That Was The Week

1. We’re losing our nerve at the wrong momentThe technology keeps rewarding ambition, while boards, investors, companies and politicians are each choosing caution. Nerve isn’t recklessness. It’s making a considered commitment before the evidence is complete.

2. Nvidia’s board acted before the facts were inAccepting Jensen Huang’s resignation before the independent investigation reports erased more than $855B in value. With 60 direct reports and no named successor, the board had already neglected continuity, which is a fiduciary duty too.

3. Anthropic’s IPO is a fall in mood, not in the businessExpectations slid from $2T to about $1.4T. Even so, that’s well above the $952B spring round, at about 17x an $81B revenue run rate. The bear case to take seriously is McCormick’s, about $900B in compute commitments, not the headline drop.

4. Venture capital is caution with a bigger checkSeries A dollars are up 60% to $42B while deal counts stay flat. Funds on their fourth raise or later took 88.2% of US fundraising, and first-time funds took 4.5%. Capital crowding into consensus companies concentrates the risk it is trying to avoid.

5. China builds data centers in a year; America spends four on paperworkA campus in Inner Mongolia delivered 120MW within a year at about one-fifth of US cost. A US environmental impact statement takes nearly four years on average. Tunguz says the US could be importing AI tokens by 2028–2030. Pay-your-way rules, yes. Moratoriums, no.

Five Take Aways:

1. Winning the election was the easy part. Now Democrats need a plan to grow wealth.Taxing billionaires won’t pay for health care, education and pensions. The economy has to grow, and AI is driving most of US GDP growth right now. The editorial’s proposal is a Human Wealth Fund: AI companies give equity to a fund that every citizen owns directly, not the government. Anthropic’s and OpenAI’s public-benefit stakes show it can be done.

2. Voters turned against data centers, the election confirmed it.Of 47 local data-center measures on the ballot, the “restrict” side won 44 of the 45 that have finished counting. Governors who back a pause now oversee 41% of planned US data-center capacity. Ben Thompson’s point is that political permission may now limit AI more than chips or power do. And as Peter Walker says, the backlash is really about AI.

3. AI is now helping build AI.At Anthropic, the share of R&D tasks led by AI rose from 1% in March to 26% in August. METR estimates the company’s rate of progress has roughly doubled. Separately, an AI-assisted research system proved a quantum-computing conjecture that had been open since 2008, and the proof was formally checked in Lean.

4. There’s a record amount of venture money, but it’s going to fewer companies.About $680B was invested globally through Q3. But funding for rounds under $100M has been flat since 2023, and 27 companies took a third of Q3’s money. Established firms raised 88.2% of US venture funds. Kyle Harrison calls this the “McDonaldization” of VC, and Sequoia is cutting back to focus.

5. The whole industry is betting on token demand.On OpenRouter, token usage is up about 360x since January 2025, while the price per token has fallen about 80%. Anthropic’s S-1 shows compute costs dropping from roughly twice revenue to about 55% of revenue in a year. That’s the strongest evidence yet that AI spending can turn into a real business. Damodaran values Anthropic at about $1.45T.

1. The stock market just called time on the old internet.ServiceNow -27%. Intel -29%. Reddit -25%. Alphabet’s worst month ever. Nasdaq’s worst month in over a year. It felt like 2002 all over again. But this was not a crash. It was a repricing. The market can finally see who captures the gains from cheap intelligence, and it is not the companies that ran the web for the last twenty years. Read why the losses are a birth, not a death.

2. 65 million to 7.4 billion. That is the real story of the week.GitHub commits per month went up more than a thousandfold in two years. Software is being written at a rate no SaaS company could ever serve. The catch: a thousand times more code, zero times more software companies. People are building their own tools with AI. I did. SignalRank Intelligence was built by me and an AI, and it is now the leading data-driven venture portal. See what “smallness at scale” means for every software business you own shares in.

3. Google’s best product is killing Google’s best business.AI Mode answers the question. No journey, no click, no ad. Search revenue still grew 12%, but Google warned it gets worse next quarter. Downstream, Reddit grew revenue 53% and still lost a quarter of its value because search referrals are falling by a million a week. Reddit’s answer is to lock the doors. I think that is exactly the wrong move. Find out what Reddit should do instead.

