI you rely on is quietly moving from 'a clever thing you chat with' to 'software that quietly does the task,' and this week's funding is a preview of which of those tasks get built first.