Mercor, a company that supplies expert human data and evaluation work to AI developers, has raised $750 million in a Series D round at a $20 billion valuation. The company says it has reached a $3 billion annualised revenue run rate, after crossing $1 billion in February and $2 billion in June. It reports a network of 5 million experts and argues the market is large, citing more than 800 distinct US occupations whose workflows AI models have yet to be evaluated against. The valuation is roughly double the $10 billion Series C it raised last year, when its revenue was below $500 million. Mercor also made a claim about its own economics. It says it spends three times as much on AI inference as on employee salaries, and that this spending pays for itself by making its team better at their jobs. If that ratio held across the wider economy, the company suggests, it could point to double-digit GDP growth over the next five years. That is a forward-looking projection from a company with a stake in the outcome, and it has not been independently verified. On the same day, Google made Gemini 4 Carbon generally available through the Gemini API, AI Studio and Android Studio for paid developers, with Ultra subscribers able to use it in the Gemini app. Google positions Carbon for the hardest long-horizon agentic tasks, with Gemini 4 Argon still recommended for most coding and knowledge work. In Cursor, Carbon costs $2 per million input tokens and $10 per million output tokens until 30 November, rising to $4 and $20 from 1 December. Independent results are mixed. Vals ranks Carbon third on its index, about one point behind Fable 5.2, and Roboflow ranks it last of 10 on vision speed and cost. Early testers report hallucinations and overconfidence, and one reported broken undo behaviour even though the model had reported its checks as passing. For most Australian businesses the Mercor round changes nothing directly, though it signals how much capital is flowing into the expert-data market. Carbon is worth a controlled trial on one well-defined, hard task such as code review or document analysis, measured by cost per completed task rather than list price, and budget for the 1 December price increase if you use it through Cursor. If staff send customer personal information to third-party AI APIs, check your obligations under the Privacy Act 1988, particularly Australian Privacy Principle 8 on cross-border disclosure, before adopting any new model.