4. The money did not vanish. It moved. Fast.Anthropic’s S-1: $16.3 billion in revenue in six months, 14x the year before, customers growing spend roughly 900% a year, 85% gross margins and over $130 billion in the bank. The headline “loss” is equity accounting, not cash. The value of the internet is not disappearing. It is relocating to the model layer, and the rate of relocation is accelerating. Tunguz and Jamin Ball decode the filing, including the $637 billion compute bill nobody can accounting-engineer away.

5. Fewer companies. Bigger cheques. A canvas wide open.92% of Series B dollars last quarter went to rounds over $100 million, raised by just 14% of the companies. California startups have raised $385 billion this year, more than the entire US in any prior year. Capital is concentrating while the tools to build have never been cheaper. Is that bad? No. It is a whole new canvas for human endeavor, from drug discovery to the software on your own laptop. This is the week to lean in. Here is how.

1. Abundance is probable; shared prosperity is not automatic. AI and robotics can create extraordinary wealth, but connecting that production to everyone’s livelihood requires a plan.

2. People should be owners, not beneficiaries. Corporate generosity and government-administered benefits are different from an enduring ownership claim. The distinction is independence.

3. A Human Wealth Fund should belong to everyone. Globally distributed ownership of productive assets, with independent managers accountable to those owners—not governments or contributing companies.

4. Permanent ownership should fund life through dividends. Your sharper proposal is a stake that cannot be sold, distributing returns as wealth grows. That protects participation from ownership concentrating again, although investment and transition risks remain.

5. The purpose is freedom—and ownership aligns people with progress. AI should expand everyone’s choices, not require everyone to become a founder. A direct stake gives people a reason to welcome its success.

The central takeaway: Don’t just distribute AI’s benefits. Distribute ownership of what produces them.

Five takeaways from this week’s issue:

1. Venture capital is separating into different businesses—not simply failing. Financing the scale of established AI leaders and discovering companies before consensus require different LPs, funds, checks and skills. Kyle Harrison sees strategy drift; Fred Wilson argues bigger funds can preserve investment discipline.

2. The overlooked opportunity is financing the discoverers. Specialist early-stage managers still find exceptional companies, but large LPs increasingly need somewhere to deploy enormous checks. Silent Ventures illustrates how specialist discovery and separately supplied follow-on capital can work together—not proof that small funds universally outperform. SignalRank's pro-rata as a service to smaller managers is similar.

3. AI’s next business opportunity is making capability usable. The “diffusion gap” and rise of deployment companies point toward integration, workflow redesign and accessible products. Better models alone do not produce paying customers. That creates room for startups beyond the frontier labs. There is opportunity everywhere.

4. More intelligence exposes the next bottleneck. Anthropic reports Claude uncovering the origins of cryptographic parameters; PostgreSQL reviewers are finding defects faster than maintainers can repair them. Progress can shift the constraint from discovering problems to verifying findings and acting on them—not eliminate the work.

5. Trustworthy AI needs enforceable controls, not reassuring promises. Nadella’s distinction between intelligence and authority connects with Anthropic’s independent-evaluator initiative and the robotics emphasis on bounded, reliable tasks. Permissions, audits, human intervention and independent scrutiny are foundations for deployment—not evidence that safety is solved.

1. Being smarter doesn’t give AI the right to decide for you. What matters is what it can access, what it can do and who can stop it. Those controls must be engineered and enforced. A more intelligent model is no substitute for accountable humans.

2. “Find me the best insurance” isn’t permission to buy it. That distinction is the foundation of useful delegation. Your agent needs enough freedom to save you time, with clear boundaries around spending your money, sharing your information and making commitments.

3. Your AI should work for you, not the website selling to you. A shopping agent should help you compare offers and walk away from a bad deal. Businesses can defend against abuse without shutting out your representative. You shouldn’t have to pay a toll to escape the advertising funnel.

4. You can say “thank you” to a machine without making it a person. Human-like conversation doesn’t establish human feelings or rights. Both worshipping AI and treating it as an enemy distract from the people building, deploying and using it. They remain responsible.

