Anthropic heads for the public markets as Mercor raises $750 million at $20 billion

Two raises, one pattern

Anthropic launched the roadshow for its proposed initial public offering of Class A common stock, the company says, and Mercor raised $750 million in a Series D round at a $20 billion valuation. The two deals look different on the surface: one is a lab going public, the other is a private company selling human expertise to labs. But the money is coming from the same place. Strategic investors including Nvidia, Alphabet, Microsoft, TSMC, Broadcom and AMD are committing $26 billion to Anthropic through a private placement, according to CNBC's Squawk Box, while Mercor's round included participation from NVIDIA, Salesforce and General Catalyst. The chipmaker sits on both sides of the week's capital formation, as a buyer of one company's future and a backer of the company that supplies the experts who train the models. m:…8985 m:…1557 m:…1274

The scale of the Anthropic deal is what makes it a funding story rather than a governance story. Anthropic's annualized revenue run rate reached $81 billion in October 2026, according to the same CNBC discussion, which puts the listing in the territory of a large operating company rather than a speculative lab. The offering itself consists of 825,000,000 shares of Class A common stock, of which 23,648,646 shares are being offered by certain existing stockholders, and Anthropic will not receive proceeds from the sale of shares by those selling stockholders. Underwriters get a 30-day option to buy up to 123,750,000 additional shares. A registration statement has been filed with the U.S. Securities and Exchange Commission but has not yet become effective, so no securities may be sold until it does. That is the normal machinery of an IPO; what is unusual is a lab arriving at it with a revenue run rate that large and a $26 billion private placement already lined up from its own customers and suppliers. m:…1274 m:…8985

What Mercor is actually selling

Mercor's network includes more than five million experts paid to train frontier AI models, the company says, and it pays out over $5 million per day to those experts, having paid out approximately $1 billion historically. The interesting part of the raise is not the headcount but the description of the business. Mercor's business model is not a marketplace but a provider of RL environments consisting of apps, content worlds and task rubrics, according to its chief executive on the Sourcery podcast. That framing matters because it explains the valuation: the company is selling the environments in which models are trained and tested, not just introducing experts to labs. It acquired DeepTune to solve the bottleneck of obtaining clones of enterprise applications like Salesforce and Adobe for those RL environments, and 70% of experts on the platform arrive through referrals, which keeps acquisition costs low while the payout rate stays high. m:…1557 m:…4423

Mercor's own spending pattern tells the same story as its pitch. The company spends three times as much on inference as on employee salaries, it says, and it has opened offices in New York and London. Inference is the cost of running models; salaries are the cost of the people. A company that spends three times more on the first than the second is a company whose product is model time, with humans as the input that makes that time valuable. Its chief executive forecasts that AI spreading through the economy could lead to double-digit GDP growth over the next five years, a forecast rather than a fact, and one that depends on exactly the diffusion his company is selling. Enterprises spending $10 million to $100 million on inference should allocate approximately 10% of that budget to evaluation and data set creation, he says, and enterprises spending over $100 million should begin post-training their own models to amortize fixed costs. That is a budget rule a buyer can test against the raise: if the advice is right, Mercor's customers keep spending, and the $20 billion valuation is a bet that the evaluation and data layer grows with inference itself. m:…1557 m:…4423

What to watch

The two deals answer different questions for an investor. Anthropic's listing will test whether public markets will pay a premium for a lab whose revenue run rate is large but whose costs are dominated by compute bought from the same investors funding the private placement. Mercor's round tests whether the human-expertise layer is a durable business or a temporary bottleneck that disappears as models get better at training themselves. The common thread is concentration: the same handful of companies appear as customers, investors and suppliers in both deals. That concentration is the risk and the opportunity at once. If the giants keep buying, both deals look cheap in hindsight; if they pull back, the two companies discover at the same time how much of their growth was one customer's budget. m:…1274 m:…1557