Today's Headlines

  • Nvidia reports third-quarter fiscal 2027 revenue of $115.8 billion, up 103% from a year ago
  • OpenAI details chip design work by its models, citing gains of up to 14% on AMD's flagship GPU
  • Google DeepMind begins rolling out the Genie 4 world model to eligible Google AI Ultra subscribers
  • Major US banks move liability for AI-agent transactions onto customers
  • Le Groupe Figaro partners with OpenAI

Today is about the chips that run AI. Nvidia more than doubled its data center revenue from a year earlier, and OpenAI showed its models improving the design of the chips themselves.

Today's Top Three

Nvidia reports third-quarter fiscal 2027 revenue of $115.8 billion, up 103% from a year ago

Nvidia on November 17 reported revenue of $115.8 billion for the third quarter of fiscal 2027, which ended October 25, 2026, up 103% from a year ago and 20% from the previous quarter.

Data Center revenue was $107.6 billion, up 110% from a year ago. Gross margin was 74.8% on a GAAP basis and 74.9% on a non-GAAP basis, and diluted earnings per share were $2.70 (GAAP) and $2.68 (non-GAAP).

For the fourth quarter, which includes an additional week, Nvidia's outlook calls for revenue of $130.0 billion, plus or minus 3%. The company says the outlook assumes no Data Center compute revenue from China. It expects gross margins between 71.0% and 72.0%.

Nvidia says it has begun shipping its Vera Rubin NVL72 rack systems, with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. "Vera Rubin is off to an explosive start," said Colette Kress, interim CEO and chief financial officer, in the release. It was Nvidia's first quarterly report since Jensen Huang resigned as CEO on November 12.

OpenAI details chip design work by its models, citing gains of up to 14% on AMD's flagship GPU

OpenAI on November 18 published results showing its models improving the circuit designs of real chips.

Working with AMD, OpenAI says its model delivered up to 14% improvement in power, performance and area against heavily optimized baselines on AMD's flagship GPU IP, which it describes as "roughly a year of progress achieved in less than 5 months."

According to OpenAI, AMD has already included some of the changes in a recent design it sent for manufacturing. OpenAI says AMD also verified single-digit percentage gains across key blocks of its upcoming next-generation flagship CPU, using designs written by OpenAI's model.

On Jalapeño, OpenAI's first custom inference chip, the team estimates that changes proposed by the models reduced standard-cell area by about 13% in the vector unit and 10% in the matrix unit. OpenAI says it started using GPT-4o to help design Jalapeño in early 2025.

In tests on two open-source RISC-V processor cores at a budget of about 112,000 tokens per run, OpenAI says GPT-6.2 Astra had the highest average score on both. OpenAI also says it is partnering with Synopsys to train GPT-Synopsys, a model built to use Synopsys's chip design tools.

Google DeepMind begins rolling out the Genie 4 world model to eligible Google AI Ultra subscribers

Google on November 18 began rolling out Genie 4, the world model it unveiled in October, to eligible Google AI Ultra subscribers inside Project Genie.

With Genie 4, users can type a new prompt mid-session to change the world they created, such as shifting the weather or adding a character. The model also generates native audio that adapts as the scene unfolds.

Google also extended Genie 3 sessions from 60 seconds to up to 3 minutes per generation. The update adds a Project Genie agent that builds worlds through conversation and a gallery where users can publish and remix worlds.

All video and audio generated in Project Genie carries SynthID watermarks. Google AI Pro members can join a waitlist.

Other developments

Policy and Safety

  • OpenAI on November 18 described how it manages frontier AI risk from training through deployment. It says it trains frontier models cut off from the live internet and has built Control Room, an internal system that stops model output within 35 seconds of an instruction to halt a job.
    Managing frontier AI risk: From training to deployment (OpenAI)
  • Citibank changed its terms this week so that customers who give third-party AI agents access to their accounts are responsible for activity through those agents, including when the AI tool leaks their credentials or is hacked, according to TechCrunch. The outlet reports that PNC changed its terms on November 18 and that JPMorgan and Bank of America announced similar policies last month.
    More banks say customers using AI agents are on their own — even if they get hacked (TechCrunch)

Products

Business

Source: AI news inbox (collected November 19, 2026; 7 items, 6 primary and 1 secondary), selected by the editors.