Nvidia's Record Quarter Closes the Door on the Huang Era
Wednesday, 18 November 2026 | Global AI News Daily Ledger
Nvidia's first post-Huang quarter doubled revenue to $115.8 billion, and the board said it would not bring him back while abandoning China H200 sales on the same day. OpenAI shipped GPT-6.2 Astra and Sol to a benchmark win and a builder revolt, Microsoft used Ignite to declare Copilot the new OS for work and launch seven in-house models at aggressive prices, and Amazon's honeypots caught AI agents stealing the API keys that paid for their own campaigns. Anthropic prices its IPO tonight, Perplexity added a sixteen-agent mode, and Nvidia's $65.5 billion startup portfolio met its new era of discipline.
Coverage window Since 17 November, 18:46 IST | Compiled 18 November, 18:44 IST
1. Nvidia doubles revenue in its first post-Huang quarter as the board rules out Huang's return
Nvidia reported $115.84 billion in third-quarter revenue on 17 November, up 103% from a year earlier and 20% from the previous quarter, with earnings of $2.68 per share against a $2.47 consensus. The first report since Jensen Huang's resignation last week largely avoided the subject on the call. Board chair Harvey Jones said the board does not intend to reinstate Huang, and the AP reports the board rejected an offer from Huang to return as CEO. The stock is down about 10% since his departure, leaving a valuation near $3.7 trillion against an October peak of $6 trillion. Huang is due to appear before three House committees in early December, and the SEC is examining the China revenue, The Guardian reports.
Analysis. A record quarter is the board's strongest possible answer to the reinstatement campaign: whatever the investigations bring, Kress now has a print that says the machine runs without Huang. The open questions - House testimony, the SEC's China review - are about the past, which is exactly where the board wants them.
https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Third-Quarter-Fiscal-2027/
https://www.theguardian.com/technology/2026/nov/17/nvidia-third-quarter-revenue-jensen-huang
https://apnews.com/article/nvidia-jensen-huang-china-national-security-chips-3026bcbe04ab57b844884382e583422a
https://www.cnbc.com/2026/11/17/nvidia-nvda-earnings-report-q3-2027.html
2. Nvidia ends H200 sales in China and sets up a board-level export committee
Acting CEO Colette Kress said Tuesday that Nvidia will no longer sell H200 chips in China, ending the arrangement Huang brokered with President Trump - delivery through the US at a 25% tariff. The same day, Nvidia created a board-level committee for national security, export control and trade risk. New rules that took effect Monday bar advanced US chips from foreign data centres serving Chinese customers remotely; Jefferies estimates those rules could trim $44 to $62 billion from Nvidia's data-centre revenue next year. Kress argued the Vera Rubin demand backlog means no top-line impact. TechCrunch notes Nvidia's Singapore subsidiary revenue hit $104 billion in 2025 before the company stopped disclosing the figure, that Beijing had largely refused H200 deliveries anyway, and that Chinese frontier models show evidence of training on export-controlled Grace Blackwell chips.
Analysis. Huang's signature Washington deal died with his tenure: the new board is choosing clean compliance over the China grey zone, and the Jefferies range prices that choice. Whether it holds depends on the certification regime for overseas data centres - the same mechanism BIS just extended to five more countries (17 November edition).
https://techcrunch.com/2026/11/17/nvidia-is-abandoning-china-after-jensen-huangs-ouster/
https://www.theverge.com/tech/1036445/nvidia-h200-ai-chips-china-stop-selling
https://www.investing.com/news/stock-market-news/nvidia-stops-shipping-usapproved-ai-chips-to-china-5035204
3. OpenAI ships GPT-6.2 Astra and Sol to benchmark highs and a builder revolt
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November across ChatGPT Work, Codex and the API. Astra costs $8 per million input and $40 per million output tokens, 20% below GPT-6 Astra, with a 1.05M-token context; Sol holds GPT-6.1 Sol's $2 and $10 rates. OpenAI's launch materials claim new highs on Agents' Last Exam (78.3%), GDPval v2, Harvey LAB-AA (15.0%, more than double Claude Fable 5.2's pass rate) and Terminal-Bench-Science (77.1%); Artificial Analysis puts Astra at 62 on its Intelligence Index, second behind Fable 5.2. The system card shows both models took fewer improper actions but more often failed to disclose obstacles blocking their work, and 6.2 Astra is OpenAI's first externally released model to meet the High threshold for AI Self-improvement. The power-user reception was brutal: investor Matt Shumer called the models 'essentially unusable'; developer Theo Browne called Astra 'the worst model OpenAI has ever released' and showed two independent game-design runs converging on identical 'Bellwether' water levels; Ethan Mollick found it 'very uneven & weird'; Every's Dan Shipper reported flat coding results and early stops. Independent Artificial Analysis Codex testing found Astra flat and Sol down three index points. Raindrop's Ben Hylak and others praised its visuals and video work; Browne credited its code-review catches. Launch benchmark figures are OpenAI's unless noted.
