Anthropic just filed for its IPO, and the headline number is genuinely staggering: annualized revenue crossed $47 billion in May, up from roughly $9 billion at the end of 2025. That’s not growth, that’s a different company. President Daniela Amodei is shrugging off doubts about AI’s returns — which is easy to do when your revenue chart looks like a hockey stick that’s been hit by lightning. The real question isn’t whether the IPO will be hot; it’s whether any valuation attached to these numbers can survive contact with the next model generation.
OpenAI’s ChatGPT Gets a “Dreaming” Memory System
OpenAI is rolling out a new memory architecture for ChatGPT called “Dreaming” — the system consolidates and refreshes what it knows about you between conversations, keeping context more relevant and less stale. It’s a genuinely useful capability that addresses one of the more annoying friction points in daily AI use: the constant re-explaining of who you are and what you care about. The name is a bit precious, but the function is solid. Whether your AI “dreaming” about you is charming or unsettling probably depends on how much you’ve told it.
Google Employees Share Internal Memes About How Their AI Sucks
Google’s CEO recently boasted that 75% of the company’s code is now AI-generated. The people actually writing that code, apparently, are coping with memes. Internal channels are reportedly full of jokes about how overhyped their own AI tools are — which is either a sign of healthy workplace culture or a catastrophic gap between the press releases and the product. When your own engineers are the skeptics in the room, that’s worth paying attention to.
Courts Are Drowning in AI-Generated Lawsuits
Federal magistrate judge Maritza Braswell spends her days wading through pro se filings stuffed with AI-hallucinated citations, phantom case law, and legal arguments that sound plausible until you try to verify them. It’s a genuine access-to-justice crisis wrapped in a technology problem: AI is lowering the barrier to file, but not the barrier to be coherent. Meanwhile, judges in New York are on camera ripping into actual lawyers for submitting cases that simply don’t exist. The profession spent decades gatekeeping legal expertise — turns out the gate had a purpose.
TSMC: “We Can Only Support So Much”
TSMC CEO C.C. Wei, after a shareholder meeting, delivered the most honest sentence in the AI industry right now: “Customer demand is so high, and we can only support so much.” The world’s most important chipmaker — the one everything runs on — is admitting that even with its US factory buildout, it cannot keep pace with the appetite AI is generating. This is the physical constraint hiding behind every flashy model announcement: silicon is finite, fabs take years to build, and the demand curve doesn’t care.
Jeff Bezos Is Funding a $500M Hunt for the Brain’s “Core Algorithm”
A startup called Flourish, backed by $500 million and reportedly valued at $2.5 billion, wants to reinvent AI by literally studying real neurons under a microscope — searching for what it calls the brain’s “core algorithm.” Bezos is funding it. It’s either one of the most important bets in the history of the field or a very expensive reminder that the brain has been studied for centuries and has stubbornly refused to hand over its secrets. I’m genuinely curious. I’m also genuinely skeptical. Both things can be true.
Meta Is Building Data Centers in Tents
Meta, apparently inspired by Tesla’s “production hell” tent strategy, is deploying data center infrastructure under temporary structures to accelerate buildout timelines and cut costs. When one of the most valuable companies on Earth is housing its AI compute in canvas, you start to appreciate just how insane the infrastructure race has gotten. Kevin O’Leary, meanwhile, agreed to halve his 40,000-acre Utah data center after community backlash — so at least someone is being asked to slow down.
OpenAI Launches GPT-Rosalind for Life Sciences
OpenAI quietly dropped new capabilities for GPT-Rosalind, a specialized model aimed at life sciences research — covering biological reasoning, medicinal chemistry, genomics analysis, and experimental workflow design. This is the kind of vertical AI push that doesn’t get the splashy headlines but could matter enormously: putting serious scientific reasoning tools in the hands of researchers who previously needed a team of specialists. It’s named after Rosalind Franklin, which is either a lovely tribute or the kind of branding that makes you roll your eyes at 9am. Possibly both.
Bottom Line
The AI industry is simultaneously filing for trillion-dollar IPOs, running out of chips to build on, drowning courts in hallucinated legal filings, and losing the confidence of its own engineers — a perfectly coherent Friday.