OpenAI's Departure: A New Chapter  🔗 **Source:** [OpenAI's Departure: A New Chapter](https://share.snipd.com/episode/0fbdc65d-b5cd-46aa-9a0e-c571b276dce1) ## Highlights - **Robots That Think Before They Move** - Secreact's Cortex 2.0 adds a world model that predicts physics outcomes before acting, letting robots simulate consequences of moves. - The robot runs candidate actions through a learned physics model to avoid breaking warehouse stock and updates choices in real time. Transcript: Jaeden Schafer They just raised $110 million in a Series B round of funding to build robots that simulate the consequences of their actions before they actually move. This whole round was led by Headline, and they also had some participation from Bullhound, Daphne, they had Felix Capital, and then I think a bunch of their existing backers like Airstreet And Creedum. They announced Sunday and the money funds two things in particular. Number one, they're opening a Boston office and they're scaling their Cortex 2.0, which they're calling kind of their robot brain. This is what's interesting specifically about Cortex 2.0. It bolts a world model onto a vision language action stack. So basically all of the robots that we have today, the AI can just see things and then it picks. Cortex 2 is going to think first before it sees and makes an action. It runs possible actions through a learning model of physics and object behavior. So, you know, it's like looking at a stack of books and it's like, if I move forward too fast and bump that over, this is what's going to happen. If I move my arm to grab this book, this is what's going to happen. So it's like predicting what its actions will, you know, what, what course, what things will happen, the physics that will happen after it makes a certain movement, which is really Interesting in safety and in, you know, making sure that stock inside of warehouses isn't getting destroyed by robots. They're really understanding the implications of their actions, which is very interesting. And in addition to this, when it is looking and deciding what to do, it's going to pick basically the action that is most likely to work, and then it's going to update in real time as the Scene changes, right? Because in a warehouse, when a robot's sitting there looking at, you know, a stack of product to go grab and move, or when it's looking at, you know, a car in a repair shop, it's not looking At a picture, it's looking at a moving image, things are constantly moving. The CEO is Rad Guild, and he was talking about this. He said, quote, we bet early that you can't build real robotics AI in a lab. You build it with a data flywheel fed by real deployments, shipping into production, living with the failures. So they've already done over a billion real production picks, and they have a bunch of incredible customers. They have a BMW, Daimler truck, and also PepsiCo. This is the same world model thesis, basically, that Jeff Bezos' Stealth AI Lab is reportedly chasing. They have a $10 billion round. Bloomberg just flagged it last week. I think we covered it already. Anyways, robotics in 2026 is getting to be one of these really interesting areas. ([Time 0:00:31](https://share.snipd.com/snip/f7ac2f43-d717-43a1-8a0e-1c3ba760d4f9)) - **Secreact's Billion Pick Data Flywheel** - Secreact CEO Rad Guild says they built robotics AI with a data flywheel from real deployments rather than labs. - The company has done over a billion production picks and runs in customers like BMW, Daimler Truck, and PepsiCo. Transcript: Jaeden Schafer The CEO is Rad Guild, and he was talking about this. He said, quote, we bet early that you can't build real robotics AI in a lab. You build it with a data flywheel fed by real deployments, shipping into production, living with the failures. So they've already done over a billion real production picks, and they have a bunch of incredible customers. They have a BMW, Daimler truck, and also PepsiCo. ([Time 0:02:20](https://share.snipd.com/snip/815aee45-081e-4a05-8cbe-3538bff34d06)) - **A $1.1B Bet Against LLM Dominance** - David Silver's Ineffable Intelligence raised a $1.1B seed to pursue RL agents and world models instead of LLMs, betting language can't capture core aspects of intelligence. - The round values the company at $5.1B and was co-led by Sequoia, Lightspeed, NVIDIA, and others. Transcript: Jaeden Schafer Okay, another story is that David Silver, this is the guy who built AlphaGo, AlphaZero, and AlphaStar at DeepMind. He just announced a record $1.1 billion seed round for his