An empty executive chair in front of a wall of server racks

AI News, Nov 18: Nvidia Beats Without Its Founder

The Big Story: Nvidia’s first quarter without Jensen Huang

Nvidia reported $115.8 billion in revenue for the quarter ended October 25, up 103% year over year and 20% sequentially, with data center revenue of $107.6 billion. Non-GAAP earnings were $2.68 a share against roughly $2.50 expected, and guidance for the current quarter is $130.0 billion, plus or minus 3%. Shares rose about 3% after a month that had taken roughly 24% off the stock.

The call mattered more than the beat. Chairman Harvey Jones opened it by saying the board “unanimously accepted his resignation and does not intend to reinstate Jensen as CEO”, days after Huang posted on X that he would be “honored to answer these calls and return as CEO.” Internal and external candidates are both in play. Colette Kress, the CFO now serving as interim chief executive, took no questions about the circumstances of the departure.

Two things got worse underneath the headline numbers. Kress said Nvidia is “no longer shipping licensed Hopper data center products to China”, walking away from roughly $2.4 billion of H200 sales booked last quarter in a market Washington had explicitly approved after the Trump–Xi summit. And gross margin is guided down from 74.9% to between 71% and 72% on the Vera Rubin ramp.

A company growing 103% a year does not usually guide margins down and abandon a licensed market in the same hour. Analysts split accordingly: JPMorgan raised its target, Jefferies and Wells Fargo both cut theirs from $300 to $250.

Tuesday’s model launches all led with price, not capability

OpenAI’s GPT-6.2 Astra arrived at $8 per million input tokens and $40 output, 20% below GPT-6 Astra, with a five-point Artificial Analysis gain that still leaves it second behind Claude Fable 5.2. Reception from power users was hostile: Matt Shumer called the models “essentially unusable” and Theo Browne called Astra “the worst model OpenAI has ever released.” Microsoft used its Ignite keynote to put MAI Code 2 on GitHub with the pitch that it almost matches rival models at a fraction of the cost. Cerebras shipped a production split architecture that runs prefill on AMD GPUs and decode on its wafer-scale systems, sold on one number: five times the throughput per megawatt.

Not one of them claimed a capability jump big enough to carry a price increase.

Wednesday’s Top Stories

Anthropic is set to price its IPO above range

CNBC’s David Faber reported Tuesday that Anthropic is likely to price above its $70 to $78 range. The order book closed at 4 p.m. ET and the stock trades Thursday on Nasdaq under ANT. The S-1 shows $33.24 billion in revenue for the nine months to September 30, up roughly 1,200% year over year, against a $104 billion net loss driven mostly by fair-value changes on convertible notes and derivatives rather than cash burn. Pricing above an already enormous range is the banks betting the public market will absorb the largest AI listing yet.

Congress’s watchdog got four chatbots to plan attacks

A GAO report says investigators obtained viable attack plans from ChatGPT, Claude, Gemini and Grok, each in under 12 hours. Two platforms produced plans to build and detonate an improvised explosive device; two produced substation attack plans using a drone or firearms, complete with maps and images. One refused the detailed bomb requests at first, and investigators got around its guardrails. Reviewers did find errors, including misplaced roads and sightlines blocked by walls. Anthropic said it takes the finding seriously but cannot verify it; Google said the tests used earlier models. Senator Maggie Hassan, who requested the report and chairs Senate Homeland Security in the new Congress, said “we need to act now” and will hold hearings in January.

Oregon sues OpenAI for $2 billion over child safety

Attorney General Dan Rayfield filed under the state’s Unlawful Trade Practices Act, alleging OpenAI misled the public about ChatGPT’s safety for minors and that the push for growth made the product “calculated to tell users exactly what they want to hear.” The suit targets ChatGPT for Teens, pleads public nuisance, and seeks at least $2 billion plus an order barring OpenAI from misrepresenting the product’s age-appropriateness. It lands two days after California’s suit, the second state in a week to attach a ten-figure number to an OpenAI safety claim.

Attackers took 149,346 Uber drivers’ license records from an AI vendor

Uber disclosed that intruders used legitimate credentials to reach a cloud storage environment it had set up for Fireworks AI, taking driver’s license images along with the names, home addresses, dates of birth and license numbers extracted from them. Because the credentials were valid, nothing tripped. Uber did not name the vendor in the state filings where the count originates, but confirmed it to TechCrunch. The inference layer is now a place companies keep identity documents, and it inherits every supply-chain problem that comes with them.

A higher-scoring coding model did not ship more work

A paper posted Tuesday treats June’s 19-day export-control suspension of Claude Fable 5 as a natural experiment. It follows 297 developers who lost access and fell back to Opus 4.8, which scores 11 points lower on SWE-Bench Pro, against 8,908 who were on Opus 4.8 throughout. Access to the higher-scoring model raised the seven-day pull-request merge rate by 1.4 percentage points, with an interval running from -1.4 to +4.3. A replication across five upgrades and 15,442 pull requests found the same null. It is a single-author paper whose identification strategy deserves scrutiny, but it is the first direct test of whether benchmark points turn into merged code.

Quick Hits

Ready to automate your busywork?

Carly schedules, researches, and briefs you—so you can focus on what matters.

See what people say

"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.

Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.

On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."

Gus Ibrahim, Founder & Director, IHR