Blog

Mark Zuckerberg’s newest manifesto casts him as a defender of the individual against concentrated technological power. He argues that “superintelligence” should be widely distributed, enabling people to create, compete, and retain leverage against institutions. Zuckerberg presents openness, individual empowerment, invention, and balance of power as the principles that should govern AI’s future.

Those principles are appealing, but they’re also hard to separate from Meta’s immediate competitive interests.

Zuckerberg hasn’t always treated decentralization as a civic imperative. In 2017, he presented Facebook as essential social infrastructure: a platform that could build a supportive, safe, informed, civically engaged, and inclusive global community. In his World Economic Forum manifesto, the language was humanitarian, but the vehicle was a closed platform controlled by Facebook. The company would supply the tools, set the rules, measure the outcomes, and decide how billions of people encountered information and one another.

That arrangement was never merely benevolent. As Ben Thompson argued in 2017, Facebook’s power flowed from its control over the social graph and, therefore, over consumption and distribution. The platform’s stated mission to connect the world sat uneasily beside an advertising model that rewarded engagement. Subsequent years made that tension unmistakable: algorithmic systems designed to maximize attention could amplify outrage, misinformation, and division while also maximizing revenue. Zuckerberg’s concern then wasn’t that one company held too much power; it was that Facebook should wield its extraordinary power more responsibly.

Now, however, the AI landscape looks different. Meta has spent aggressively to compete in the race for frontier models, yet its models are nowhere near the leaders. The company recently delayed its Avocado model after tests showed it trailing rivals. Meta was even considering licensing Google’s Gemini while reorganizing its AI efforts. Meanwhile, its enormous infrastructure spending has spooked investors concerned about debt, cash flow, and uncertain returns, even as its core advertising business keeps humming along.

Against that backdrop, Zuckerberg’s attack on centralized AI power strikes me as strategically convenient. Openness can be a genuine public good, especially if it prevents a handful of labs from becoming gatekeepers to an essential technology. But it can also be a useful argument for a company trying to weaken better-positioned rivals, reduce regulatory barriers, and turn access to models into a competitive equalizer. His call for “personalized superintelligence” should be judged not only by its rhetoric but by who controls the data, distribution, infrastructure, and economic gains behind it.

The broader lesson extends beyond Meta. As AI expenditures swell at Microsoft, Oracle, Meta, OpenAI, and other firms, executives may frame their investments as public missions: empowerment, national competitiveness, safety, productivity, or democratization. Those claims may contain truth. But investors, regulators, and users should recognize CEO posturing for what it often is—an attempt to dress up an expensive, uncertain corporate strategy as an unimpeachable public mission.

• WSJ: OpenAI’s High-Stakes, High-Touch Push to Make AI Work for Business (Jul. 22, 2026)
  OpenAI is pushing into enterprise with a $150 million partner program and DeployCo, a services firm embedding forward-deployed engineers to build AI apps for clients. 
• WSJ: OpenAI’s Clothing Ignited Debate. Its Design Director Explains the Vision. (Jul. 22, 2026)
  OpenAI launched Supply Co., a merch line including $15 socks, $175 sweatshirts, and a green ChatGPT basketball, which drew mixed online reactions but is selling out. AI companies use apparel to build brand loyalty, attract superfans, and provoke backlash.
• WSJ: Anthropic Doubles Midterm Spending to $40 Million to Push AI Regulation (Jul. 22, 2026)
  Anthropic is doubling midterm spending to $40 million, giving $20 million to Public First Action to push for stronger AI safeguards, transparency, and regulation. 
• AP News: Judge approves a $1.5B Anthropic settlement over books used to train Claude (Jul. 21, 2026)
  A federal judge approved a $1.5 billion settlement after Anthropic used pirated books to train its Claude chatbot. Anthropic will pay thousands of authors about $3,000 per book.
• WSJ: Nvidia Supplier Wistron Opens $700 Million Texas Site Producing AI ‘Superchips’ (Jul. 22, 2026)
  Wistron opened a $700 million Fort Worth factory to build Nvidia superchip modules, scaling to tens of thousands monthly. The move signals Taiwanese suppliers pivoting to AI hardware, and a U.S. supply-chain buildout.
• WSJ: Utilities Join Trump Pledge to Limit AI-Driven Increases in Electricity Bills (Jul. 21, 2026)
  Nearly 200 utilities, data-center developers, and tech firms pledged to pay more for electricity and fund grid upgrades to prevent AI-driven bill increases.
• WSJ: Google Search Was a Lifeline for Publishers. Now They’re Thinking of Cutting it Off. (Jul. 21, 2026)
  Reddit and several major publishers are weighing blocking Google’s crawlers or ending deals as AI search reduces referral traffic, ad revenue, and web clicks.
• WSJ: White House to Redirect Billions in Research Funds Toward AI, Away From Colleges (Jul. 21, 2026)
  The Trump administration will shift federal research funding to individual scientists, AI-driven projects, and fellowships, instead of universities, to speed discovery. 

• TechCrunch: OpenAI's new voice mode makes it to the ChatGPT desktop app (Jul. 24, 2026)
  OpenAI added ChatGPT Voice to the desktop app, using ChatGPT-Live, to control agents, run multi-step tasks, and access apps and websites. 
• WSJ: Exclusive | OpenAI’s Planned Cloud Spending Hits $750 Billion as Computing Efforts Ramp Up (Jul. 22, 2026)
  OpenAI raised projected computing spending to about $750 billion, and will spend $20 billion to start Project Camellia in Georgia, securing 3.2 gigawatts. 
• WSJ: Nvidia to Invest $1 Billion in Naver’s AI Project (Jul. 26, 2026)
  Nvidia will invest about $1 billion in Naver, buying a 4.5% stake to fund an AI data center. 
• www.alexinch.com: Kimi K3 Is Not Cheap
  Kimi K3, a new open-weights Chinese LLM, is powerful but not cheap. Benchmarks show similar per-task cost to OpenAI’s top model, higher than other Chinese models, cheaper for coding, and pricier for office work.
• WSJ: Google Study Says AI Is Helping Workers, Not Replacing Them (Jul. 23, 2026)
  A Google study of millions of de-identified interactions finds AI mainly aids workers, rather than automating jobs. Use is shallow across occupations, including skilled blue-collar roles, office jobs, and home tasks, but it could shift toward automation.
• WSJ: Gen Z Is Turning to AI Chatbots to Guide Even Their In-Person Social Lives (Jul. 26, 2026)
  Young adults increasingly use AI to draft dating-app lines, texts, and in-person responses, easing anxiety and smoothing awkward moments. 
• TechCrunch: Librarians are hosting viral 'Avoiding AI' workshops for people who are fed up with Big Tech (Jul. 25, 2026)
  Librarians run “Avoiding AI” workshops to teach people how to disable AI features, explain risks, and reclaim control over devices. 
• University of Chicago Law School: Rethinking Legal Education in the AI Era (Jul. 6, 2026)
  The University of Chicago Law School will adapt legal education for the AI era, emphasizing AI-resilient pedagogy, essential human skills, and responsible AI use. Policies include limited device bans in 1L classrooms, AI-integrated research and writing, and elective experimentation.