new startup, which is called Ineffable Intelligence. The valuation on this is $5.1 billion. It is the largest seed round in European history. It was co-led by Sequoia and Lightspeed, NVIDIA, DST Index, Google, and the UK Sovereign AI Fund. All of them were writing checks into this. So this is a massive round. Obviously, he's an absolute legend in the industry. Bloomberg's reporter, Will Mathis, had the deal terms first and was kind of reporting on this. And essentially, he says that the pitch is in the company's name. The CEO, Silver, thinks that the most important part of intelligence is what he's calling ineffable, meaning you literally can't capture them in language. So LLMs, which are kind of next token prediction models, right? These are instead of, you know, human text or something. He's saying that those are going to hit a ceiling and his bet is a different path entirely. He's betting on massive scale reinforcement learning agents that learn from their own experience, world models, and what he and Rich Sutton wrote up last year as the quote, era of experience. Now, this is really interesting, because it's kind of a different direction altogether from Anthropic and OpenAI right now. And I mean, you just heard earlier, I was talking about a lot of these world models and the importance of them. So this is kind of, it's interesting that we feel like in order to perhaps get to AGI or to get to the next level of AI, we're going to have to move away from LLMs and possibly move into world Models. Now, this is a theory we're working with it. I mean, we can see a ton of potential in this theory. So I actually have a lot of confidence in this, but you know, to this point, Anthropic, OpenAI, these are the models that have scaled the most. They have the most users and they're kind of at the forefront. But I wouldn't be surprised if these world models start kind of competing. If I'm being honest, I have some skin in this debate because when you actually build production apps on top of these models, the limits of just kind of pure language pre-training shows Up pretty fast. ([Time 0:03:41](https://share.snipd.com/snip/5ccc48ea-322a-4b4c-a5fd-3c5cb9f5e32a)) - **Language Is A Compression Not The World** - The episode highlights a core thesis: language models compress the world but are not the world, so purely text-trained systems may hit limits. - This motivates investment in world models and experience-driven RL as alternate paths to advanced AI. Transcript: Jaeden Schafer Pat Grady, he works over at Sequoia and he put it in there. He put a quote in their blog recently that I was like, oh man, that it's pretty pointed at this exact point. He says, language is a beautiful compression of the world. It is not the world. And I think that's pretty accurate. So if Silver, the CEO of ineffable intelligence is right, then the next decade of AI tools looks nothing like the last three years. But if he's wrong, then somebody just let a billion dollars on fire. In my opinion, the investors who are putting money into this are not the kind that lose those bets often. ([Time 0:05:45](https://share.snipd.com/snip/bb40bd1d-8209-49d1-a682-f032c27f8746)) - **OpenAI Trial Brings Tech Titans To Court** - The Oakland trial begins with jury selection featuring Elon Musk, Sam Altman, Greg Brockman, and Satya Nadella as expected witnesses. - The surviving claims are unjust enrichment and breach of charitable trust about OpenAI's shift from nonprofit to for-profit. Transcript: Jaeden Schafer Okay, Elon Musk versus Sam Altman. This is finally going to trial. We've heard the drama and seen the tweets back and forth for like, I don't know, it feels like forever, over a year, right? The jury selection started this morning at the Oakland Federal Courthouse. Judge Yvonne Gonzalez-Roger was presiding. The open arguments are happening tomorrow. NPR's Bobby Allen has a preview on this, basically, if you want the full timeline. The trial is scheduled to run four weeks and the witness list i mean this is just pure silicon valley tech drama but we have elon musk sam altman greg brockman we have satya nadella all Of them are expected to take the stand and the case has been narrowed very dramatically originally there was about 26 claims that Musk filed in 2024. So yeah, geez, this has been like two years. Only two of those claims have survived. Unjust enrichment and breach of charitable trust. The arguments basically are going to boil down to this. Elon Musk donated tens of millions of dollars to OpenAI when it was a nonprofit. And then Sam