Docs: AI data startup Mercor had $614M in gross revenue in H1 2026, up 70% from all of 2025, with ~91% from foundation model makers like OpenAI and Anthropic
THE SO WHAT
A three-year-old data vendor doing $614 million in H1 revenue — 91% from top model labs — shows that high-quality human feedback is a primary cost center in frontier model training. If you’re not a lab, expect tighter competition and higher prices for the same annotator and evaluator capacity you need for your own systems.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIThe AI Copyright Lawsuits Have Finally Produced an Actual Payout
$1.5B moving from an AI lab to rights holders turns copyright risk from theoretical to priced-in. If your models or data vendors touch copyrighted corpora, you now have a reference number for downside exposure — legal, PR, and cash — and boards will start asking for that line item.
Applied AIGoogle releases three new Gemini models — but no 3.5 Pro
Google doubling down on 3.6 Flash, 3.5 Flash-Lite, and Flash Cyber — and still holding back 3.5 Pro — says the near-term fight is on cost, latency, and specialization, not max IQ. If you’re building on Gemini, design for a portfolio of cheap, task-tuned models rather than betting your roadmap on a single flagship that keeps slipping.
Applied AIAnthropic’s $1.5 billion book piracy settlement approved by judge
Judicial approval of a $1.5B author settlement cements a template for how training-data disputes might be resolved at scale. If you’re training or fine-tuning models, expect counterparties to benchmark against this outcome when they price licenses, indemnities, and audit rights.
Applied AIGoogle shipped three cheap Gemini models and a Mythos rival, but not the one that matters
Google’s answer to being outrun is a fleet of low-cost Gemini Flash models and a Mythos competitor — not a new Pro flagship — ahead of earnings. If you’re choosing a foundation partner, assume Google’s near-term edge will be on price-performance and verticalized variants, not on having the single strongest general model.