Nvidia's Board Is Investigating Its CEO. Anthropic Still Plans Its $1.5T IPO.
6 min read · 14 sources
- Anthropic plans to start trading by end of next week at a target valuation around $1.5 trillion, despite investor pushback.
- Nvidia's stock slid over 4% after its board launched an investigation into CEO Jensen Huang's ties to an alleged Chinese spy.
- Rosen Law Firm filed a class action against Nvidia covering purchases between Oct 9, 2025 and Nov 8, 2026.
- Claude Fable 5.2 approaches "Jagged AGI," matching most human experts on most computer-based intellectual tasks.
- Storybook 11 beta ships with a 25% smaller package and up to 10x faster context delivery for coding agents.
Anthropic is going public at a valuation that could hit $1.5 trillion, and Nvidia’s CEO is being investigated for allegedly knowing a spy. Both things are true on the same Tuesday, and both are moving markets in opposite directions.
Start with the numbers that matter. Anthropic plans to start trading by the end of next week. Investors were pricing a ~$1.5 trillion valuation last week, but some major funds now want it lower after the AI selloff. Delaying would risk sparking panic. Nvidia, meanwhile, watched its stock slide over 4% after its board announced an investigation into CEO Jensen Huang’s relationship with Alice Huang - an executive U.S. intelligence believes was spying for China. The class action lawyers are already circling.
Rosen Law Firm’s class action covers Nvidia purchases between Oct 9, 2025 and Nov 8, 2026, accusing Huang of misleading investors about his ties to Megaspeed.
Claude Fable 5.2: The Jagged AGI That Still Needs a Human
Anthropic’s pre-IPO moment is riding on a model that one tester describes as approaching “Jagged AGI.” Claude Fable 5.2, tested before public launch, is “almost as good as most human experts at most intellectual tasks that can be completed with a computer,” according to the One Useful Thing review.
The demo that matters: the author had Fable create an embeddable video in one shot, with no feedback loop. That’s creativity and judgment at a human-expert level. But the “jagged” part is real - there are still important areas where the model falls well behind human capability. The point the author makes is that these aren’t simple errors or hallucinations anymore. It’s a different class of failure, closer to a brilliant but inconsistent colleague than a broken tool.
For engineers, this sets expectations. The models you’re integrating into production are no longer just autocomplete engines. They’re approaching parity on the tasks you’d hand to a competent contractor. The gaps are now in judgment under ambiguity, not in raw capability. Plan your human oversight accordingly.
Google's TPU Gambit: Confidential Computing and Long Contexts
Google is rolling out two new TPU variants aimed squarely at Nvidia’s weak spots. The 8tx targets confidential computing - secure, encrypted workloads for government agencies and regulated industries. The 8i Max is built for AI applications that need long-context windows, letting models access a wider range of information when handling queries.
This is the agent era play. Agents need to hold more context, and they need to do it in environments where data sovereignty and encryption are non-negotiable. Google’s TPU business has already netted Apple, Meta, and Oracle as clients. These new chips give engineers real alternatives to Nvidia’s GPUs for specific workloads - if you’re building for regulated industries or running long-context agents, the 8tx and 8i Max are worth benchmarking against whatever Nvidia SKU you’re currently renting.
Storybook 11 Beta: Agents Get a Workshop
Storybook 11 beta is the first release built for coding agents, not just humans. The headline numbers: a 25% smaller package, up to 10x faster context delivery than v10, and type-safe stories across React, Vue, Angular, Web Components, and Svelte.
The interesting part is the architecture. The new ui-workflow skill and headless CLI tools let agents query component APIs and docs without spinning up a running Storybook instance. That’s a token-efficiency play - agents get component context directly, without the overhead of a full browser session. The test/review tooling is powered by Vitest, which keeps the stack familiar.
For teams already using agents to write UI code, this removes a real bottleneck: the context-gathering step that used to eat tokens and time. Storybook is positioning itself as the agent’s source of truth for component behavior, which is a smart bet if agentic development is going to be the default workflow.
chrome-capture: Native 120fps Tab Recording, No ffmpeg
Paul Irish’s chrome-capture leverages the new Page.startScreenRecording API (landed September 2026) to record Chrome tabs at 60 - 120fps, up to 2160p, straight to an AV1 .mp4 file - tab audio included, no ffmpeg required.
