Big morning in AI land. OpenAI built its own chip, Anthropic is somehow making money, and there is a fresh fight over who copied who. Let me walk through it.
OpenAI Made Its Own Chip And Named It Jalapeño
OpenAI pulled the wrapper off its first custom chip, the Jalapeño, built alongside Broadcom. The pitch is simple. It is supposed to make running their models about 50 percent cheaper.
Here is why that matters. For years now Nvidia has had the whole industry by the wallet. Every lab pays whatever Nvidia asks because there was no other game in town. If OpenAI can run real workloads on its own silicon, the cost of every answer you get back drops, and so does their dependence on one supplier.
My take. The name made me laugh, but the move is dead serious. Google has done its own chips for years and it shows up in their margins. OpenAI loses a pile of money on every model run, so cutting inference cost in half is not a nice to have, it is survival. I will believe the 50 percent when somebody outside the company measures it, but the direction is right.
Anthropic Is Turning A Profit While OpenAI Bleeds Billions
The reports out today say Anthropic is tracking toward its first operating profit, somewhere around 559 million dollars in the second quarter, on an annualized revenue run rate near 47 billion. Same day, OpenAI is projecting a loss of about 14 billion dollars for 2026.
Why it matters. These are two of the biggest names in the business and they are pointed in opposite directions on the one number that keeps a company alive. A good chunk of that gap is Claude Code, which is reportedly grabbing a big slice of the AI coding market.
My take. Money talks. Everybody loves to argue about benchmarks and who has the smartest model this week, but the company that figures out how to charge more than it spends is the one still standing in five years. Anthropic selling coding tools to developers who happily pay for them looks a lot smarter than burning cash to win a popularity contest.
Anthropic Says Alibaba Ran 28 Million Fake Chats To Copy Claude
Same company, busy day. Anthropic accused Alibaba of running 28.8 million fraudulent exchanges against Claude to pull off what folks call a distillation attack. That is when you hammer somebody else's model with questions and use the answers to train your own. Anthropic sent a letter to the Senate asking for export controls and screening of high volume API traffic.
This matters because distillation is the new version of copying homework, except the homework cost billions of dollars to produce. If a competitor can clone most of your model's behavior for the price of an API bill, the moat you spent years digging gets a lot shallower.
My take. I get a little skeptical any time a company runs to Washington asking for rules that just happen to box out a rival. Maybe Alibaba did exactly what they say. Maybe it is convenient timing. Either way, get used to this fight. The labs spent the cash to build these things and they are not going to sit quiet while folks siphon them off. Expect lawyers and lobbyists, not just engineers.
That is the rundown for today. Chips getting cheaper, one lab making money and another losing it, and a copying fight headed for Congress. Same circus, new tent. Catch you tomorrow.