The summer slowdown panic, the labs turn profitable, and a practical AI playbook for leaders

The annual 'AI is cooling off' story is back, but the numbers underneath point the other way. Meanwhile the leading labs are suddenly profitable and racing to go public, and there is a clear, actionable way for leaders to actually get value out of all this.

• The 'AI is slowing down' panic is back for the summer
  My take: Every year around this time a story goes around that AI has hit a wall. This year it is hanging on a few data points: Uber said its heavy AI spending did not translate into more useful features, and the number of new coding-tool installs in one editor flattened out. The catch is that most of that flattening is just people moving to newer tools that the chart does not count, and the actual demand signal, the price of AI compute, is still up sharply because buyers want far more than there is supply for. The honest read is that the free experimentation phase is ending and costs are getting real, not that AI is fizzling. Do not pull back your AI efforts because of the headlines, but do start treating token spend like a real budget line with an ROI attached.
• AI labs turn their first profit and head for the public markets
  My take: Anthropic is projecting its first profitable quarter, which would be the first ever for any major AI lab, at an annualized revenue run rate around 44 billion dollars. OpenAI, close behind on revenue, has engaged bankers and could file IPO paperwork within weeks. For the past year the skeptic argument shifted from 'these companies will never make real money' to 'they can never make money profitably,' and that argument just took a serious hit. The practical takeaway: stop treating your AI vendors like fragile startups that might vanish. They are becoming durable, well-funded infrastructure you can plan a multi-year roadmap around.
• Cheaper models are quietly closing the gap with the expensive ones
  My take: As AI bills climb, the market is responding with much cheaper options that are nearly as good. DeepSeek made its steep discount permanent and is raising a large round, putting a capable model at a fraction of the price of the top US labs. Cursor's new coding model lands just behind the best frontier models on quality while costing 10 to 60 times less. The lesson for your business is simple: you probably do not need the most expensive model for most tasks. Run a few of your real workflows on a cheaper model and see if the quality difference actually matters for what you are doing.
• AI vendors start selling 'guaranteed capacity' like a utility
  My take: OpenAI launched a program letting companies commit to one to three year AI budgets in exchange for discounts and priority access when supply gets tight. This makes buying AI look a lot more like buying cloud computing than buying software subscriptions. The signal underneath is that the people running these companies expect demand to outstrip supply for years, so access itself is becoming something worth locking in. If AI is becoming critical to how your business operates, start thinking about it like electricity or bandwidth: something you plan capacity and budget for in advance, not something you assume is always there at a flat monthly price.
• A practical playbook: the AI 'team members' every leader should set up
  My take: There is a useful framework going around for leaders who feel behind: instead of chasing tools, set up four AI 'roles' that work the way you do. A research analyst you brief properly instead of treating like a search box, a strategic advisor loaded with your real context, a writing partner trained on your actual voice, and an operational assistant that handles your briefings and prep. The point that stuck with me is that the leader's own quality of AI use is the single biggest predictor of how well their whole team adopts it. You cannot do this in 15 minutes a week. Block real time, pick one of these four, and build it properly before moving to the next.
• More automation is creating more human work, not less
  My take: The most AI-native companies report the same surprising thing: the more they automate, the more expert human work there is to do. The reason is that when everyone has the same AI trained on the same past work, the default output becomes generic and samey, which creates demand for the human judgment that makes work actually good. The pattern that works is a human setting the goal, AI doing the heavy lifting, and a human judging and extending the result. For your team, this means AI is not a reason to freeze hiring. It is a reason to invest in people who can frame problems well, exercise judgment, and tell good work from passable output.