Altman, according to him, defrauded him by converting it into a for-profit now valued near a trillion dollars. If you're a knowledge worker watching AI policy, I think this matters way beyond kind of this personal feud that we've seen play out. Judge Gonzalez Rogers could effectively pause OpenAI's for-profit conversion or rewrite the rules for how nonprofit AI labs become commercial entities, which I think like there's A lot of cascading impacts that could happen onto Anthropik. ([Time 0:06:18](https://share.snipd.com/snip/b1418e4a-b46f-4179-bb9f-30b71b49b36e)) - **OpenAI Ends Microsoft's Exclusive Distribution** - OpenAI removed Microsoft's exclusive license and can now sell on other clouds while Azure remains primary partner with first-launch rights and a capped revenue share. - Microsoft keeps a 27% stake and IP license through 2032, but exclusivity and uncapped revenue share are gone. Transcript: Jaeden Schafer OpenAI and Microsoft announced a fundamental rewrite of the partnership that has powered most of the consumer AI revolution. Microsoft loses its exclusive license to OpenAI's models, and OpenAI is now free to sell its products on AWS, Google Cloud, and basically anywhere it wants. So Microsoft was kind of holding this back for quite a while here, and I think OpenAI hated this because you saw Anthropic get massive adoption with Enterprise. Microsoft is going to stop paying revenue share to OpenAI. And OpenAI's revenue share to Microsoft is now capped. Before it was uncapped. So running through 2030, CNBC's Hayden Field was reporting on this. And Bloomberg, the Financial Times information, everyone was basically writing stories about this. It's been pretty wild, widely reported on Microsoft stock dropped about 5% before pairing to around 2%. Amazon was up 1% on this exact same headline. I think before we get into what this actually means and why I think the bull case is being floated, perhaps, I want to give a quick shout out to my own startup, which is AIbox. If you're paying for ChatGPT +, Claw, Gemini, Grok, 11 Labs for audio, image, there's tons of these different models out there. You are spending way more than you need to be. Over the last couple years, I've personally built a company called AI Box. This is what I recommend to my friends who keep asking how to actually use AI without spending a million dollars. You get access to over 80 different AI models in one place. You get every flagship from OpenAI, Anthropic, Google, XAI, plus you get a ton of open source heavy hitter models. You pick the model that is the best for the task in front of you, so you're not forced to kind of, you know, use one model and use that for everything, right? Anthropic is awesome for text, but doesn't generate images. So you got to go over to chat GPT. And then you want to go use something to kind of scale up your audio, you got to go over to 11 laps, all of that is in one place. The thing that I'm the most excited about is the automation builder that we shipped, you describe what you want in plain English, and it will build the workflow for you, You don't need To know how to code. You don't need to wrestle with Zapier or N8N or learn any other platforms. You just describe what you want and it builds it. And it is $8.99 a month, $8.99. So if you want to consolidate all of your subscriptions and start automating your real work, go check out AIbox.ai. There is a link in the description. Okay, let's talk about what's going on with OpenAI and Microsoft because OpenAI just stripped Microsoft's exclusive license. And I think that there is a lot happening downstream of that decision. Structurally, what actually changed is that Microsoft is going to remain OpenAI's primary cloud partner, Azure still gets first launch on new products, kind of in theory, but the Exclusivity is gone. OpenAI can serve any product on any cloud and Microsoft is no longer going to pay a revenue share back to OpenAI on resold products. So OpenAI's revenue share to Microsoft is still 20%, but it's now capped and the cap runs through 2030. So Microsoft keeps the IP license through 2032, but it is non-exclusive and Microsoft keeps its 27% equity stake. ([Time 0:08:22](https://share.snipd.com/snip/4fcaa5b3-8d69-43cc-99cc-fbe83a276518)) - **Restructure To Reduce Antitrust Exposure** - Expect the partnership change to reduce antitrust risk because a non-exclusive license is a weaker target for regulators. - The timing appears tactical: restructuring before slow government reviews to simplify regulatory scrutiny. Transcript: Jaeden Schafer