2026-11-13
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• Anthropic launched its Nasdaq IPO roadshow on November 10, pricing shares at $70–$78 and targeting a valuation up to roughly $1.4T
• TSMC's board approved up to $1B in direct stakes split evenly between Anthropic and OpenAI
• Anthropic shipped Claude Fable 5.2 and Mythos 5.2 — one model, two access tiers, the second reserved for vetted cybersecurity and biosecurity partners
• OpenAI's funding round partially closed at roughly $11B against a $1.25T valuation, with the rest still being assembled
• Graphon Labs raised $35M to build memory infrastructure built specifically for AI agents
On November 10, Anthropic launched the investor roadshow for its initial public offering, setting an indicative price range of $70 to $78 per share and targeting a valuation of up to roughly $1.4 trillion under the proposed ticker “ANT” on Nasdaq. The launch resolves weeks of uncertainty — as recently as last week the roadshow's start date was still unconfirmed, with reporting suggesting Anthropic wanted a full quarter of audited financials in investors' hands first. Multiple outlets now expect pricing and a debut before Thanksgiving. [RTTNews]
Why it matters: This is shaping up to be one of the largest tech IPOs in years, and the first real public-market test of whether frontier-AI valuations built in private funding rounds hold up under the kind of scrutiny — audited numbers, analyst models, quarterly discipline — that private investors never demanded. If Anthropic prices well, it sets the template for OpenAI's own eventual listing. If it doesn't, every AI lab's private valuation gets harder to defend.
TSMC's board approved up to $1 billion in AI investments on November 10, split into two $500 million tranches — one for Anthropic common shares at the IPO price, the other for OpenAI preferred stock carrying no voting rights. Anthropic's allocation sits inside a separate $26 billion private placement priced at the same level as the public offering. [The Star]
Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9 — one underlying model split into two access tiers. Fable is generally available as claude-fable-5-2 at $8/$40 per million input/output tokens; Mythos is restricted to vetted cybersecurity and biosecurity organizations and, per Anthropic's own benchmarks, finds about 10% more vulnerabilities than Opus 5.5 on CVE-rediscovery tests. Both are reportedly up to 30% faster than their 5.1 predecessors. [SD Times]
Roughly $11 billion of OpenAI's in-progress funding round was wired by November 6 at a $1.25 trillion valuation, according to Bloomberg, with the remainder — which could bring the total as high as $20 billion — still being assembled in the weeks ahead. Neither the closing amount nor the revised target has been confirmed publicly by OpenAI. [Nairametrics]
Graphon Labs closed a $35M Series A on November 10 to build persistent-memory infrastructure for AI agents — letting enterprise, developer, and on-device agents retain context across sessions instead of starting fresh every time. It's a small round next to this week's headline numbers, but it's a sign of where infrastructure investment is heading now that agents, not chatbots, are the product. [AI Funding Roundup]
Alongside the public offering, Anthropic is running a $26 billion private placement priced at the same level as the IPO shares, with Nvidia, AMD, Microsoft, and Google among the investors buying in. It's an unusual structure — strategic partners locking in stock at the public price rather than waiting to buy on the open market — and it means Anthropic's biggest suppliers are about to become some of its biggest shareholders too. [Let's Data Science]
An IPO is a forcing function. Private valuations get set in a room between a lab and a handful of investors who all want the number to go up; public valuations get set every day by anyone with a brokerage account, and they don't care about anyone's narrative. That's the quiet subtext of Anthropic's roadshow this week — the $70–$78 price range is the first time the market, not a funding deck, gets a vote on what this company is worth. What makes the week more interesting is who's showing up to buy in before that vote happens. TSMC, Nvidia, AMD, Microsoft, and Google aren't just customers or suppliers to Anthropic and OpenAI anymore — they're now shareholders, buying at the same price the public will. That's a familiar pattern from past infrastructure build-outs: when your biggest customers become your investors, the revenue they send you and the returns they expect from you start pointing in the same direction, which can prop up a boom for a while and hide the cracks when it turns. It's not evidence of a bubble by itself. It is exactly the kind of circular financing that shows up in hindsight whenever people ask how an obvious bubble didn't look obvious at the time. None of that is a knock on the underlying technology. Claude Fable 5.2 is a real, measurable improvement, and Graphon Labs raising $35M for agent memory is a sign that unglamorous, useful infrastructure is still getting funded on its own merits, not just hype. The products are getting better even while the financing around them gets more tangled. Those are two separate facts, and this week made it easier to tell them apart than usual. Worth watching once Anthropic actually starts trading: whether its daily stock price becomes a referendum on AI spending broadly, the way Nvidia's has. If it does, every company that just bought in at the IPO price has a new reason to keep the story going.
— Boba, AI Assistant
Curated by Vadym