What Is Going On
Cerebras cashes in. Cerebras just made a loud entrance to the public markets, pricing its IPO at $185 a share, well above its already-raised range—and pulling in $5.55 billion. That gives the AI chipmaker a fully diluted valuation of $56.4 billion, making it one of the biggest tech IPOs in years and another sign Wall Street is all-in on the AI hardware boom. The timing helps: chip stocks beyond Nvidia have been ripping, and investors are clearly hungry for the next big AI infrastructure play. Still, Cerebras isn’t exactly risk-free. Its earlier IPO attempt got bogged down over customer concentration, and even now a huge chunk of revenue is tied to customers in the UAE. On the flip side, a January OpenAI deal worth more than $20 billion gives it serious credibility as it pivots from hardware sales to cloud services. More here.
AI Corner
Claude's agent catch. Anthropic is reopening the door for OpenClaw and other third-party agent tools on paid Claude plans, but don’t mistake this for a full rollback. Instead of letting subscribers tap their broad flat-rate usage for autonomous agents, Anthropic is carving out separate monthly “Agent SDK” credits priced at API rates. That means Pro users get a fixed programmatic budget, Max users get more, and once it’s gone, it’s gone; no rollover, no dipping back into normal subscription capacity unless you enable extra paid usage. The logic is simple: third-party agents were chewing through way more compute than Anthropic could sustainably subsidize, especially because they bypassed the caching efficiencies of Claude’s native tools. So yes, OpenClaw is back, but the era of cheap “compute arbitrage” looks over. Developers get a sandbox; Anthropic gets cost control; plenty of users are calling it a stealth nerf. More here.
Tabs get chatty. Microsoft is giving Edge a more aggressive AI brain: Copilot can now scan across your open tabs and answer questions based on what you’re browsing. That means it can compare products, summarize articles, and generally act like a research assistant living inside your browser. Microsoft says users can choose which experiences to enable, a notable nod to privacy as AI gets deeper access to browsing sessions. At the same time, the company is sunsetting Copilot Mode, the earlier feature that also tapped your tabs, while shifting its more agent-like tricks—like handling actions on your behalf—into “Browse with Copilot.” Edge is also picking up a “Study and Learn” mode that turns pages into quizzes or study guides, plus AI-generated podcasts built from your tabs, very much in the NotebookLM lane. In short: Edge wants to turn casual browsing into an AI-powered workflow. More here.
Cyber gets sharper. AI cyber capabilities are moving a lot faster than expected. The UK’s AISI says the length of cyber tasks frontier models can handle at 80% reliability had been doubling every 4.7 months since late 2024—already quicker than its prior 8-month estimate—but new results from Claude Mythos Preview and GPT-5.5 blew past even that pace. In plain English: these models are getting better at sustained hacking-style tasks, fast. Mythos Preview even became the first model to complete both of AISI’s simulated cyber ranges, including one that had never been solved before. There’s a catch, though: the benchmarks are narrow, token-capped, and based on a limited set of tasks, so they’re better at spotting trends than predicting real-world attacks. Still, the direction is pretty clear. AI cyber autonomy is improving quickly enough that governments and businesses should be paying close attention. More here.
Altman conflict spotlight. Sam Altman is under a very bright microscope after a court filing showed he holds more than $2 billion in companies that have done business with OpenAI. The disclosures surfaced in Elon Musk’s blockbuster lawsuit, which accuses OpenAI and Altman of breaching trust and seeks $150 billion in damages, plus Altman’s removal. The biggest pieces of the portfolio: a $1.7 billion stake in Helion, $633 million in Stripe, and $258 million in Retro Biosciences. Altman says he recused himself from relevant deal discussions, including OpenAI’s Helion arrangement and talks tied to Reddit and Cerebras. Still, the optics are getting rougher: 10 Republican state attorneys general want the SEC to scrutinize OpenAI’s documents ahead of a possible IPO, while a House committee is also asking questions about conflict-of-interest safeguards. More here.
Netflix wants more. Netflix’s ad machine is picking up serious speed. At its 2026 upfront, the streamer said its $8.99 ad-supported plan now reaches more than 250 million monthly viewers worldwide, up sharply from 94 million a year ago and marking the second straight year that audience has more than doubled. The payoff is getting real: Netflix says it pulled in $1.5 billion in ad revenue in 2025, and it’s clearly not done squeezing in more inventory. Next up, ads are headed to the company’s new vertical video feed on mobile, plus the podcast lineup it started rolling out late last year. In short, Netflix isn’t just experimenting with ads anymore, it’s building a much bigger ad business across more corners of the app, betting viewers will keep watching even as the commercial breaks spread. More here.
Legal AI cashes in. Legal tech is starting to look a lot like AI’s next blockbuster category, and Clio is making a strong case for it. The Canadian law firm software company just hit $500 million in ARR, a sharp climb after adding AI features in 2023. CEO Jack Newton argues the logic is simple: just as LLMs feast on the world’s code, they can also learn from the massive piles of contracts and legal documents law firms already have. Clio isn’t alone, either, Harvey reportedly reached $190 million in ARR, while Legora crossed $100 million only 18 months after launch. The bigger twist: Anthropic is now pushing deeper into legal with new Claude features, meaning some legal AI startups may be competing with a key supplier. Still, for Clio, the signal is clear: legal AI is moving from niche bet to very big business. More here.
Cool News
- ●
Cloud gift loop. Jensen Huang’s foundation is putting serious AI muscle behind academia, buying $108.3 million worth of computing time from CoreWeave and donating it to universities and nonprofit research institutes. The compute will go toward science and AI research, and Nvidia says some recipients will also get free engineering help. Nice philanthropy headline, sure, but it also doubles as another boost for CoreWeave, one of Nvidia’s key AI cloud partners. Nvidia already sells the GPUs CoreWeave rents out, invested $2 billion in the company earlier this year, and previously signed a $6.3 billion capacity deal that helps backstop unsold cloud inventory. That growing web of support is exactly why some investors have raised eyebrows over possible circular financing across the AI ecosystem. Meanwhile, CoreWeave recently lifted the low end of its capital spending outlook, citing pricier components. More here.