Nvidia starts work under interim CEO Colette Kress with an all-hands set for Wednesday, Google puts Gemini 4 Hydrogen in front of everyone, SpaceX pledges more than $74 billion for two AI supercomputing campuses, a nonprofit coalition sues five AI giants to freeze frontier models, Notion confirms customer chats were exposed in the Fireworks breach, Nebius shows the neocloud economy's two faces, and United tells AI agents they can look but can't book. In the afternoon, Congress summons Jensen Huang and the Commerce Secretary to testify, OpenAI launches ChatGPT Wallet with all three big card networks, and agent-tooling startup LangChain more than doubles its valuation to $3 billion.
1Nvidia's first day after Huang: an all-hands Wednesday and an open CEO search
On Nvidia's first day without Jensen Huang, its new leadership moved to steady the company: interim CEO Colette Kress told employees they will hear more "in the coming days" and called an all-company meeting for November 18 at 9am PT, in staff emails obtained by CNBC. Harvey Jones, the board's first-ever chair, told employees the board will consider internal and external candidates for the permanent job and praised Kress - chief financial officer for over 13 years - as "a key architect of NVIDIA's growth and capital strategy."
In plain English
The caretakers of the company at the center of the AI boom want everyone to hear one message - business as usual while the board finds a permanent CEO.
Why it matters
The all-hands lands the day after Nvidia's scheduled November 17 earnings report, now the first public test of investor confidence in a Nvidia without its founder.
2Google ships Gemini 4 Hydrogen, a faster everyday model, to everyone
Google released Gemini 4 Hydrogen on Thursday, a speed- and cost-focused model for coding and agent work that replaces Gemini 3.8 Flash across its products - the Gemini app, Google Search, AI Studio, the Gemini API and enterprise tiers. Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default setting, edging past its own frontier Argon model's 50, runs 2.5 times faster than 3.8 Flash, and costs a flat $1.50 per million input tokens and $7.50 per million output across a one-million-token context window. A new Orchestrator in Google's Antigravity tool automatically switches between Hydrogen and the frontier models inside one session.
In plain English
The model most people will actually talk to just got smarter and cheaper, not only the flagship.
Why it matters
The model race has moved to the cheap workhorse tier that runs inside products billions of people use, where small price and speed edges compound. The benchmark figures are Google's own.
3SpaceX pledges more than $74 billion for two AI supercomputing campuses
SpaceX announced plans on Thursday for two giant AI computing campuses: about $50 billion in Cochise County, Arizona, which it says would bring roughly 5,500 construction jobs and 1,000 permanent roles, and about $24 billion at Starbase, Texas, on company land near the Port of Brownsville. The Texas site will train frontier AI systems and generate its own electricity, and SpaceX is putting $220 million into expanding the area's water and sewage systems. The Arizona project, planned for dry cooling with no industrial groundwater use, still needs county approval.
In plain English
Elon Musk's rocket company is becoming a power-and-compute company too.
Why it matters
The pledge lands as data centers turn into a political flashpoint - Texas's governor-elect campaigned on a statewide moratorium - and SpaceX's own statement concedes investments of this scale require "stable, competitive policies."
4Nebius revenue jumps 571%, and the stock still falls 16%
Nebius, one of the largest "neoclouds" renting out AI computing power, reported third-quarter revenue of $980.8 million - up 571% from a year earlier and ahead of the $907.5 million analysts expected - with an adjusted profit of 31 cents a share against an expected 69-cent loss. But it guided full-year revenue to the low end of its $3 billion to $3.4 billion range, raised 2026 capital spending plans to $27 billion from $20-25 billion, and its shares fell 16% on Thursday.
In plain English
Business is booming, and investors are still nervous about what the boom costs to build.
Why it matters
Nebius expects $40-45 billion of capital spending in 2027 - the buildout's appetite for money keeps growing even as markets punish the companies feeding it, and backer Nvidia's own turmoil adds to the sector's volatility.
5AMD expands its MI400 chip line and claims a benchmark edge over Nvidia
AMD on Thursday added the Instinct MI450X - 288 gigabytes of HBM4 memory - to its MI400 GPU family for its Helios rack-scale systems, detailed the science-focused MI430X with 432 GB of memory and native double-precision compute, and said its lab testing shows the MI455X reaching 21,333 tokens per second per GPU on DeepSeek-R1 in an MLPerf-style offline setup - up to 30% above Nvidia's latest published Vera Rubin NVL72 result.
In plain English
AMD is attacking Nvidia's crown with more memory options and a carefully chosen benchmark.
Why it matters
The 30% figure is AMD's own testing against a published Nvidia result, not an independent contest - but with Nvidia distracted by a leadership crisis, every credible alternative has an opening.
6Nonprofits sue five AI giants to halt frontier model development
A coalition of four nonprofits - the Irreplaceable Education Fund, Public Citizen, the Heat Initiative and the Alliance for Secure AI - sued Anthropic, Google, Meta, OpenAI and SpaceXSI in San Francisco Superior Court on Thursday, asking a judge to freeze development of more advanced AI systems, the Associated Press reports. The suit, brought under California's Unfair Competition Law by former New Jersey attorney general Matthew Platkin, argues the companies keep releasing systems their own executives warn could escape human control and enable bioweapons and catastrophic cyberattacks.
In plain English
Critics are asking a court, not Congress, to hit pause on the AI race.
Why it matters
The filing tests whether judges will treat the labs' own safety warnings as evidence against them - and it lands days after midterm elections that put AI regulation on the new Congress's agenda. The claims are allegations, not findings.
