Three things crossed my desk this morning and all three say the same thing. AI quit being a demo and started being a line item. Washington is using it, Anthropic is getting rich off it, and the chip wars under the hood are heating up. Here is where I land on each.
Uncle Sam Hands the Auditors a Chatbot
The Department of Health and Human Services said it is going to feed the annual audit reports from all 50 states into ChatGPT and other AI tools, looking for fraud, waste, and abuse in federal health spending. The program is run by Assistant Secretary Gustav Chiarello, and they have already given every governor and treasurer a heads up that it is happening.
Why it matters: nobody reads those audit reports. I mean it. They are giant, they pile up, and no human team on earth has the hours to read all 50 cover to cover every year. Pointing a model at that haystack is the kind of grunt work AI is actually good at.
My take: I like the idea and I do not trust the rollout. A model that hallucinates a number is annoying when it writes your fantasy football recap. It is a different animal when the output is a flag that says your state misspent federal money. Read the audits, sure. But a human signs the accusation, every time. No exceptions.
Anthropic Went From $9B to $47B in Five Months
Anthropic's annualized revenue hit $47 billion in May, up from roughly $9 billion at the end of 2025. That is about five times bigger in five months. Numbers like that do not happen in software very often, and when they do everybody starts asking how long it can last.
Why it matters: this is the proof that companies are paying real money for AI work, not just kicking the tires. You do not back into a run rate like that on free trials. Somebody is putting Claude into production and writing checks.
My take: the revenue is loud and the costs are louder. Anthropic decided not to build its own data centers and is buying compute instead, including up to 5 gigawatts from Amazon and a reported $1.25 billion a month from xAI's Colossus. That is a smart move when you want to grow fast and not get stuck owning concrete. It is a scary move if the music ever stops. Big top line, bigger appetite. I am watching the gap between the two.
Claude Might Soon Run on Microsoft's Own Chips
Anthropic is in early talks with Microsoft to run Claude inference on Microsoft's custom Maia 200 chips through Azure. The Maia 200 launched in January on TSMC's 3nm process, it is built specifically for inference, and Microsoft claims more than 30 percent better performance per dollar than the rival silicon.
Why it matters: inference cost is the whole ballgame. Training grabs the headlines, but answering the millions of everyday questions is where the money actually burns. Shaving 30 percent off that bill changes what you can charge and still keep the lights on.
My take: this is Anthropic refusing to marry one supplier. Amazon, xAI, and now maybe Microsoft. That is three horses in the race and a lot less leverage handed to any one of them. Smart. The less you depend on a single chip vendor, the better you sleep. I would do the exact same thing.
Bottom line: the demo era is over. The bills, the revenue, and the silicon are all real now. That is a better problem to have than wondering if any of this was going to work.