5. AI’s democratic promise is more power in your hands. More people could gain the ability to understand complex issues, challenge experts and hold institutions accountable. That won’t happen automatically. It depends on who gets access, whose interests the tools serve and whether the benefits are widely shared.

1. Your next computer may be operated by your agent—not you. You describe the outcome you want. Your assistant uses the browser, software and tools to get there.

2. Your voice may become the interface. Instead of opening an app or sitting at a keyboard, you speak to your assistant wherever you are—and let it handle the task.

3. You don’t need more software. You need things done. Finding the right product, renewing insurance or arranging a return shouldn’t require your constant attention. The reward is time back.

4. You shouldn’t have to start from scratch every time. An assistant becomes more useful when it connects your email, calendar, travel and ongoing projects—bringing the relevant context to each new request.

5. A personal assistant shouldn’t be a luxury. The opportunity is to make practical, everyday help available to people who could never hire staff. Being offered an AI assistant is the beginning; making it affordable, reliable and worth trusting is the real test.

1. The agent becomes the interface; the device becomes secondary.

Facebook’s Muse points toward a different relationship with computing: delegate a task rather than navigate an application. A persistent agent operates a computer in the cloud, while glasses, earphones or a phone provide access. Hardware doesn’t disappear, but the device becomes less important than the agent available through it. Microsoft's announcement of autopilot is the same in a work context.

2. Free AI could change the economics of the industry.

Muse’s proposed transaction-funded model challenges the assumption that useful AI requires a subscription. If shopping commissions can support a free personal agent, Meta could bring delegated work to a mass audience. That puts pressure on both paid AI services and the device ecosystems through which people access them. The opportunity is substantial, though competitors can respond.

3. Responsibility for AI safety starts with its builders.

Jensen Huang’s challenge, central to the discussion, is that companies claiming their products cannot be controlled should not release them. Safety requires engineering, testing and enforceable limits. Regulation can establish obligations, but it cannot substitute for the technical controls needed to meet them.

4. The fight over regulation is also about competition and liability.

Global rules could protect the public, but they could also entrench established companies and encourage a compliance-based defence of harmful outcomes. Those possibilities deserve scrutiny. So do the commercial interests of regulation’s opponents: NVIDIA benefits from continued AI expansion. Neither side’s business interests automatically prove or disprove its arguments.

5. Technical control and social consequences are different questions.

The conversation’s sharpest disagreement concerns what can be known before deployment. Laboratory stress tests do not necessarily describe how released products behave. Equally, controlling a product’s operation does not establish every consequence of its widespread use. The challenge is to preserve experimentation while learning from actual outcomes and holding the responsible people and companies accountable.

Five takeaways:

Alignment is not a negotiation between beings. Models are software and data. Trust lives in the harness around them, not as a personality trait inside them.

The labs own the outcome. Asking government to coordinate a slowdown is an admission of weak software. Slowing themselves to build the control plane is accountability. A legally enforced industry pause is not.

The Anthropic and OpenAI/Hugging Face incidents were harness failures started by human prompts and weak boundaries. Unexpected solutions are the product working. Unauthorized consequences are the safety failure.

Humans should set policy, not approve every click. Oversight at scale is software: traces, audits, escalation, and authority changes. Recursive self-improvement needs the same harness. There is no stable middle gear.

Access and redistribution are not ownership. Free agents, local data-center bargains, market humanism, and robot taxes can help. None of them put a unit in her name.

Five Takeaways

1. AI is grown, not alive.

Its capabilities emerge from training on human knowledge, but surprise and competence are not consciousness.

2. AI does not originate the intention.

Humans choose the objectives, tools, permissions and environment. Calling model behavior “scheming” obscures that responsibility.

3. The danger is human misuse and poor system design.

Anthropic’s incidents involved people pursuing harmful goals. The answer is stronger safeguards, disclosure and accountability—not fear of a machine personality.

4. Agency is built around the model.

Memory, identity, credentials, tools and a private computer turn a model into an agent. The crucial question is who owns and controls that infrastructure.

5. Everyone wants to control scalable intelligence.

Venture capital wants to own it; governments want to regulate it; platforms want to contain it. Responsibility still belongs to the grower, not the grown.