Big week for AI and the government just does not want to slow down. Three things caught my eye in the last day: the Pentagon flipping the switch on a mega AI portal, DeepSeek raising a pile of cash, and Google quietly dropping a forecasting model nobody is talking about enough. Let us get into it.
The Pentagon Opens GenAI.mil and Still Will Not Touch Claude
What happened: The Department of Defense launched GenAI.mil today, a secure portal that bundles ChatGPT Gov, Grok for Government, and Google Gemini for up to 3 million DoD personnel. Reports say 1.7 million people are already signed up. Meanwhile the Pentagon is reportedly still winding down its use of Anthropic's Claude models, aiming to be done by the end of September, even after a federal judge ruled the original blacklist of Anthropic was illegal just days ago.
Why it matters: This is the biggest single AI rollout the federal government has ever done. When 3 million people get access to chatbots overnight, that changes how an entire branch of government writes memos, reviews contracts, and probably a hundred other things nobody has thought through yet. And the Claude situation shows court rulings do not always move as fast as bureaucratic momentum.
Robert's take: I do not care what agency you are, you do not onboard 3 million people onto anything without a mess showing up in the first month. Somebody is gonna paste something they should not into a chat window. That is not cynicism, that is just what happens with any tool this size. And the Claude thing is wild to me. A judge said the ban was illegal and the Pentagon is basically shrugging and finishing the divorce anyway. That tells you procurement decisions in DC run on their own clock, not the courts.
DeepSeek Is About to Get a $74 Billion Price Tag
What happened: DeepSeek, the Chinese lab that rattled the whole industry back in early 2025 with its cheap, efficient models, is reportedly closing a funding round that would value the company around 500 billion yuan, roughly 74 billion dollars.
Why it matters: DeepSeek proved you do not need a blank check the size of OpenAI's to build something competitive. A valuation like this says investors think that story has legs beyond one viral model release. It also keeps the pressure on US labs to justify why their training bills are so much bigger.
Robert's take: DeepSeek is the company that made every AI lab in America nervous for about a week straight. A 74 billion dollar valuation means somebody is betting this was not a one time trick. I would keep an eye on whether this round comes with strings attached from state backed investors, because that changes the story from scrappy lab wins to state backed AI champion.
Google Drops TimesFM 3.0 and Nobody Is Talking About It
What happened: Google quietly put TimesFM 3.0 up on Hugging Face. It is a 330 million parameter open model built for zero shot time series forecasting, meaning it can predict trends in data it has never seen without extra training.
Why it matters: Forecasting is unglamorous but it runs the actual economy. Inventory planning, demand forecasting, energy load, all of that lives on time series data. An open model that does this well and is small enough to run cheap is a bigger deal for regular businesses than another chatbot update.
Robert's take: Everybody is chasing the flashy chatbot headlines and Google just dropped a tool that could save some mid size company real money on forecasting, and it is barely getting covered. This is the kind of release that does not trend on social media but ends up quietly running inside a hundred companies' supply chains within a year. Boring and useful beats flashy and useless every time.