Mercor's $750M Round Leads a Big Day for AI Builders
Four AI companies raised more than $915 million between them this week, and they're all working on a different layer of the stack. Mercor supplies the human experts who train and grade AI models. NatureSelect.ai builds AI companions for consumers. Graphon Labs wants to give AI agents a working memory. Tendrils Compute is trying to redesign the processor itself.
On the surface they don't have much in common. One is a $20 billion marketplace, another is a Shenzhen consumer app maker and two are small research-heavy teams. But together they show where investor money in AI is going right now. It's no longer just about building bigger models. It's about everything that makes those models useful, trustworthy and cheaper to run.
Here's a closer look at each deal.
Mercor: $750 million to grade the AI agents
Mercor raised $750 million in a Series D that values the company at $20 billion. The round, announced on November 10, didn't have a named lead. Instead it pulled in a mix of corporate and venture backers. Nvidia, Salesforce and Zoom came in on the corporate side, and returning investors Benchmark, Felicis Ventures and General Catalyst also took part.
If you haven't followed Mercor, the business is simpler than the valuation suggests. AI labs need people with real expertise to write training data and judge whether a model's answers are any good. A chatbot can't learn to review a legal contract or read a medical scan from random internet text alone. Someone who actually does that work has to show it what good looks like. Mercor runs a network of more than 5 million domain experts who do that work for AI companies.
The new money is aimed at a shift in what those experts do. Mercor is moving into evaluating AI agents for large companies, building custom benchmarks for each customer. That's a smart move. Businesses are starting to put agents into real workflows, from customer support to finance. The obvious next question is how anyone knows the agent is doing the job right. Generic benchmarks don't answer that. A bank wants to know how an agent handles its own loan files, not a public test set.
The investor list fits that story. Salesforce and Zoom both sell agents to businesses, so they have a direct interest in better ways to test them. Nvidia, meanwhile, benefits from anything that keeps AI spending growing.
A $20 billion price tag is a lot for what started as a hiring marketplace. But Mercor has turned human expertise into a supply chain for AI, and right now demand for it is far ahead of supply.
NatureSelect.ai: $100 million-plus for AI companions
Shenzhen-based NatureSelect.ai closed new financing of more than $100 million, according to DealStreetAsia. The company didn't name a lead. Backers include 5Y Capital, Qiming Venture Partners, SCV, iMPC and HyT Capital, a group that includes some of China's best-known consumer tech investors.
NatureSelect makes two main products. EVE is a 3D AI companion with long-term memory, meaning it remembers past conversations and builds on them over time. Elys is what the company calls an agentic social network, a social app where AI agents take part alongside people. The company describes its mission as building experiences that promote long-term relationships between humans and AI.
That might sound like science fiction to some readers, but AI companionship is already a real business. Millions of people chat with AI characters every day, and the most popular apps keep users engaged for long stretches. The hard part has been memory. A companion that forgets what you told it last week doesn't feel like much of a companion. NatureSelect is betting that long-term memory is what separates a toy from a product people stick with.
The category also raises fair questions about loneliness, dependence and how these apps make money from emotional attachment. Regulators in several countries are starting to pay attention. For now, though, investors clearly see a large consumer market, and a nine-figure round shows how seriously they're taking it.
Graphon Labs: $35 million to help agents remember
Graphon Labs, a San Francisco AI lab, raised a $35 million Series A led by RRE Ventures. That brings its total funding to $43.3 million.
Graphon is working on what it calls "Persistent Intelligence." In plain terms, it's building infrastructure that lets AI agents store information and recall it later, so they don't start from zero every session. The company's approach is based on graphons, a mathematical tool for describing very large networks.
Memory is quickly becoming one of the biggest problems in AI agents. An agent that handles your expenses or manages a project needs to remember what happened yesterday, last month and last quarter.
Today most agents get around this by stuffing as much text as possible into each prompt, which gets expensive and still loses details. A real memory layer could make agents cheaper to run and far more reliable. Graphon is one of a growing group of startups chasing that problem, and a $35 million Series A gives it room to prove its math works at scale.
Tendrils Compute: $30 million to rethink the CPU
Tendrils Compute, based in Cambridge, UK, raised $30 million in a round led by Benchmark. Tech.eu described the round as hotly contested. Visionaries Club, the UK Sovereign AI Fund, FR8, Antimatter, Cloudberry VC, Conviction, Embassy Ventures and SV Angel also invested.
Tendrils is going after one of the hardest problems in tech: building a new kind of processor. Its chip is based on a computing model called interaction nets, which is designed to spread work across thousands of cores running in parallel. The goal is to challenge both conventional CPUs and GPUs.
There's a real caveat. Tendrils hasn't published hardware benchmarks yet, so for now the claims are mostly on paper. Chip startups also need a lot of time and money before they ship anything. Still, landing Benchmark as lead is a strong signal, and the presence of the UK's Sovereign AI Fund suggests the British government sees this as a strategic bet.
What ties these deals together
Look at the four rounds side by side and a pattern shows up. The money is flowing to the parts of AI that sit around the model, not just the model itself.
Mercor supplies the human judgment that models still can't do without.
Graphon Labs is working on memory, which agents need to be useful over time.
Tendrils Compute is trying to make the hardware faster and cheaper.
NatureSelect.ai is building a consumer product on top of all of it.
For founders, the lesson is that there's still plenty of room in AI outside the frontier labs. You don't need to train a giant model to raise a big round. You need to solve a problem that gets bigger as AI adoption grows. Evaluation, memory, compute and consumer experience all fit that description.
The risk is that valuations are running ahead of results. Mercor is priced at $20 billion, and Tendrils has yet to show benchmark numbers. Plenty of companies raising today will struggle to grow into their price. But for this week at least, investors are still writing large checks, and they're spreading them across the whole AI stack.
