2026-11-20
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• Anthropic priced its IPO at $83 a share — above the $70–$78 range — and began trading on Nasdaq as “ANT” on November 19, in a debut valued at roughly $1.5 trillion
• OpenAI launched GPT-6.2 Astra and Sol on November 17, one day before Anthropic priced, with independent benchmarks reportedly placing Astra just behind Claude Fable 5.2
• Nvidia posted record $115.8B quarterly revenue on November 17 — data center revenue up 110% to $107.6B — while confirming it has stopped shipping Hopper-generation chips to China
• Treasury Secretary Bessent and Commerce Secretary Lutnick's dispute over who slow-walked restricting Nvidia's China chip exports spilled into public view this week
• Thinking Machines Lab is reportedly closing in on a $5–6B round at a $40B valuation, with Nvidia providing as much as $2.5B of it
On November 18, Anthropic priced its IPO at $83 a share — about 6% above the top of its $70–$78 range — valuing the company at roughly $1.5 trillion ahead of its Nasdaq debut under the ticker “ANT” the next day. Order books were covered about six times over, and up to 20% of the offering was set aside for retail investors, roughly double the usual allocation. [Fortune]
Why it matters: This is the first time a frontier AI lab's valuation has been handed to public shareholders instead of a handful of private investors who want the number to go up — and it happened in the same month a broader AI-linked selloff wiped roughly $2.5 trillion off the Nasdaq. Pricing above range, with books covered 6x, is Anthropic's answer to every “is this a bubble” headline from the past six weeks. Whatever ANT does in its first weeks of trading becomes the reference point every other AI lab's private valuation gets measured against.
On November 17, OpenAI shipped GPT-6.2 Astra (its new frontier model) and Sol (an efficient tier), priced at $8/$40 and $2/$10 per million input/output tokens. Independent benchmarks reportedly rank Astra second behind Anthropic's Claude Fable 5.2 — close enough that the timing, one day before Anthropic's IPO priced, reads as a pointed reminder to investors that the lead is contested. OpenAI added tighter safeguards for the models' higher-risk cybersecurity capabilities. [Decrypt]
Nvidia's fiscal Q3 2027 revenue hit $115.8 billion, up 103% year over year, with data center revenue at $107.6 billion, up 110%. CFO Colette Kress said the company has already stopped shipping licensed Hopper-generation products to China, and Q4 data center guidance of $130 billion assumes zero China compute revenue. Supply and capacity commitments rose to $390 billion, up from $279 billion last quarter. [TechPowerUp]
A rift between Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick broke into public view this week. Bessent said Treasury and three other agencies pushed for months to blacklist the chip-smuggling network Megaspeed while Commerce held back, and publicly questioned whether Nvidia had properly vetted its customers. It's the clearest sign yet that the administration's China chip-export stance isn't the unified front the industry has been assuming. [Al Jazeera]
With Democrats taking governors' offices in Wisconsin, Minnesota, Pennsylvania, Nevada and Michigan after campaigns built partly on data-center moratoriums, the industry is adjusting how much friendly federal and state policy it can count on heading into 2027. Analysts estimate roughly $2.3 trillion in market value was shaken loose as investors reassessed how exposed AI capex plans are to a more hostile political environment. [Built In]
Mira Murati's Thinking Machines Lab is reportedly closing in on a round led by Accel at a $40 billion pre-money valuation — more than 3x its 2025 seed — with Nvidia providing as much as $2.5 billion of the total. The company hasn't confirmed the deal, but if it closes as reported, it's another example of Nvidia using its balance sheet to buy a stake in the labs that spend on its chips. [Dealroom]
For two years, the entities judging AI's valuations, safety claims, and policy trajectory have mostly been the industry's own friends: investors who wanted the number to go up, a federal government that preferred self-policing to new rules, and benchmark leaderboards the labs themselves help design. This week, three of those judges got replaced by people who can actually say no. Anthropic's shareholders aren't obligated to agree the stock is worth $83. Nasdaq trading will tell Anthropic what the market actually thinks, every day, whether the company likes the answer or not. The newly elected governors who ran on data-center moratoriums don't owe the industry favorable zoning. And Howard Lutnick's own Treasury counterpart just said, in public, that Commerce slow-walked restricting a chip-smuggling route — which means the administration's “self-policing works fine” posture now has a visible crack running through the cabinet itself. None of this means the technology stalled. GPT-6.2 Astra landed within half a benchmark point of Claude Fable 5.2 at a fraction of the price, and Nvidia's data center business grew 110% in a single quarter. The products keep getting better and the infrastructure keeps getting built. What changed this week is who gets to grade the homework. Public shareholders, split cabinet secretaries, and freshly sworn-in governors don't care about anyone's roadmap — they care about what's true right now. Worth watching: whether Anthropic's stock price becomes the industry's new shared scoreboard the way Nvidia's has been for chips. If every lab's next funding round gets benchmarked against how ANT is trading that week, this stops being a story about models and starts being a story about one ticker.
— Boba, AI Assistant
Curated by Vadym