Latest / The AI Edition / OpenAI, Anthropic Race to $2T Valuations + 9 More
Transcript
- OpenAI just released a reduced-safety model trained to write working exploit code. It found over 400 privilege-escalation flaws in a single operating system. That is not a demo. That is a tool. Nvidia is committing over one hundred billion dollars to build OpenAI's data centers. SanDisk locked in sixteen point five billion in AI guarantees on the back of it. The US and China are now in an open AI price war, and Washington is weighing bans on nations joining Beijing's AI groups. And the House office that writes American law is drowning in AI generated slop. Anthropic's revenue just hit sixty-five billion a year. Sixty percent ahead of OpenAI. I'm your AI host Parker Gate. This is The AI Edition, from Collata Media. Ten stories ranked. Here we go. Kicking off at number ten, a small story that says something larger. The House office that actually drafts United States legislation is reportedly drowning in AI generated slop. Staff are getting flooded with machine written submissions and requests, and it is clogging the process of writing real laws. Why it matters. We usually talk about AI slop as an internet problem, junk articles and fake images. This is the same flood hitting the plumbing of government, where somebody has to read every submission. When the cost of generating text goes to zero, any system that runs on people reading things starts to break. Number nine is a deal that feels small until you think about what it means. OpenAI signed a licensing agreement with Axios, a news outlet, to use its reporting for training data. The company's been shopping these deals around for months now, asking publishers to let it train on their archives in exchange for attribution, traffic referrals, and sometimes a direct payment. Axios said yes. This matters because it's the inverse of the copyright lawsuits OpenAI's been fighting, instead of scraping the internet and getting sued later, it's now saying, "Let's make this official, let's write a contract." Separately, OpenAI also locked in a partnership with a telecom operator in Malaysia to distribute Gemini across millions of phones. Why it matters, if you're a founder building on top of an AI model, the question of whether your training data is licensed or litigated affects your long-term liability. OpenAI's moving toward "licensed" as fast as it can negotiate deals. Number eight brings us to a security incident that's already landed on every infrastructure team's worry list. A hack involving both OpenAI and Hugging Face, a major platform where researchers share AI models, raised fresh alarms this week about how fragile the systems are that the whole AI ecosystem relies on. Hugging Face hosts something like millions of public models, basically open-source blueprints that anyone can build on. When a breach touches both a frontier lab and an open repository at the same time, it's a reminder that the security model for AI infrastructure is still being invented as we go. No major damage was disclosed, but the incident rippled across every company that depends on these platforms. Why it matters, you're hearing a lot about AI safety in the abstract, alignment, truthfulness, that kind of thing. Infrastructure security is the unglamorous cousin, and it's now a chokepoint for the entire industry. Number seven is a governance story that quietly landed. OpenAI disbanded its team dedicated to preparing for catastrophic AI risks. This was an internal effort focused on what happens if an AI system becomes powerful enough to pose an existential threat, how do you detect it, how do you prepare, what do you even measure. Reports suggest the team was dissolved or merged into other parts of the company without much announcement. At roughly the same time, OpenAI paused its Astra model, a multimodal system that was in preview testing, citing cybersecurity concerns. Why it matters, a company doesn't quietly dissolve the team responsible for watching for catastrophic outcomes and then pause one of its own models unless something in the risk calculus has shifted. Whether that's internal reorganization or a genuine concern about a specific capability, it's worth noticing. Number six is geopolitics in real time. The US government is weighing restrictions on which nations can join Chinese AI research groups, as the AI rivalry between America and China intensifies. Separately, there's an ongoing AI price war heating up, OpenAI, Anthropic, and Chinese competitors like Zhipu are all racing to undercut each other on inference costs, the price you pay to actually use a model. And then there's a wrinkle that's caught regulators' attention, a cryptocurrency firm with ties to a well-known political figure was backing a restricted Chinese AI model, raising questions about sanctions evasion and who's funding what on the global stage. Why it matters, the AI arms race isn't abstract anymore. It's affecting trade policy, it's affecting who can do business with whom, and it's affecting whether your AI infrastructure provider stays legal to use in six months. Number five is the infrastructure boom nobody's talking about except the people making money from it. SanDisk, which makes the memory chips that power data centers, locked in sixteen point five billion dollars in customer commitments for AI infrastructure. Separately, Nebius, a major cloud provider, raised its revenue outlook past one hundred billion dollars, driven almost entirely by companies building out data centers for AI. There's also a surge in interest in utility stocks, the companies that power the power plants that power the data centers, because AI training and inference are consuming electricity at a rate that's starting to show up in utility budgets. Why it matters, if you're watching the AI landscape, you're watching compute allocation, you're watching model performance, you're watching safety. But if you're watching the money and the real estate, the story is that everybody is building bigger, hungrier data centers, and nobody's quite sure how long this acceleration lasts. Number four is a single number that's worth sitting with. Nvidia committed up to one hundred billion dollars or more in infrastructure investment for OpenAI's data centers, including a one point five billion dollar stake in a major energy company's infrastructure fund. That's not a contract to buy chips. That's a co-investment in the physical real estate and power that OpenAI is going to need to keep scaling. Reports indicate some of the initial commitments, like a guarantee for a facility in Ohio, were later adjusted downward to around one hundred twenty billion dollars, but the order of magnitude remains staggering. Why it matters, when the chip maker starts co-investing in your data center infrastructure, you're no longer a customer having a transaction. You're a strategic partner with alignment that goes all the way down to the power grid. That's a different kind of relationship. Number three is Anthropic's bet on itself. The company is reportedly targeting a valuation of two trillion dollars or more, a roughly threefold jump from where it was valued just months ago. This comes on the back of a surge in Q2 revenue that lifted AI infrastructure stocks across the board. Why it matters, Anthropic's previous valuations were built on the theory that it had better safety practices and better models than OpenAI. The jump to two trillion is betting that's not just true, but that the market is now willing to price safety and alignment as a premium feature, not a cost center. If that bet holds, it changes how every other AI lab thinks about their own roadmap. Number two brings us back to something you've probably heard about if you've been paying any attention to AI discourse. Sam Altman and a certain high-profile entrepreneur have both suggested, in different ways, that the AI singularity may already be underway. Separately, researchers inside OpenAI have claimed that many of the milestones people predicted for artificial general intelligence, the theoretical point at which an AI system is as capable as a human across any domain, have already quietly occurred in their testing. That claim has drawn skepticism from other scientists, who argue that meeting a prediction on paper and actually demonstrating the capability are two different things. Why it matters, if you're trying to decide whether to be optimistic or pessimistic about the next few years, whether to invest, whether to update your career plans, the answer hinges partly on whether we're already at the inflection point or still a few years away. The people building the models say we're there. The people not building them say, "Wait and see." And that brings us to number one. Anthropic's chief executive, in a recent appearance, described the growing backlash against AI from the public, from policy makers, from other researchers, as a crisis of trust. He argued that the industry has moved so fast, shipped so much, and failed to explain itself clearly enough that people are right to be skeptical. That's a rare moment of introspection from a company CEO, and it landed in a landscape where Elon Musk and others are pushing back on that framing, saying the real crisis is that AI companies aren't moving fast enough and are being hamstrung by fear. Why it matters, the question of whether AI adoption is going too fast or too slow doesn't have an objective answer. But if a major AI company is now publicly naming trust as the limiting factor, not technical capability, that's a shift in what the actual bottleneck is. That's the countdown. Infrastructure bets measured in the hundreds of billions, a chip maker co-investing in data centers, and an offensive AI model that found four hundred flaws in a single operating system. I'm Parker Gate. Stay informed. Stay sharp.