← Back to articlesAI news
AI News November 13, 2026: Jensen Huang Leaves Nvidia, Outlook Down for 12+ Hours and Fable 5.2 at $8
Jensen Huang resigns from Nvidia, Microsoft 365 stays down for 12+ hours, Fable 5.2 cuts token prices by 20% and Anthropic sets its IPO for November 19.

Credit: Numerama
Jensen Huang resigned on Thursday, November 12, as president, CEO and board member of Nvidia, effective immediately, after 33 years at the head of the company he co-founded. The stock lost up to 14% intraday, and CFO Colette Kress is serving as interim CEO.
The same day, a Microsoft 365 outage hit Outlook, Teams and Copilot in 74 countries and was still running more than 12 hours after it began. Anthropic launched Fable 5.2 and Mythos 5.2 on November 9, with a token priced 20% below the previous generation. OpenAI announced it would stop supplying its models to Cursor, now owned by SpaceX, with November 12 as the proposed cutoff date.
Anthropic, for its part, has started its roadshow: the IPO price range is set between $70 and $78 a share, with listing planned for November 19.
Breakdown.
Microsoft 365 Outage: Outlook and Teams Down for More Than 12 Hours
A Microsoft 365 outage began on Thursday, November 12 in the early afternoon (US East Coast time) and was still disrupting Outlook, Teams, SharePoint and Copilot on Friday morning. Microsoft first cited elevated load, then corrected itself: servers that became unavailable were overwhelming the others. No root cause had been published at that point.

Click to enlarge
Credit: AP / NPR
The incident carries the number MO1550821 in Microsoft's admin center (Exchange Online EX1550543, Teams TM1550488). According to Al Jazeera, more than 130,000 Outlook reports were logged on Downdetector across 74 countries. Microsoft 365 claims about 430 million users, and the outage also hit Exchange Online, SharePoint Online, Microsoft 365 Copilot and Defender XDR.
The diagnosis changed during the day. Microsoft first announced a traffic spike, then explained that a core process on the affected servers was overloaded and that these servers were failing automated health checks (regular tests that pull a server out of service as soon as it responds badly). Deployments were paused, traffic redistributed and servers restarted. On Friday morning, Microsoft described "marginal success": restored servers can become unresponsive again as soon as they receive traffic. A fix to an authentication and routing component was applied at 2:24 a.m. Eastern time, and a network configuration change was being reverted, with no confirmed link to the outage (NPR). Microsoft sees no sign of malicious activity and announced a post-incident review.
The most recent precedent dates back to August, when a code update disrupted the service for several days. Concentrating usage on a single email and video-conferencing provider is measured here in lost working hours, not in revenue.
Another case of single-provider dependency played out the same day on the developer tools side.
OpenAI Cuts Its Models to Cursor, Acquired by SpaceX
OpenAI announced it would stop supplying its models to Cursor, which SpaceX acquired for $60 billion on August 14, with November 12 as the proposed cutoff date. The stated reason is trust: OpenAI says it cannot guarantee that SpaceX will comply with its terms of use. OpenAI models account for about 5% of Cursor's traffic.

Click to enlarge
Credit: Reuters / CNBC
The announcement dates from August 28, fourteen days after the acquisition closed, and the proposed cutoff date is November 12, with no public confirmation that it has taken effect. OpenAI justified its decision by citing "our experience with Elon Musk's companies violating contracts," CNBC reports. The contract contained a change-of-control clause (a provision that lets a supplier terminate an agreement when its customer is acquired), and OpenAI had a limited window to exercise it, according to Reuters. OpenAI also says no future models will be supplied to Cursor, which includes the upcoming GPT-6 Astra.
Michael Truell, CEO of Cursor, replied that "OpenAI models serve about 5% of Cursor's user traffic." Personal OpenAI API keys keep working in the editor. Anthropic announced, through Tom Brown, that it was increasing the compute dedicated to Claude in Cursor, and Fable 5.2 was available there on its launch day, November 9.
For development teams, the episode illustrates a supply-chain risk: a change of ownership can suspend access to a model within weeks. The topic of dependence on a single coding-agent provider is detailed in the architecture of AI code agents.
On the Anthropic side, the week is mostly defined by a new model and a new price grid.
Fable 5.2 and Mythos 5.2: One Model, Two Levels of Safeguards
Anthropic launched Fable 5.2 and Mythos 5.2 on November 9, two versions of the same model with different safeguards. Fable is open to all paid plans and the API at $8 input and $40 output per million tokens, 20% less than Fable 5.1. Mythos remains restricted to verified programs.

