AI Sector Daily Digest: August 12, 2026 — Nebius' AI-cloud numbers and the push for model control

In brief

1. Nebius shows what AI-cloud demand looks like in revenue

• Amsterdam-based Nebius reported $582.3 million in second-quarter revenue, ahead of the $572.75 million analyst average. Revenue from its core AI-cloud business rose more than 500% and made up about 98% of group revenue. 1
• The company closed four AI-cloud contracts during the quarter, each with an average total contract value above $1 billion. Total contract value nearly quadrupled from the previous quarter, and Nebius reaffirmed its full-year 2026 outlook. 1
• About 70% of the period's deals included customer prepayments covering 50% to 60% of related capital expenditure. That is the important operating detail: customers are helping fund capacity before Nebius builds it, while demand is also lifting prices for both new and older GPUs. 1

2. River AI raises $1.1 billion for custom models

• River AI, founded by xAI co-founder Igor Babuschkin, said it raised $1.1 billion to help companies build personalized AI models on their own data. General Catalyst and AMP PBC led the round; Nvidia, AMD Ventures, Y Combinator, and Temasek also invested. 2
• River's thesis is that enterprises will move from renting general-purpose models from large labs toward customizing and owning open-weight models. Its API is designed to run reinforcement-learning jobs in 15 to 20 minutes without a dedicated infrastructure team. 2
• River claims its approach is two to four times more cost-effective than closed-source alternatives, but it did not disclose a valuation and the performance and cost figures are company claims. The round's investor mix still matters: both leading GPU vendors are backing a company that wants more model control to sit with customers. 2

3. Anthropic adds watermarks to Claude output

• Anthropic said models released after August 2 will automatically watermark generated text and files, including output from Claude, Claude Code, Claude Cowork, Claude Tag, and the API. The announcement follows the EU's AI-transparency rules taking effect on August 2. 3
• For files, Anthropic is using the C2PA open standard. The company also says it will extend watermarking to older models and that text-level marks travel with copy-and-paste, may survive some editing, and are applied at the model level rather than only inside one product. 3
• Anthropic has not said how much editing is needed to remove the text watermark, and TechCrunch was still seeking clarification. For teams building detection or provenance workflows, the practical status is therefore partial: the policy is announced, but its robustness against rewriting is not yet established. 3

4. Manus returns to independence as its Meta deal unwinds

• Chinese AI startup Manus said it will resume operating independently from Meta. Some user data generated on or after December 29, 2025 will be deleted later this month to comply with regulatory requirements in specific jurisdictions. 4
• Manus said affected users will be notified in the app and by email and will have a chance to back up their data. The company described the deletion as part of its separation from Meta, rather than as a normal product migration. 4
• Beijing ordered Meta to unwind its $2 billion-plus acquisition in April, and Reuters previously reported that Tencent was discussing becoming Manus' largest shareholder. The immediate issue for users is data continuity; the larger industry signal is that Chinese scrutiny can reshape the ownership of a frontier AI company after a deal is announced. 4

5. Brad Lightcap leaves OpenAI

• Brad Lightcap, one of OpenAI's longest-serving executives, said he is leaving to "start something new." He joined OpenAI in 2018, served four years as CFO, then became COO in 2022 before moving to special projects during an executive reshuffle earlier this year. 5
• Lightcap said he helped build OpenAI's Finance, Legal, People, corporate security, go-to-market, government relations, and partnerships teams. He did not identify his next venture or give a departure date in the message shared with staff. 5
• The departure follows other senior changes, including Fidji Simo's move away from OpenAI's No. 2 role in July. It is not evidence of a strategic failure by itself, but it adds another senior transition while OpenAI prepares for an IPO and reorganizes responsibilities around the next phase of the company. 5

The read-through