Welcome to today's Daily Pulse from Nicolas's AI Lab - the AI briefing for founders, operators, and busy professionals. About a 6-7 minute read. Straight to what matters.

Anthropic just closed the largest copyright settlement in AI history, $1.5 billion for training Claude on pirated books. Amazon cut jobs inside its AGI research unit the same week its AI infrastructure spending hit $200 billion for the year. Microsoft committed billions more to Mistral, betting on Europe's home-grown lab over building everything itself. Here's what each move changes about how you budget, hire, and choose AI partners this week.

A federal judge granted final approval to Anthropic's $1.5 billion settlement with authors and publishers over pirated books used to train Claude, the largest copyright recovery in history. The deal covers more than 482,000 titles at roughly $3,000 per work, and about 91% of eligible works have already been claimed. Judge Araceli Martinez-Olguin approved the deal on July 21, building on an earlier ruling that training on copyrighted books is fair use, but that Anthropic wrongfully sourced the books from piracy sites.

Why It Matters: This closes the first major AI copyright case to reach a final number, and that $3,000-per-work figure now gives every other AI copyright suit, and every author weighing whether to join one, a concrete benchmark to negotiate against.

What To Do:

Source: Fortune

Amazon eliminated an undisclosed number of roles inside its AGI research unit, including model customization and post-training teams, the same week its 2026 AI infrastructure capex hit $200 billion. AWS infrastructure chief Peter DeSantis, who now runs AGI, admitted it's "a fair narrative that our models haven't been at the very frontier for the very largest, most demanding workloads." The cut roles overlap with what Amazon's Nova Forge product now automates for customers, and the company still has more than 100 AGI roles open.

Why It Matters: A company cutting frontier-research jobs while pouring $200 billion into infrastructure is telling you where it thinks the real AI money sits. Amazon is betting on being the infrastructure layer other companies build AI on, and letting Anthropic and OpenAI fight over whose model is smartest.

What To Do:

Source: Tech Times

Microsoft expanded its partnership with Mistral, committing a multibillion-dollar sum to expand the French lab's GPU capacity in Europe on Nvidia's new Vera Rubin chips. The deal integrates Mistral's Medium 3.5 and OCR 4 models into Microsoft Foundry and Copilot Studio, letting customers run them across Azure, Azure Local, or fully disconnected environments. Microsoft Vice Chair Brad Smith said the goal is giving "Europe access to the world's most capable AI without compromising control over their data."

Why It Matters: This is Microsoft underwriting a European alternative to its own frontier models, aimed at regulated industries, banks, hospitals, governments, that need frontier AI but can't or won't send data to a US-only stack. If you sell into Europe, sovereignty is turning into a real product requirement for anyone selling into regulated industries there.

What To Do:

Source: Microsoft

Deezer said AI-generated tracks now account for more than 50% of daily uploads to its platform, up from about 44% in April and from roughly 10,000 tracks a day in early 2025 to nearly 90,000 a day now. The company's fraud-detection system flags fully AI-generated tracks so they can be excluded from royalty payouts and recommendation algorithms.

Why It's Interesting: The volume of machine-made music now physically outweighs what humans upload, on a platform that still has to pay someone for every stream.

Key Takeaway: Detection and labeling infrastructure, not generation quality, is becoming music streaming's real AI battleground.

Source: TechCrunch

A small study had 16 volunteers talk with AI chatbots built to resemble people who had died, and many said the experience felt comforting and offered a sense of closure. Participants cared more about the chatbot's tone and personality matching the person than about factual accuracy, and researchers flagged risks around emotional dependence and misleading answers.

Why It's Interesting: It's a rare case where users explicitly say they don't want the AI to be accurate, they want it to feel right.

Key Takeaway: Grief tech needs its own safety standards built for the job; the accuracy benchmarks made for every other chatbot don't apply here.

Source: Science News

Tesla expanded its robotaxi service into Orlando and Tampa, just 18 days after launching in Miami, giving its camera-only self-driving system its toughest weather test yet: Florida's sudden downpours. The rollout landed one day before Tesla's Q2 earnings call.

Why It's Interesting: Tesla is betting its whole self-driving approach on cameras alone, no lidar, and picked one of the hardest weather environments in the country to prove it out in public, right before investors get to grade it.

Key Takeaway: Watch how robotaxi services perform in bad weather before trusting any self-driving safety claim built on clear-day miles.

Source: TechCrunch

Context: Amazon just cut AGI research jobs while defending its infrastructure bet, and Microsoft just backed a second AI lab instead of only pushing its own models. Both moves are a reminder that a vendor's headline model isn't the only thing worth evaluating, its actual strategy is.

Prompt: I'm evaluating [vendor name] as an AI partner for [specific use case]. Here's what I know about them: [paste public statements, recent news, or their pitch]. First, identify whether their recent moves, hires, layoffs, partnerships, point toward doubling down on frontier research or toward enterprise deployment and services. Second, tell me what that direction implies about how their pricing, support, and roadmap will likely change over the next year. Third, flag anything in their recent public statements that contradicts their own sales pitch to me.

Do Not: Do not evaluate a vendor only on their current model's benchmark scores without checking what direction their recent business decisions actually point.

If/Then: If a vendor's recent layoffs or hiring pattern contradicts the roadmap they're selling you, then ask them directly about it before signing, don't assume it's unrelated to your deal.

Example: An operations director used this prompt before renewing a multi-year AI infrastructure contract, noticed the vendor's own recent layoffs matched exactly the team that had built the feature she relied on most, and negotiated a shorter renewal term instead of the multi-year deal they'd proposed.

European Commissioner for Technology Henna Virkkunen told the Financial Times that Europe's heavy dependence on foreign AI and cloud infrastructure is becoming a national security risk in her framing, a sharper worry than the usual competitiveness talk, and that the bloc urgently needs its own domestic capacity.

Why It's Important: A senior EU official saying this in an interview, not a policy paper, signals Brussels is treating foreign AI dependence as a lever other governments could pull against Europe.

Business Takeaway: If you sell AI-powered products into Europe, expect sovereignty requirements, like the kind driving the Microsoft-Mistral deal, to become a standard line item in procurement decisions.

Source: EU Today

Nvidia said its next-generation Vera Rubin chip platform, a seven-chip system pairing a new Vera CPU with Rubin GPUs, has entered full production, with OpenAI set to deploy it at scale in the third quarter. Anthropic and Meta are also on board as early customers.

Why It's Important: This is the infrastructure generation that every major lab's next round of model training and inference will run on, and OpenAI committing to deploy at scale this quickly is a stronger signal than any spec sheet.

Business Takeaway: Expect a fresh wave of compute-cost and capacity conversations across every AI vendor you use as Vera Rubin rolls out, since pricing and availability usually shift around a chip generation change.

Source: Nvidia

London-based Humanoid raised $152 million in a Series A at a $1.35 billion valuation, becoming Europe's first pure-play humanoid robotics unicorn. Bosch will manufacture its wheeled humanoid robots at scale, aimed at industrial and warehouse work.

Why It's Important: A legacy industrial manufacturer agreeing to build a robotics startup's hardware at scale is a concrete signal that humanoid robotics is moving past prototypes and into an actual supply chain, the harder step most humanoid startups haven't reached yet.

Business Takeaway: If your business involves warehouse, logistics, or manufacturing labor, start tracking humanoid robotics timelines now, the manufacturing partnerships needed to scale them are starting to lock in.

Source: Forbes

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas