AI capital kept stacking up last week, and the top of the board was almost entirely about agents, training data and compute. Mirendil and Mercor set the pace, but the deals further down show where investors are placing their bets. That's the more useful read. Headline rounds grab attention, while the middle of the list shows which categories have moved from experiment to budget line.
Agents and human expertise at the top
Mirendil raised $1 billion in a Series A, the largest U.S. round of the week. The San Francisco startup builds agentic AI for research and development, and Kleiner Perkins and Clearlake Capital led the round. A billion-dollar Series A is rare enough that it's worth watching how Mirendil deploys the money. Early-stage rounds are usually sized to get a company to its next proof point, so a check this large suggests investors are underwriting something much bigger than a demo. This is more significant than it looks, and the real test will be whether the company can turn that capital into research output that customers pay for, not just compute bills.
Mercor came in second with a $750 million Series D. The company connects AI developers with human experts for model training and evaluation, and the round put a $20 billion valuation on a San Francisco business that's less than four years old. Here's the thing: that valuation tells you investors see human-in-the-loop work as a durable layer of the AI stack, not a stopgap until models improve. I'd argue that's a bet on labor more than software. Models need people to judge their answers, label edge cases and tell them when they're wrong, and that work doesn't disappear just because the models get better. If you're building or buying AI systems, you'll want to ask who is grading the outputs and how that process holds up at scale.
Infrastructure and physical AI
Salt Lake City-based Torus closed a $500 million growth financing led by Magnetar Capital. Torus makes distributed power and compute infrastructure, and it used the occasion to launch Foundry, a modular system for AI inference compute. Power and compute are now being financed as one package, and that's a meaningful shift. For most of the current cycle, the bottleneck story has centered on chips, with power treated as someone else's problem. Infrastructure companies that bundle generation and compute are betting that the grid, not the GPU, is where the constraint bites hardest. (And this is the part most people overlook: a modular system is only as useful as the sites where it can actually be plugged in, which makes siting and power access the real product.)
Path Robotics, which started with welding and now describes its work as physical AI for robotics and manufacturing, raised $300 million in Series E funding. Welding first. Then the pitch got much bigger. That's a common path for industrial robotics companies that find a narrow, paying use case and then argue for a wider platform. Georgian led the round, which valued the Columbus, Ohio company at $2.45 billion. The valuation rests on whether the broader physical AI story can be demonstrated on factory floors, where timelines are long and customers are conservative. Manufacturing buyers tend to adopt new automation slowly, so a company that can prove its existing revenue while expanding its scope is in a much stronger position than one that only has a vision.
Software, tooling and the cost of governing agents
On the software side, LangChain closed a $300 million Series C at a $3 billion valuation. The company offers a platform to run, improve and govern AI agents, and CapitalG led the round for the four-year-old San Francisco business. That's a lot of money for the tooling layer, and it suggests investors expect agent governance to become a budget line, not a feature. Agents that take actions across systems create a different kind of risk than chatbots that only produce text. Someone has to track what they did, why they did it and whether they stayed within bounds. Tooling that handles that work is starting to look less like a developer convenience and more like basic operational infrastructure.
Nominal, an Austin company building intelligence tools for hardware engineering, raised $240 million at a $3 billion valuation. Lightspeed Venture Partners and Insight Partners led the round. Hardware engineering has historically generated less venture excitement than software, partly because data is messy and sales cycles are slow. Tools that make sense of test data and engineering workflows are a natural fit for AI, since much of that work is pattern recognition across large, poorly labeled datasets. A $3 billion valuation on that thesis is a strong statement, and it's the kind of number that will need years of customer expansion to justify.
The rest of the list
Non-AI deals were scarce, but they weren't absent. Imprint, a payments, loyalty and customer intelligence platform for brands, took $130 million at a $1.9 billion valuation. Khosla Ventures led the financing for the six-year-old New York company. Loyalty and payments have been attractive to investors for years, and a platform that sits between brands and their customers' purchase data can be valuable if it keeps the relationships sticky. The money here is less about novelty and more about established commercial plumbing.
Biotech made the cut through Delix Therapeutics, a Bedford, Massachusetts company that says it has "reverse-engineered and optimized psychedelics" to treat psychiatric and psychological disorders. It closed a $100 million Series C, led by Johnson & Johnson Innovation and the Dementia Discovery Fund from SV Health Investors. That's a bold claim, and I'd treat it with the skepticism it deserves. Psychedelic drug development has a long and uneven history, and any company making this kind of pitch will ultimately be judged by clinical results, not by the language in a funding announcement. The presence of a large pharma-linked investor is a signal of interest, but it isn't proof.
Several $100 million-plus rounds closed in a tie at the bottom of the list. Rox, a San Francisco company building an AI revenue agent for enterprise sales, announced a $100 million Series B led by General Catalyst and March Capital. Databento, a finance data platform based in Salt Lake City, raised a $100 million Series C at a $1.1 billion valuation, led by Fifth Down Capital. Sales and finance are two areas where AI tools can point to measurable outcomes, which makes them easier to pitch to investors than vaguer productivity claims. Still, measurable outcomes have to show up in customers' books before the valuations look sensible.
Mavria also belongs in this group. The startup, which is focused on standardizing how AI infrastructure gets developed, built and deployed, launched publicly and said it has raised $130 million across pre-seed and seed rounds. Its latest round, led by Paradigm, sets a $1 billion valuation, and its founders previously worked at the U.S. Department of Energy. A company this early reaching a billion-dollar valuation is unusual, and it says a lot about how investors are pricing standardization bets. Whether the market rewards that kind of abstraction layer will depend on whether teams actually adopt a shared standard rather than building their own.
What this week's list is telling you
Step back from the individual deals and the pattern is hard to miss. Capital is concentrating around three things: the people who train and evaluate models, the physical systems that power and house them, and the software that keeps agents under control. Each category has a different risk profile. Labor-based businesses can be undercut if automation improves faster than expected, while infrastructure carries heavy capital requirements and long construction timelines. Tooling companies face the usual question of whether customers will pay for governance once the novelty wears off.
That said, announced rounds aren't the same as revenue. A $3 billion valuation on a tooling company or a $2.45 billion valuation on an industrial robotics firm is a claim about future outcomes, and those claims are only as good as the adoption curves behind them. If you're working in this space, the useful question isn't which company raised the most money. It's which customers are paying for these products today, and how much of that spending is discretionary versus embedded in core operations.
And yet the funding itself is a real signal. Investors with access to deal flow are moving large sums into AI-adjacent businesses at a pace that suggests they see the current categories as durable rather than experimental. Whether that conviction holds will show up over the next few cycles, when some of these companies have to explain their growth to a market that's less forgiving than it was when the rounds closed. I'd watch the infrastructure and governance names most closely, because their success depends on things that can be measured.
How the list was built
The rankings cover the largest announced rounds raised by U.S.-based companies between Nov. 7 and 13, 2026, drawn from the Crunchbase database. Most announced rounds get logged quickly, but some surface late in the week, so the list can shift slightly as more deals are reported. That matters because "announced" and "closed" aren't always the same thing, and some rounds are reported in stages or with terms that change later. Treat the ranking as a snapshot of what was disclosed in that window, not a final tally of who raised what.
