Can OpenAI and Anthropic Justify Their Valuations? + Flock, Privacy, and Plummeting Crime

Scott Galloway (0:01)

Support for the show comes from Apollo. The world's leading companies don't need off-the-shelf financing solutions, that's where Apollo comes in. Apollo provides access to long-duration capital and a broad range of financing capabilities, helping companies develop customized financing solutions designed around their strategic objectives and structured at an efficient cost of capital. Learn more at apollo.com/capital.

Support for the show comes from Yahoo! Scout. Whatever you're searching for, Yahoo! Scout can turn hours of research into clear, trustworthy answers in seconds, keeping up with culture, navigating the ups and downs of parenting, planning a vacation and more. Make sense of life's everyday questions and personal decisions with Yahoo! Scout. Yahoo!'s personal finance page has been my homepage for about 30 years, and so far, I find the user interface from Yahoo! Scout really elegant, and I find the results to be really helpful, quite frankly. Go to scout.com to start exploring today.

SPEAKER_2 (1:00)

This monday.com ad was created by a team of people and AI agents. Reese, our content agent, wrote the copy based on our best practices, like mentioning monday.com three times. That was the second. Johnny, our coordination agent, built the timeline and kept everyone aligned. Olivia, our human creative director, stayed in the loop because agents are great, but they don't always know when a joke lands. She had one note. Tell listeners it only takes minutes to build an agent. So minutes. Create your own AI agent today on monday.com.

Scott Galloway (1:37)

Welcome to Office Hours with Prof G, where we answer your questions about business, big tech entrepreneurship and whatever else is on your mind. Do you have a question? Send us a voice recording at officehoursforprofgmedia.com or post your question on the Scott Galloway subreddit, and we just might feature it in an upcoming episode.

Question number one comes from Fred, who texted us. Dear Prof G, how can Frontier AI Labs sustain a subscription-based revenue model when free local AI is getting so powerful? Every week, the setup process for local AI gets simpler and the local models get more powerful. I've cut my spend on clod code by 50 percent using local AI. Plus, local AI solves some compliance issues for enterprises. So, what's the future of AI's business model? So, first off, thanks for the question, what does local AI mean? It's essentially open-weight models make their core components publicly available, allowing users to download, customize, and run them on their own hardware. This can lower costs and keep sensitive data from being sent back to a model provider to help it learn and do a better job with other people's queries. The idea is you own your data and you're not using it to train. It doesn't jump the firewall and get absorbed or strip mined by one of the frontier LLMs. The trade-off is that providers have less ability to monitor misuse, restrict access, or quickly deploy safeguards. And some, there's some enterprise-wide benefits protecting you from cybercrime and ensuring certain debugging that you don't get with your own internal, almost like call it desktop, open-weight AI. Open models are getting more competitive. In July, the U.K.'s AI Security Institute tested leading open-weight models against frontier models, released four to seven months earlier. It found that the gap in their cybersecurity performance had narrowed since 2025

The open models were also substantially cheaper. On tasks both models completed reliably, GLM 5.2 cost $6 per task versus $15 for Claude Opus 4.6.

DeepSeek version 4 Pro cost about $0.30 versus $12.50 for Opus 4.5. Many of the strongest open-weight models now come from Chinese developers, including DeepSeek and Moonshot. In July, Nvidia, Microsoft Meta, and 20-plus other companies published an open letter warning Washington against premature restrictions on open-weight AI models. Consumer subscriptions remain a small market. It's all about the enterprise. Bank of America found that only about 3% of US households paid for an AI subscription as of February. Adoption was highest among households earning more than $125,000, basically the domain of businesses and rich households. OpenAI reported 50 million paid subscribers in June, just as estimates report the company hit 1 billion users around the same time. This means only about 5% were paying.

So the reason why Anthropic has accelerated ahead of ChatGPT, kind of the initial gangster from a valuation standpoint, is because of that. It's because the larger opportunity appears to be an enterprise. In December of last year, OpenAI claimed they had hit more than a million business enterprise customers. Anthropic reportedly tells investors that its local addressable market is more than 30 trillion. Okay, that number is meaningless. Plus, Anthropic has cited gross margins above 80%, but that excluded some major expenses, including model training costs and revenue share with partners. It's sort of like WeWork adjusted EBITDA.