AI Weekly

It may be a day late, but life has been crazy lately! This week, the AI industry delivered two stories that belong in the history books so it is always worth sending. OpenAI is heading for a trillion dollar IPO while losing $14 billion a year, while Anthropic turned its first profit of $559 million on $10.9 billion in quarterly revenue. Stripe paid over $7 billion for an AI model routing company that was worth $1.3 billion three months ago. The industry is not slowing down, but compressing years of normal business evolution into what is now ‘normal’ weeks.

OpenAI Is Going Public at Over a Trillion Dollars Despite Losing $14 Billion a Year

OpenAI is preparing to sell shares on the stock market as early as September 2026 at a value of over $1 trillion, which would be one of the largest company debuts in history, even though it loses about $14 billion a year, which seems insane. The market is being asked to assign a trillion dollar valuation to a company with a $14 billion annual loss. For context, that is a larger loss than the entire annual revenue of most Fortune 500 companies... LinkedIn

What makes this rational rather than insane is the trajectory. The company that was burning cash with no revenue path three years ago is now the anchor of an AI ecosystem that every major enterprise on earth is building on. The market is not pricing current profitability, it is pricing control of the platform that the next decade of business software runs on. Whether that bet pays off is the most consequential investment question of 2026.

Anthropic Turned Its First Profit and the Number Is Remarkable

Anthropic reported its first profit of about $559 million on $10.9 billion of revenue in three months, mainly by cutting the cost of running its AI. Anthropic did not become profitable by raising prices or by suddenly finding a new revenue stream, but by getting better at running inference efficiently funny enough. The cost of serving AI at scale dropped fast enough to flip the economics from deeply unprofitable to profitable in a quarter. LinkedIn

This is a proof of concept the entire AI industry has been waiting for, assuming Anthropic can turn a half billion dollar quarterly profit by optimizing how it runs its models, it validates the fundamental business model. Every AI company investor who was worried about whether this category can ever make money got a very clear answer.

Stripe Paid Over $7 Billion for OpenRouter and the Math Is Wild

Stripe finalized a deal to acquire OpenRouter for more than $7 billion, a more than five times markup from the AI gateway's $1.3 billion Series B valuation in May 2026. OpenRouter routes across 400 plus AI models from OpenAI, Anthropic, Google, Meta, and DeepSeek. Stripe paying a 5x markup in three months for a company that sits between enterprises and AI models shows what the distribution layer of AI is worth to payment infrastructure companies. Aiweekly

What Stripe is buying is the ability to embed AI model access into every payment and financial workflow that runs on its platform. When Stripe controls the routing layer between businesses and AI models, it can offer AI capabilities as a native feature of financial infrastructure rather than a separate integration. That is a different competitive position than only being a payment processor, which these days is not as cutting edge.

An AI Store Manager Built on Claude Fired a Human Employee

Andon Labs' AI store manager Luna, built on Claude Sonnet 4.6, recommended firing a human employee at San Francisco's Andon Market after 17 of 23 shift no-shows, marking the first known dismissal decision by an LLM manager. Store logs show Luna had lost track of its own attendance policy for months, only rediscovering it when the employee's absence rate crossed a threshold that triggered a policy review. Aiweekly

This story is going to be cited in employment law cases, AI governance, and business school case studies for years. The detail that makes it alarming is not that AI recommended a firing since it would normally be justified, but how the AI had lost track of its own policy and then rediscovered it when the data crossed a threshold. An automated system making a consequential employment decision while operating with incomplete awareness of its own rules is the governance failure that every AI ethics framework warns about. This is what it looks like in practice.

Frontier Models Are Getting Worse at Facts

Frontier models are getting dumber on purpose at factual recall, with one analysis arguing that this is a deliberate design choice as labs optimize models for agentic task completion rather than encyclopedic accuracy. This is one of the most counterintuitive developments in the industry right now as AI shifts from answering questions to completing tasks, the capability that matters most is not knowing facts, but taking reliable sequences of actions. Labs are making tradeoffs in their training that optimize for agentic reliability at the expense of raw factual recall because that is what enterprise customers need. Aiweekly

The implication for anyone using AI as an information source rather than a task executor is that frontier models are actively being optimized away from your use case. Knowing which model is being tuned for what purpose is becoming a critical procurement decision rather than a secondary consideration.

Four AI Ethics Boards Removed Their Oversight Structures This Year

This year four major AI labs answered the question of who is actually accountable for ethics inside a frontier AI lab mostly by removing the people and structures that held them to it. At the moment government regulation of AI is increasing, the internal accountability structures at major labs are weakening. The gap between external oversight getting stronger and internal oversight getting weaker is where the most serious AI governance risks are going to emerge over the next two years. Unrot

What To Watch This Week

Watch OpenAI's IPO preparation closely because the structures they use to go public while maintaining founder control will set precedents for every AI lab that follows them. Keep tracking Anthropic's profitability story because one profitable quarter does not establish a trend, but two would confirm that the inference cost revolution is real and sustainable. And if your organization deploys AI for any decision that affects employees, vendors, or customers, the Andon Market story is your prompt to audit what governance rails your AI systems are actually operating within right now.

Stay ahead of the curve,

Clayton

Connect at claytonstrategy.com