Mercor, the AI training startup founded by Brendan Foody when he was 19 years old, is in discussions to raise at a $20 billion valuation, doubling its $10 billion valuation from just nine months ago. The company also crossed $2 billion in annualized revenue and acquired Deeptune, an AI agent training environment startup backed by Andreessen Horowitz. For founders watching the AI infrastructure stack, Mercor's trajectory is a signal that the data and training layer may be where the most concentrated value is being created right now.
CEO Brendan Foody is now 23 years old and worth billions on paper. His company's growth rate puts it in rarefied air. Mercor went from $1 million to $2 billion in annualized recurring revenue in just 24 months, which Foody called the fastest growth trajectory ever for a company at this scale. The last $1 billion of that came in only four months.
The Deeptune Acquisition Fills the Missing Layer in AI Agent Training
Mercor announced the acquisition of Deeptune on July 9, bringing in a team that builds simulation environments where AI agents practice real-world tasks before touching production systems. Think of it as a flight simulator for AI agents learning to navigate Excel, Salesforce, Slack, and other enterprise tools. Deeptune had raised a $43 million Series A led by Andreessen Horowitz just three months earlier, and Foody was listed as an angel investor in that round.
Foody told Fortune that the angel check was written with the acquisition already in mind. This is a strategic pattern worth noting: foody effectively used his position as an early-stage investor to evaluate and then acquire a company that filled a critical gap in Mercor's offering. For founders building in adjacent layers of the AI stack, this kind of strategic angel investing followed by acquisition is becoming a recognizable playbook, and it works because the investors funding the target (a16z in this case) get an exit and the acquirer gets technology that was already battle-tested by their own customers.
Mercor's network of more than five million domain experts already builds the tasks and scoring rubrics that tell AI models whether they completed a job correctly. Deeptune builds the software environments those tasks run inside. Together, they cover the full training stack from environment creation to expert evaluation and scoring. Mercor claims every member of the Magnificent Seven except Tesla is already a customer.
The $2 Billion ARR Milestone and What It Means for the AI Training Market
Mercor hitting $2 billion in annualized revenue is not just a number. It validates a thesis that many dismissed two years ago: that the data and training layer for AI would become a standalone infrastructure category worth tens of billions. When Mercor raised its $350 million Series C at a $10 billion valuation in October 2025, skeptics questioned whether an outsourced training operation could command that kind of multiple. The answer, it turns out, is yes and then some.
The growth from $1 billion to $2 billion ARR in four months implies a quarterly growth rate that most SaaS companies would describe as unbelievable. But Mercor is not a SaaS company in the traditional sense. It operates a marketplace of domain experts who create training data for frontier AI labs. As those labs scale their models and agents, the demand for high-quality, expert-curated training data scales with them. This creates a flywheel: better models need better training data, which makes models more capable, which creates demand for even more training data.
For founders, the lesson is straightforward. The companies building the core capabilities for frontier AI model providers are seeing revenue growth that traditional SaaS benchmarks cannot explain. If you are building in the AI infrastructure layer, Mercor's trajectory suggests that the market is willing to pay premium prices for quality, scale, and reliability. The bottleneck has shifted from compute to data quality and training environment fidelity.
The Data Breach That Did Not Slow Them Down
Mercor's rapid growth came despite a significant security incident earlier this year. In March 2026, hackers exploited a supply chain vulnerability in LiteLLM, an open source Python library widely used to build AI tools, and exfiltrated an estimated four terabytes of Mercor data. The stolen data allegedly included tax and banking information, passport scans, interview recordings, facial biometrics, and internal records. The hacking group Lapsus$ claimed responsibility and offered the data for sale on Telegram.
A class action lawsuit was filed in California in April. Mercor responded by saying that only a very limited subset of sensitive information was affected and that there was no evidence the data had been used fraudulently. When Fortune asked about the breach, Foody said every frontier lab had expanded its relationship with Mercor since the incident. The business grew from $1 billion to $2 billion in ARR during the four months after the breach was disclosed.
That response is either a testament to customer loyalty or a sign that frontier AI labs have few alternatives at this scale. Either way, it underscores a reality for founders: once you become critical infrastructure for the most important companies in AI, your customers may tolerate significant operational risk because switching costs are too high. That is a powerful position to be in, but it is not one you can manufacture. It comes from being the only provider that can deliver at a certain level of quality and scale.
What Comes Next for the AI Training Category
The $20 billion valuation target implies that Mercor's existing investors and new backers believe the company can sustain or accelerate its growth trajectory. With the Deeptune acquisition, Mercor now owns the full chain from environment creation to expert verification to model scoring. That vertical integration makes it harder for competitors to replicate what Mercor does. A competitor would need to build both the expert network and the software platform, which is a multi-year effort even for well-funded teams.
For founders building in AI, the Mercor story reinforces a few key observations. First, the AI training and evaluation layer is producing infrastructure companies with revenue profiles that look more like platform businesses than services businesses. Second, strategic acquisitions using angel investments as a pipeline are an effective way to fill product gaps without the competitive auction dynamics of a traditional acqui-hire. Third, security incidents at this scale are survivable if your product is indispensable, but that is a gamble that not every startup can afford to take.
The next twelve months will test whether Mercor can maintain its growth rate as it integrates Deeptune, navigates the class action lawsuit, and potentially becomes a public company candidate. A $20 billion valuation puts it in the conversation for one of the largest private AI companies by market cap. If the growth continues, an IPO within eighteen months is a reasonable expectation. For now, Mercor is demonstrating that the AI training layer is not just a support function for frontier models but a category-defining business in its own right.