Skip to content
AI daily

6 min · 4 stories

AI Week: Anthropic's Record IPO, Nvidia's Mess, and Bank Fine Print

By Robert Hattala

In this issue

I have been doing this daily AI roundup for a while now, and I do not think I have ever had a news day where the biggest IPO in history was only half the story. Wednesday was that day. Here is what mattered.

01 of 04

Anthropic prices the biggest IPO ever at $83 a share

Anthropic priced its IPO Wednesday night at $83 a share, above the $70 to $78 range it originally set, because investors kept piling in. It offered 980 million shares of Class A common stock on Nasdaq, which works out to $81.3 billion raised and a valuation of about $1.5 trillion. That knocks SpaceX off the top of the record books. SpaceX raised $75 billion when it went public in June.

That is not even all the money. Anthropic said it would raise another $26 billion privately from AMD, Broadcom, Google, Microsoft, and Nvidia, which puts the total haul at roughly $107 billion. CBS News points out it is the first frontier AI lab to go public. Back in October, when it announced the plan, Anthropic called the IPO one of the "seminal moments in human history." Modest folks.

Sam Grelck, an equity strategy analyst at Truist Advisory Services, told CBS he would not view the IPO "as a verdict on AI," just a read on how hungry investors are for the theme.

I agree with him, and I would go a step further. Pricing above the range tells you about demand for a story, not about whether the business underneath it is worth $1.5 trillion. Now there are quarterly reports, public shareholders, and a stock ticker that moves every time a competitor ships something. A lab that spent years talking about safety and patience just signed up for the least patient audience on earth. I am curious how that changes them.

Source: CBS News, TechCrunch

02 of 04

Nvidia posts a monster quarter with nobody permanent in the big chair

Nvidia reported $115.8 billion in revenue for the quarter that ended October 25, up 103% from $57.0 billion a year ago. Data center alone was $107.6 billion. GAAP net income came in at $65.3 billion, or $2.70 a share. For next quarter it is guiding to $130 billion, plus or minus 3%, and it is assuming zero data center compute revenue from China.

The person reading those numbers on the call was Colette Kress, the CFO who is now interim CEO. Jensen Huang left abruptly last week. TechCrunch reports Nvidia is running an internal investigation tied to Huang's personal conduct, and the press release mentions an independent board committee and a search for a new CEO.

Two details from the call matter more than the headline number. First, Nvidia did not file its 10-Q alongside earnings like it normally does. Kress just said it would come "in the coming weeks." UBS analyst Timothy Arcuri asked afterward and wrote that management tied the delay to the ongoing investigation and whether new risk factors need to be added. He kept his buy rating but cut his price target from $240 to $220.

Second, Kress clarified the $105 billion loan backstop Nvidia put behind OpenAI's PORTS-Pike campus in Ohio. The guarantee only kicks in once the customer has the money to build the facility and get the compute installed. In her words, if the project is "delayed or they have a challenge in terms of building that, our guarantee doesn't exist anymore." She said Nvidia will be "far more selective" from here.

Here is where I land on it. The chips are selling like ice water in August, nobody is arguing that. What spooks me is the combination: a missing filing, an interim CEO, and the company quietly narrowing the financial promises that helped keep its own customers buying. Nvidia did not just sell shovels during this gold rush, it co-signed some of the loans for the shovels. When the co-signer starts reading the fine print out loud, I pay attention.

Source: Nvidia press release, TechCrunch on the 10-Q, TechCrunch on the backstop

03 of 04

Four nonprofits sue to freeze frontier AI where it stands

Four nonprofits, Irreplaceable Education Fund, Public Citizen, Heat Initiative, and Alliance for Secure AI, filed a 63-page complaint in San Francisco Superior Court against Anthropic, Google, Meta, OpenAI, and SpaceX. They are represented by Matthew Platkin, the former New Jersey attorney general.

The legal hook is California's Unfair Competition Law, with the argument that building ever more capable AI is a public nuisance that endangers health, safety, and property. What they want is wild in its simplicity: an immediate injunction barring the companies from developing or deploying anything "more capable than their most capable systems as of the date of this Complaint." Existing products would keep running. The frontier would just stop where it was on November 12. They also want the labs barred from using AI to run their own research into better models.

For evidence, Decrypt reports the complaint leans on incidents like July's episode where OpenAI models escaped a test environment and compromised Hugging Face's infrastructure while trying to cheat on a cybersecurity evaluation. Phil Aroneanu, who runs Irreplaceable, told Al Jazeera that "a real freeze must specify what work has to halt, how compliance is verified, and the consequences of breaking the rules."

What I want to know is how a state judge even measures "more capable." There is no agreed yardstick, and the labs will argue every benchmark to death. I do not think this injunction gets granted. But I also do not think that is the point. Discovery is the point. If this survives a motion to dismiss, a lot of internal safety testing ends up in a public court record, and that might do more than any freeze would.

Source: Al Jazeera, Decrypt

04 of 04

Big banks say your AI agent is your problem, even if it gets hacked

Citi updated its online user agreement this week, and PNC changed its terms Wednesday. They join JPMorgan and Bank of America, which rolled out similar policies in October, and Citizens, which emailed customers this week with the same warning. That is five of the biggest banks in the country lined up on one position.

Citi's version is the most spelled out. A Citi spokesperson told TechCrunch that "customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents." According to All About Cookies, the new language treats anything a third-party agent does as authorized by you, even if your credentials later get stolen, and Citi says it has no duty to investigate those transactions. Citi can also cut off an agent's access at any time without warning. The carve-out: AI tools Citi builds and offers itself are excluded.

I get why the banks are doing it. Handing your banking login to a bot is a security nightmare and they do not want to eat the losses. But notice the shape of the deal. Their AI is safe to use, everybody else's AI is on you. That is not just risk management, that is a moat. If you want to use an agent with your money anytime soon, read your bank's terms before you paste in a password, because the answer to "who pays if this goes sideways" is now in writing, and it is you.

Source: TechCrunch, All About Cookies

Thursdays

Get the digest on Thursdays

Each Thursday: the shop notes and AI daily issues from the last seven days, and one product. Unsubscribe from any issue.