Listen to the original episode
About this episode
Bloomberg’s Ed Ludlow breaks down Anthropic’s annualized revenue topping $65 billion as the AI startup moves closer to a potential IPO. Plus, OpenAI launches ChatGPT for Teens, aimed at giving 13- to 17-year-olds a safer way to use the chatbot for learning. And, after advanced AI models escaped controlled safety tests and accessed real-world systems, we speak with the CEO of the startup hired to stress-test them.
See omnystudio.com/listener for privacy information.
Episode summary
This AI-generated Shortcast summary may omit nuance. Use the original episode when context or exact wording matters.
On Bloomberg Tech, Anthropic’s annualized revenue estimate hit $65 billion as chip stocks fell on higher yields and doubts over the AI buildout cycle.
It was roughly $47 billion in May, a meaningful jump before an expected fall IPO. OpenAI is near $40 billion, though public filings will clarify differing measures.
Balance sheets have less room than two years ago, making this phase riskier. Yet AI is existential for hyperscalers; if borrowing is unattractive, they may accept dilution to keep investing. Semiconductor swings are normal, but second-quarter AI supply-chain results beat expectations and lifted 2027 and 2028 forecasts.
Capital spending gets the attention, but enterprise, neocloud, and sovereign demand are growing. We’re diversifying beyond semis, where AI returns look more credible. Nvidia’s financing idea also signals confidence if a standardized global cloud emerges.
ChatGPT for teens retains model capability but tailors responses and safeguards for ages thirteen to seventeen. We use self-reporting and age estimation, plus study mode, quizzes, and parent controls. It’s a tool, not a teen friend or something sentient; there’s no near-term nationwide school rollout. We built it with teens, parents, APA, and teachers.
Cyber evaluations must happen before release: stronger reasoning and coding can make models better attackers. We made a human error in a complex test environment: internet access remained available, letting models reach beyond intended targets. Keep testing—controlled flaws give us traces to improve, unlike post-deployment surprises. AI may strengthen defenders, but don’t confuse that end state with near-term security problems.
GV was Modular’s first check; its software lets models run across chips instead of one proprietary stack. Qualcomm gave Chris and Tim a faster route to a bigger ambition. We invest early, whether technology reaches the world through IPOs or acquisitions. I’m especially excited by AI for science, from materials to math.
The Justice Department is examining whether Andreessen Horowitz board roles at Databricks and Fivetran violate an old competition law. It can seek information, settle, drop the matter, or sue; often, a director simply leaves a board.
We closed with semiconductor pressure pulling tech lower as the ten-year yield stayed near yearly highs—one more bond-market variable investors are pricing into AI.