7Notion confirms the Fireworks breach exposed some customers' AI chat transcripts
Notion says attackers who breached AI infrastructure provider Fireworks - through Fireworks' own cloud provider - accessed Notion AI chat transcripts from its Insights feature, affecting fewer than 1% of workspaces. In a security advisory updated Friday, Notion said the transcripts were mostly "unattributable" - not labeled with who wrote them or which workspace they came from - and contained no tool-call results. Notion disconnected from Fireworks on November 3, told affected customers on November 9, revoked exposed API keys on November 11, and has disabled Insights while it investigates. TechCrunch reports the attackers also took Fireworks customers' billing information and datasets, and appear financially motivated.
In plain English
The hack happened at a supplier, not at Notion - but it is Notion customers' conversations that leaked.
Why it matters
AI features quietly multiply the places your data sits - every third-party model provider in the chain is another door to guard. Notion advises affected users to rotate any credentials they pasted into Notion AI.
8United tells AI agents: you can look, but you can't book
United Airlines will block unauthorized third-party AI agents from completing bookings on its website and app, the Wall Street Journal first reported, citing privacy risks, unintended purchases and federal rules that require a traveler to attest to things like not packing explosives. Agents can still show United's flights; they just can't finish the transaction. "To be honest, I don't want to create another middleman," CEO Scott Kirby said Tuesday, and eBay and several major banks have taken similar steps, The Verge reports.
In plain English
One of the biggest US airlines just closed its checkout to the AI assistants that want to shop for you.
Why it matters
Agentic commerce only works if merchants let agents in, and big consumer brands are starting to say no until protections exist - a real brake on the agent economy. United says it could reconsider as the technology matures.
9Congress summons Jensen Huang and Commerce Secretary Lutnick to testify
Senate Banking Chairman Tim Scott summoned former Nvidia CEO Jensen Huang to testify December 3 - threatening a subpoena if he does not cooperate - and Commerce Secretary Howard Lutnick for December 10, over national security concerns and sales of sensitive US technology to China, UPI reports. Lutnick agreed to appear voluntarily almost immediately. Three House committee chairmen sent their own letters requesting appearances by Huang, Nvidia general counsel Tim Teter and a Commerce under secretary, calling the situation a prime reason to pass a package of chip-security bills.
In plain English
Congress is not letting Huang's resignation close the case - both chambers now want him under oath about what Nvidia knew and what the administration did once it learned of his relationship with a suspected Chinese intelligence asset.
Why it matters
The questioning has bipartisan support, and Senator Warren says neither Huang nor Nvidia is off the hook "for placing national security at risk or for any potential violations of our laws" - the scandal is turning into chip-export policy, with four security bills now riding on it. US officials assess the woman, Alice Huang, worked for China's Ministry of State Security; Jensen Huang has not been charged with anything.
10OpenAI launches ChatGPT Wallet so agents can pay - with a human approving each purchase
OpenAI introduced ChatGPT Wallet, built with American Express, Mastercard and Visa, which it calls the first wallet that lets agents pay without ever seeing a user's actual card details. The user adds a card once, then reviews and approves each purchase with a security check like Face ID; the card network then issues a one-time payment token, so the merchant and the agent never receive the real number. It rolls out today through dots, OpenAI's agent feature, for US ChatGPT Pro subscribers with US-issued cards, and OpenAI published a playbook telling businesses how to stop their fraud systems from blocking legitimate agent purchases.
In plain English
ChatGPT can now check out for you, but you still tap approve on every purchase - and your card number stays hidden from everyone but your bank.
Why it matters
One day after United Airlines said it will block agents from finishing bookings, OpenAI and the three biggest card networks are building the trusted lane those merchants say doesn't exist yet - the fight over who controls agentic checkout is now fully joined. OpenAI says delegated purchasing with preset spending limits comes later.
11LangChain more than doubles its valuation to $3 billion as agent tooling consolidates
LangChain raised a $300 million Series C at a $3 billion valuation led by CapitalG, with Singapore's GIC, DFJ Growth, IVP and Sapphire Ventures participating, TechCrunch reports - more than double its $1.25 billion price about a year ago. The company, which began in 2022 as an open source library, says its LangSmith agent-management platform now serves 7,000 customers, including Harvey, Rippling, Nvidia, Toyota and Salesforce, pitching model-neutral tooling that lets businesses own their own intelligence.
In plain English
Investors are betting that companies will pay for independent plumbing to run their AI agents rather than buying it all from one big lab or cloud.
Why it matters
As agents move from demos to checkout lanes and factory floors, the money is following the pick-and-shovel layer - and LangChain's round, on the same day OpenAI launched its own agent payments stack, shows both ends of the agent economy attracting capital at once.
Checked against original announcements and named press reports on November 13, 2026, morning and mid-afternoon. Gemini 4 Hydrogen's benchmark and pricing figures are Google's own; AMD's performance claim is its own lab testing against Nvidia's published results. SpaceX's investment and jobs figures are the company's own announcements. The lawsuit's claims are the plaintiffs' allegations, not court findings. The Gemini, SpaceX, lawsuit, Notion, Nebius and AMD items were announced November 12 and the United item November 11, after Thursday's briefing was published - they are included as catch-up coverage. ChatGPT Wallet's security design and rewards features are OpenAI's own description, and LangChain's customer count is the company's own figure. The espionage assessment of Alice Huang is by US officials as reported by UPI, not a court finding.