Click to enlarge
Credit: Anthropic
According to Anthropic's launch page, Mythos 5.2 is available only to participants in the Cyber and Life Sciences Verification Programs, while Fable 5.2 is available on the API (
claude-fable-5-2), AWS, Google Cloud and Azure. The context window reaches 1 million tokens and maximum output 128,000. A fast mode charges $16 input and $80 output for output up to 30% faster. According to Anthropic, false blocks by the safeguards drop by about 55%.On published benchmarks, Fable 5.2 reaches 49.1% on FrontierCode 2.0 at the Xhigh effort level, against 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra. On FrontierSWE v2, the score is 69.9% against 65.5% for Astra and 61.2% for Gemini 4 Carbon. The Artificial Analysis Intelligence Index puts Fable at 65, against 59 for Opus 5.5. On Zapier's AutomationBench, Gemini 4 Carbon (53.9%) beats Fable (51.4%). Anthropic also states that 1,000 Fable agents built an operating system in about 8 days (CNBC).
A pricing nuance: the 20% cut applies to the standard rate. The 80% gap with GPT-6 Astra, sometimes cited, only concerns tokens read from cache ($0.20 against $1 per million). Opus 5.5 remains half the price ($4 and $20), and Anthropic still recommends it for most tasks. Analyst Sanchit Vir Gogia of Greyhound Research sums it up in InfoWorld: "Anthropic has cut the price of a token, not the price of a thought."
The same day, Meta opened the other end of the range with Muse Ember, priced at $0.10 input and $0.50 output per million tokens, the same rate as Claude Haiku 5.5 and GPT-6 Luna on short requests (VentureBeat). Google, for its part, opened Gemini 4 Carbon to paying developers on November 10, at $2 input and $10 output (Google blog). Details on using Claude Code in production are in the Claude Code field report.
Jensen Huang Resigns From Nvidia
Jensen Huang resigned on Thursday, November 12 as CEO, chair and board member of Nvidia, effective immediately. Colette Kress, CFO since 2013, is interim CEO, and Harvey C. Jones becomes the board's first chair. Nvidia lost up to 14% intraday, five days before its quarterly results.

Click to enlarge
Credit: MacGeneration
The timeline spans two weeks. The New York Times published an investigation into a relationship between Jensen Huang and Alice Huang, a former executive at Megaspeed, a Singapore-based cloud provider that the United States suspects of helping China obtain restricted chips. US intelligence officials believe she was recruited by China's Ministry of State Security, which Beijing denies. The board opened an independent investigation on November 8. On Sunday, Jensen Huang acknowledged in an email: "I was unfaithful to my wife." On November 11, the New York Times reported a second relationship, with a subordinate, which Reuters attributes to a second employee of the company (Gulf News). Nvidia says it has "no reason to believe" that national security or company secrets were compromised, and NPR notes it could not independently confirm the reports.
The market reacted in two stages. The stock had hit a six-month low of $165.89 on Tuesday, before dropping 13 to 14% at Thursday's open and closing the session down 4.3%. Nvidia lost nearly $493 billion in market capitalization over the week, for a value of about $4.01 trillion (Tasnim, citing Reuters). Reuters puts the loss at more than $550 billion since the investigation was announced. The stock trades at around 11 to 12 times forward earnings, its lowest level since 2012 according to the same sources.
Analysts are divided on how to read the event. Baird downgraded its rating to "Neutral" with a $200 target, Bernstein cut its own in half (from $400 to $200), and Morgan Stanley removed Nvidia from its top-picks list. Vijay Valecha of Century Financial separates the man from the competitive advantage: in his view, the CUDA developer base, the hyperscalers' locked-in orders and the Blackwell roadmap remain in place. He calls Colette Kress a "stabilizing choice" and the arrival of an independent chair a "meaningful governance upgrade."
Colette Kress is the only executive besides Jensen Huang who regularly speaks on earnings calls. According to The Information, she had pushed back on some deals championed by the founder, including a $250 billion credit backstop proposed to OpenAI (Capital Brief). Donald Trump suggested on social media that Tim Cook succeed Jensen Huang. Nvidia will look for a permanent successor internally and externally, and will publish its fiscal third-quarter results on November 17: Wall Street expects about $109 billion in revenue and $2.47 per share (BigGo Finance). Jensen Huang holds 3.61% of the capital, making him the company's fourth-largest shareholder.
Nvidia is also one of the investors placing billions in Anthropic, whose stock market debut serves as a barometer for the whole sector.
Anthropic Sets Its IPO Range at $70-$78 a Share
Anthropic launched its roadshow on Tuesday, November 10 with 825 million Class A shares offered between $70 and $78, a raise of $57.75 to $64.35 billion and an implied valuation of up to $1.404 trillion. Pricing is set for November 18 and the Nasdaq listing, under the ticker ANT, for November 19.

