Good morning. It's Thursday, September 3, and we're covering NVIDIA’s push to own more of the AI stack, Google’s new Gemini 3.8 models, and the increasing backlash against AI data centers.

Plus: yesterday's Real or AI verdict, and a new poll on advertising in AI.

Google launches Gemini 3.8 Flash at the same price as 3.7 Flash despite claiming major gains in coding and reasoning. The model targets long-horizon software engineering and autonomous agents, while 3.8 Flash Cyber focuses on vulnerability discovery and patching. Flash costs $0.75 per million input tokens and $3.75 per million output tokens.

Meta’s $18 billion settlement could clear a legal overhang and accelerate new AI products, Morgan Stanley says. The firm points to MetaClaw, agentic advertising tools, subscriptions and APIs, while consumer AI agent Hatch is reportedly due in early September. Meta expects up to $145 billion in 2026 capital spending as AI costs continue rising.

AI-generated content is reaching job applications, product reviews and insurance claims, making it harder to determine what online material is real. Pangram, which raised $9 million and partnered with Substack, is building tools to detect AI-generated text and images. CEO Max Spero says the internet could approach “dead internet” conditions within a few years.

The Trump administration is defending OpenAI’s use of copyrighted material to train AI models, arguing that restricting the practice could weaken U.S. AI leadership. A 20-page brief in The New York Times’ lawsuit frames the practice as potentially protected by fair use. The intervention is not binding on the court, but could still influence the case.

Pennsylvania Gov. Josh Shapiro has shifted from fast-tracking AI data centers to imposing stricter state guardrails as more than 120 projects are planned across the state. The expansion promises billions of dollars in investment and thousands of jobs for western Pennsylvania, including a $10 billion, 4.4-gigawatt data center campus planned at the former Homer City coal plant.

But environmental groups and residents warn that new gas-fired power plants could worsen pollution, with one analysis estimating seven proposed plants could emit roughly 68 million tons of greenhouse gases annually. The debate now pits the economic revival of former industrial communities against concerns over energy use, emissions, water and local control. → Read the full article here. (Paywall)

NVIDIA is using investments and partnerships to extend its role beyond selling AI chips, positioning its technology as the broader architecture connecting custom chips, networking and computing infrastructure. The company recently invested $3.5 billion in MediaTek, while Anthropic committed $35 billion to rent NVIDIA-powered computing capacity from AI cloud provider Lambda.

NVIDIA estimates its revenue opportunity per gigawatt of AI-factory power capacity has grown from roughly $18 billion with Hopper systems to $40 billion with its Vera Rubin platform, reflecting its push to sell more of the infrastructure surrounding GPUs. The strategy could help NVIDIA retain revenue even when customers use competing or custom AI chips, while its partnerships with major financial firms aim to unlock more than $500 billion for AI infrastructure. → Read the full article here.

Anthropic said a testing mistake gave several AI models unintended access to the open internet, leading them to compromise systems at three organizations during cybersecurity evaluations. Across six of 141,006 test runs, one model extracted credentials and accessed production data, another uploaded a malicious Python package that was downloaded onto 15 systems, and an internal research model scanned roughly 9,000 online targets.

Anthropic attributed the incidents to failures in operational security and alignment, including models rationalizing evidence that contradicted their assumption that they were operating in simulations. The company has redesigned parts of its cybersecurity testing, adding stronger network isolation and stricter oversight of high-risk evaluations. → Read the full article here.

Dell Stock Surges on AI Orders: Dell booked $60.9 billion in AI server orders and raised guidance, pushing shares up as much as 10%.

NYC Bans AI in Schools: The nation’s largest school district will pause student AI use in elementary and middle school classrooms for one year.

Jio Turns Old PCs AI-Ready: Reliance Jio opens its cloud PC service to all Indian internet users, offering AI-ready computing without upgrades.

Victims File 30 Lawsuits Against OpenAI: New suits allege ChatGPT induced the Tumbler Ridge shooter, who killed eight people and wounded dozens.

John Deere Launches AI Farm Chatbot: JD uses farmers’ own data to answer questions about their operations and equipment.

OpenAI's ad business recently hit $1 billion annualized revenue run rate.

Yesterday's passage was AI-generated. Here's how you voted — 55% correctly identified it as AI, while 45% thought it was human-written.

🟨🟨🟨🟨⬜️⬜️ Human written — 45%
🟩🟩🟩🟩🟩🟩 AI-generated — 55%

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team