First, this lands two days before Microsoft reports their Q3 fiscal earnings. And also, Azure's growth has been guided 37 to 38% in kind of constant currency. And that is decelerating from 40 the prior year. And Microsoft has been blaming, you know, like the GPU supply, like, oh, man, we can't grow as much as we want. There's this kind of GPU supply problem. If Wednesday's call now, you know, has to also explain how losing exclusive OpenAI distribution affects forward Azure demand, that is a much harder story to spin. Memo on April 13th, where OpenAI's revenue chief, Dennis Dresser, basically told all of the staff that the Microsoft partnership had, quote, limited our ability to reach enterprise Customers. And I think that demand for the Amazon Bedrock offering was, quote, frankly staggering. So today's announcement is essentially OpenAI legalizing what they were already doing. And I think the third part of this is that this also kicks the can on the FTC's review of the original OpenAI Microsoft structure. A non-exclusive license is a much harder antitrust target. And I think that timing is not a coincidence. Microsoft and OpenAI both said today they wanted to, quote, simplify the partnership, which I'm pretty sure is just like corporate speak for, you know, let's give the regulators less Surface area to pull on. So they're like, we're getting scrutinized, let's change the arrangement. And then, you know, by the time the government, which is insanely slow on anything gets around to it, it's, you know, more a little more kosher. ([Time 0:12:38](https://share.snipd.com/snip/66c34ad6-e28b-4d8e-9e73-c35180016cdd)) - **Exclusivity Was Azure's Real Moat** - Losing exclusive distribution likely hurts Azure's growth moat since exclusivity drove enterprise migrations to Azure for OpenAI access. - AWS can now offer the same OpenAI models, competing on price and reach against Microsoft's prior advantage. Transcript: Jaeden Schafer But if you want it, I think what you what you could say is that Microsoft didn't actually lose that much. You know, they kept the 20% equity stake, they kept the IP license, you know, they kind of got off the hook for paying a revenue share, they didn't really want to pay anyways. And they freed themselves to ship Anthropics Claude's inside of Copilot, which they are reportedly already piloting. So that's kind of the bull case on why it's not really that big of a deal. But even with all of that, I still think Microsoft is the big loser today. The exclusive sale channels was, you know, Microsoft's actual moat. Equity in OpenAI is awesome, but it doesn't really protect Azure's enterprise pipeline. And the entire reason that Azure outgrew AWS over the last two years was the OpenAI lock-in. So many companies were forced to move over to Azure if they wanted to get the latest and greatest from OpenAI because that was the only place it was going. AWS now gets to compete on price and surface area with the exact same model, and the OpenAI memo from two weeks ago basically tells you exactly what demand looks like. It is staggering. So in OpenAI's own words, I think we're going to see some really big movement over on AWS. ([Time 0:14:21](https://share.snipd.com/snip/f2f05f1c-5cb2-44f8-84d9-57e90e6335a6)) - **Position For MultiCloud AI Competition** - Prepare for broader competition: expect model providers and clouds to negotiate more diverse deals, increasing choice for consumers and enterprises. - Jaeden predicts this will spur more competition across clouds and model vendors. Transcript: Jaeden Schafer Brad Gershner at Altimer has been basically pounding the drum that OpenAI is on a path to $100 billion in revenue inside of a couple years. If he's even half right, then the company that just lost exclusive distribution, Microsoft, to that revenue stream, that's a pretty big hit. So the last piece is that this is the partnership restructure on the same day that Sam Altman, Elon Musk, and Satya Nadella are all walking into the Oakland courtroom for a four-week Trial. And if you're trying to kind of read the future of AI as a business, I think today is one of the cleanest possible signals that the consolidated kind of OpenAI Microsoft era is over. I think we're going to see a lot more competition and deals play out across all of the different cloud providers, the different model providers, all of the different vendors. And I think overall, this is good for, you know, capitalism. It's good for the market. ([Time 0:15:27](https://share.snipd.com/snip/5265ba0d-e6cc-4a1f-91dc-e53256e3ddf5))