The mechanics are interesting: frames go from the compositor to an out-of-process encoder over shared memory, which makes stopping near-instant. The CLI runs headless for ≤60fps and switches to headful on macOS for >60fps to lock onto ProMotion’s 120Hz display. Puppeteer v25.10+ wraps it as experimental page.record(), and Playwright has an open PR.
For anyone who’s fought with screen recording pipelines - capture, encode, mux, sync audio - this removes an entire class of tooling. If you’re building demo tools, testing suites, or anything that needs high-fidelity browser capture, this is the new baseline.
The Nvidia Mess: Board Investigation, Class Action, and a Timeline
The race to sue Nvidia is on. Rosen Law Firm filed a class action on behalf of a retail investor covering purchases between Oct 9, 2025 and Nov 8, 2026, accusing Nvidia and Huang of misleading investors about his ties to Megaspeed. Securities firms including Schall, Brown & Schwartz and Levi & Korsinsky have announced their own investigations.
The WSJ timeline of the drama notes that Huang told President Trump he had no idea Alice Huang was a spy. Whether or not that holds, the legal and reputational risk is now priced into the stock. For engineers, this matters at the supply-chain level: Nvidia’s operations and market position are now subject to litigation uncertainty, which could affect allocation, pricing, and roadmap timing. If you’re planning capacity around Nvidia hardware, build in some buffer.
Anthropic vs Meta: Room to Be Wrong
The MBI deep dive puts the two AI giants side by side. Anthropic is expected to go public at ~$1.5 trillion, though some investors push for ~$1 trillion after the selloff. Meta’s enterprise value is also ~$1.5 trillion. The difference is in the fallback position.
Meta’s core is advertising, with a bet on Meta Superintelligence Labs (MSL). If that bet fails, the ads business still prints money. Anthropic’s core is a fast-growing API, with a bet on first-party products. If the API growth stalls, there’s no safety net. Meta plans $135 - 145 billion capex this year (~3.5x 2024), with ~60% going to MSL and AI. Two-thirds of next year’s ~$212 billion consensus capex goes to AI.
The uncomfortable truth: Meta can be wrong about AI and survive. Anthropic can’t. That asymmetry is why investors are pushing for a lower IPO price, and why the IPO is still happening anyway - the window may close.
The Rest of the Day
Product Hunt almost died, according to founder Ryan Hoover. The company needed an acquisition to survive, but is now seeing record engagement - 3x more makers launching than a year ago and ~25% of unicorn founders having accounts. The lesson: each consecutive fundraising round is 10x harder than the last.
Kevin Kelly argues that difference becomes the valuable scarcity when AI is the main producer of content. Cheap variation isn’t valuable difference - the distinction is what pays.
An experimental mRNA therapy reawakened heart muscle cells in mice, boosting blood pumped per heartbeat by ~10% for at least 12 weeks. It activates Myc and Cyclin T1 temporarily, localized to reduce overgrowth risk. Human trials are a long way off.
Thom Wolf’s “future of software” describes “everflux software” (companies merging 30,000+ code changes daily), “translucent sources” (LLMs reverse-engineer closed software), and “bespoke features” (everyone’s their own PM). The challenge: customizing without cutting off upstream updates.
And writing as a thinking tool gets the Manuel Blum treatment: writing elevates you from a finite automaton to a Turing machine. It’s “out-of-core execution” for the brain.
Alphabet’s Wing will build a 15,000-square-foot factory in Fort Worth, Texas with capacity for tens of thousands of delivery drones per year.
You May Also Like
Cook exits, FTC sues Amazon ads, and Anthropic signs a $35B Nvidia-backed cloud deal
Tim Cook's 15-year run as Apple CEO ends with John Ternus, Apple's hardware SVP, taking over; Cook stays as executive chairman. The FTC sues Amazon for …
Anthropic teaches agents to drive lab gear, Cloudflare sheds 100TB of DNS bloat
Anthropic shipped the Model Hardware Standard, a driver layer that lets AI agents control microscopes, cameras and lasers over a common interface, aiming to …
Nvidia's CEO is gone, $1T vanished, and Apple wants Siri to actually do things
Nvidia CEO Jensen Huang resigned after admitting an affair with a woman US intelligence identified as a Chinese spy, wiping out over $1 trillion in market value …




