October 20, 2026
TechCrunch reported that Anthropic is having trouble securing enough compute from the same cloud providers it depends on, because those providers increasingly see Anthropic as a rival rather than just a customer. This is happening while Anthropic's infrastructure costs keep climbing ahead of a planned November IPO. That timing is not a coincidence, and it is not a small detail buried in a filing -- it is the whole story.
Every major AI lab has the same structural problem: the companies that sell you GPUs and data-center capacity are also building their own foundation models. Amazon has its own model ambitions, Google obviously does, and Microsoft has leaned on OpenAI while quietly diversifying. Anthropic's bind is just the most visible version of an arrangement that was always going to get awkward once the labs stopped being pure customers and started looking like direct competitors for enterprise AI budgets. Going public while your supply chain has a conflict of interest baked into it is a tough story to tell investors, and I think it is fair to expect this to show up in Anthropic's IPO pricing and in the terms of whatever new compute deals it signs between now and then.
For businesses evaluating AI vendors, the lesson isn't about Anthropic specifically -- it's about concentration risk. If your AI roadmap depends on one lab whose own supply chain is shaky, that's a dependency worth interrogating before you build on it, the same way we'd tell teams to think hard before locking into any single vendor's CRM or automation stack without an exit plan.
While Anthropic's cap table drama (TechCrunch also noted Eric Schmidt holds roughly half the equity stake Dario Amodei does, about 1% of the company) dominates the conversation, SpaceX posted a genuinely striking number: $4.4 billion in net income for the third quarter, with AI cited as the primary driver, according to TechCrunch.
What's notable here is that this isn't a model company burning cash on a growth story -- it's a company using AI as an operating lever inside an existing, capital-intensive business and turning a real profit doing it. That's a useful contrast for anyone tired of AI headlines being dominated by valuation math and IPO speculation. It's a reminder that the most durable AI wins right now may not be the labs themselves, but the companies quietly embedding AI into operations, logistics, and existing revenue lines. That's the same thesis behind why we built workflow automation around AI rather than a chatbot bolted onto a legacy stack -- the money is in the application, not just the model.
Two smaller stories today point at the same trend from different angles. OpenAI added the ability to lock individual ChatGPT conversations, explicitly for moments when you hand your phone to someone else and don't want them stumbling into a sensitive chat. And Yelp rolled out an AI agent that can call restaurants on your behalf to check wait times, part of a broader push into personalized, memory-driven recommendations.
Neither feature is revolutionary on its own, but together they show AI tools quietly taking on two things people used to do themselves: managing privacy boundaries and making phone calls. The ChatGPT lock is a tacit admission that these tools now hold genuinely personal information -- plans, secrets, half-formed ideas -- and that privacy has to be designed for at the conversation level, not just the account level. Yelp's calling agent, meanwhile, is a small but real step toward AI acting as your proxy in the physical world, not just answering questions about it. I'd watch this space closely: once an AI agent can place a phone call convincingly on your behalf, the line between assistant and agent gets a lot blurrier, and businesses on the receiving end of those calls need to start thinking about how they verify who, or what, they're actually talking to.
Where do you draw the line on AI agents acting on your behalf -- checking a wait time feels harmless, but would you trust one to negotiate a contract?
Sources