Big week for lawsuits, big checks, and a model release that's about to make every context window look small. Here's what happened in AI the last day or so, and what I think about it.
Apple Sues OpenAI Over Its Own People
Apple filed suit against OpenAI in federal court in Northern California, claiming trade secret theft. The filing points to more than 400 former Apple employees who now work at OpenAI, mostly folks who used to work on silicon engineering, on-device AI, and hardware design. Apple's argument is that this isn't normal hiring, it's a coordinated effort to pull confidential tech out the back door.
This matters because it's the clearest sign yet that the AI talent war has moved past salary bidding wars and into the courtroom. When a company as buttoned-up as Apple decides to sue instead of just losing engineers quietly, you know they think something real walked out with those people.
My take: I don't know if Apple wins this one, trade secret cases are hard to prove and 400 employees is a lot of people to tie to one conspiracy theory. But I do think this is the first of many. Every big AI lab has been raiding every other big tech company for two years now. Somebody was going to get sued eventually. Won't be the last time either.
SambaNova Pulls In a Billion Bucks
SambaNova closed a $1 billion Series F at an $11 billion valuation, led by General Atlantic with BlackRock, Intel Capital, Qatar Investment Authority, T. Rowe Price, Battery Ventures, Capital Group, and Vista Equity Partners all chipping in. That's on the heels of Together AI raising $800 million just a few weeks back.
Why it matters: this is the AI infrastructure race, not the chatbot race. SambaNova builds chips and systems for running big models, and money like this shows investors still think there's real money to be made in the picks and shovels, not just the flashy apps on top.
Robert's take: a billion dollars at an $11 billion valuation for a chip company most regular folks have never heard of tells you the smart money still thinks compute is the bottleneck, not ideas. I keep seeing people say the AI bubble is about to pop and then another billion dollar round lands like clockwork. Maybe it pops eventually. Not this week though.
Gemini 3.5 Pro Is About to Show Up With a Huge Context Window
Google DeepMind's Gemini 3.5 Pro is set for general availability on July 17, according to a launch plan that leaked out a couple days ago. The headline spec is a 2 million token context window, double anything else in the frontier field right now. The extended reasoning mode, called Deep Think, is getting locked behind the $250 a month Ultra subscription.
This matters because context window size has quietly become the thing that actually changes what you can do with these models day to day. A 2 million token window means you can dump entire codebases or huge document sets in at once instead of chunking everything up and hoping the model remembers what you told it three messages ago.
My take: the context window number is impressive, but locking the good reasoning mode behind a $250 a month tier is going to rub people the wrong way. That's not a consumer price, that's a "we know businesses will pay it" price. Google's betting enterprise customers care more about raw capability than sticker shock, and honestly they're probably right.
ChatGPT Work Wants to Live Inside Your Job
OpenAI shipped ChatGPT Work, an agentic productivity product that folds Codex right into the ChatGPT desktop app, plus a directory of 15 integrations to plug into the rest of your tools.
Why it matters: this is OpenAI stopping being just a chat window and starting to act like actual workplace software, sitting alongside your calendar, your code, and whatever else you run your day on. That's a different product than the one most people started using two years ago.
Robert's take: every AI company is racing to become the thing you never close, the app that sits open all day doing tasks instead of answering one question at a time. ChatGPT Work is OpenAI's clearest move yet in that direction. Whether it actually gets work done or just adds another tab to manage is going to depend a lot on how well those 15 integrations actually work, not on the pitch.