Click to enlarge
Credit: Reuters / TechCrunch
The roadshow (the presentation tour to institutional investors before pricing) relies on amendment no. 2 to the S-1 form, the document filed with the US securities regulator. About 801 million shares are sold by Anthropic. The implied valuation is between $1.26 and $1.404 trillion on a fully diluted basis (TechCrunch).
A $26 billion private placement of Class N shares accompanies the offering, with AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC, Google and Nvidia committing $10 billion each. The founders hold a Class F share that gives them 50.1% of the voting rights, within Anthropic's public benefit corporation status and its Long-Term Benefit Trust. Compute commitments stand at $900.9 billion, against about $132 billion in cash.
Other News in Brief
The Paris Call on AI safety. On November 9, the Élysée had representatives of eight countries and ten AI companies, including OpenAI, Anthropic and Mistral, sign a text. It sets three priorities: independent evaluators, common standards for reporting major incidents and clearer responsibilities, with no sanctions or timeline (franceinfo, Le Monde).
Gemini 4 Carbon and the Signal proof. Google DeepMind says it used Gemini 4 Carbon to verify security properties of the reference code of the Signal protocol, with about 6.5 million lines of Lean proofs. The scope covers Google's version of the code, not the Signal apps (The New Stack).
Meta Muse Ember. The model offers 1 million tokens of context and up to 512,000 in output. It matches Claude Haiku 5.5 on the Artificial Analysis index (36), with a cost per task of $0.14 against $2.39 for Muse Spark 1.3.
OpenAI opens local inference in India. Since November 12, Indian companies can process some API workloads on servers located in the country.
Conclusion: Dependence on a Single Player Across Chips, Email and Models
A token that dropped 20%, an outage of more than 12 hours, a founder gone after 33 years: several topics this week rest on a single point of support, whether a leader at the head of Nvidia, an email host used by 430 million people or a model provider whose access depends on a change-of-control clause.
Nvidia's results on November 17, the pricing of Anthropic's IPO on the 18th and the listing on the 19th will give the first numbers on how markets value these dependencies.
Alex
- Jensen Huang resigned from Nvidia on November 12, Colette Kress is interim CEO and results land on November 17
- The Microsoft 365 outage hit 74 countries and lasted more than 12 hours, with no root cause published at that point
- Fable 5.2 costs $8 and $40 per million tokens, 20% below Fable 5.1, while Mythos 5.2 remains restricted to verified programs
- OpenAI announced it would cut its models to Cursor, where they account for about 5% of traffic, after SpaceX's $60 billion acquisition
- Anthropic targets $57.75 to $64.35 billion in its IPO, with a Nasdaq listing planned for November 19
Related articles

AI News November 6, 2026: Anthropic at $17 Billion a Quarter, OpenAI Cuts Its Raise and ChatGPT at 20% of Google
Anthropic reports $17 billion in Q3 revenue, OpenAI cuts its raise to $20 billion, Meta borrows at higher yields and ChatGPT claims 20% of Google's query volume.
AI News November 1, 2026: Anthropic Files Its S-1, Gemini 4 Carbon and Mistral Large 4
Anthropic reveals $16.3 billion in revenue in its IPO S-1, Google launches Gemini 4 Carbon, Mistral opens Large 4 and the Nasdaq loses $2 trillion.
AI News October 24, 2026: Anthropic's Claude Duet, Google's Agent Compute Engine, OpenAI on Google Cloud
Anthropic launches Claude Duet, Google unveils Agent Compute Engine, OpenAI lands on Google Cloud, and Muse passes 10 million users.Comments
Comments
Got a take on this article?
Create a free account in 10 seconds to comment, like, and get the next articles straight to your inbox.
Don't have an account yet?