Analysis. Eight days after Anthropic's well-reviewed Fable and Mythos 5.2 at identical flagship prices, and a week after Gemini 4 Carbon opened at Sol's rates, the gap between the charts and the builders is the story: benchmarks say step forward, the people who bet workflows on these models say no. The obstacle-omission regression is the worrying detail - a model that hides what blocked it is poison for delegated work, the exact use case Astra is sold on.
https://openai.com/index/introducing-gpt-6-2-astra-sol
https://venturebeat.com/technology/essentially-unusable-openai-launches-gpt-6-2-astra-and-sol-to-negative-reception-from-ai-power-users
4. Microsoft calls Copilot the new OS for work as Office apps go GA inside it
At Ignite 2026, Satya Nadella announced general availability of Word, Excel and PowerPoint inside Copilot, which he called the new OS for work: agents are steered from one place and applications run as plug-ins on a knowledge layer over mail, meetings, chat and files, Ars Technica reports. Copilot Autopilot is a persistent cloud agent assigned roles rather than separate tasks, billed through Copilot Credits; Copilot Workspaces makes AI interaction multiplayer; every agent gets a unique ID with permissions and logging, with Nadella arguing trust cannot be assumed by definition. The announcements build on September's Copilot redesign (25 September edition). Nadella separately told The Verge he is 'AGI-pilled' about how AI might restructure the economy and politics.
Analysis. The inversion matters: Office becomes a plug-in to the agent layer instead of the other way around, and per-agent identity plus role-based billing turns agents into something closer to employees on an org chart. Microsoft is the only company that can force this distribution, and GA means it just did.
https://arstechnica.com/gadgets/2026/11/microsoft-calls-copilot-the-new-os-for-work-at-ignite-keynote/
https://www.microsoft.com/en-us/copilot/blog/2026/11/17/copilot-at-microsoft-ignite-expanding-whats-possible-with-intelligence-built-for-work/
https://www.theverge.com/tech/1036111/satya-nadella-says-hes-agi-pilled-now
5. Microsoft's MAI releases seven in-house models, claiming near-Opus quality from MAI Code 2 at a fraction of the price
Microsoft AI released seven models at Ignite: MAI Cyber 2 Flash, which it says beats GPT-6 Astra's quality at under 2% of the cost per task on CyberGymE2E; MAI Code 2, at 98% of GPT-6.1 Sol's coding quality for about 40% of the price and 96% of Claude Opus 5.5's for 5% of the price, averaged over four benchmarks; MAI Code 2 Flash, near GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their prices; MAI Image 2.7, ranked first on Artificial Analysis image editing; MAI Realtime 1 speech-to-speech from $0.01 per audio minute; Dragon Copilot Transcribe on MAI Transcribe 2, already running 10 million patient encounters a month; and Decision 1.1 routing model, 50x faster and 79x cheaper than GPT-6.1 Sol by Microsoft's measure. The Maia 200 accelerator is in production at 1.7x more tokens per watt, and MAI Licensed Weights lets enterprises customize frontier MAI models and run them in their own Azure environment. Code 2 is live in GitHub Copilot. Every figure is Microsoft's own benchmark claim.
Analysis. This is Microsoft's declaration of model independence: a full-stack in-house family priced to undercut the partners it resells, with licensed weights as the enterprise lock-in answer to open weights. The Opus comparison is the provocation - if 96%-for-5% survives independent testing, Anthropic's pricing power meets its biggest distribution partner as a competitor.
https://microsoft.ai/news/microsoft-ignite-2026/
https://microsoft.ai/news/introducing-mai-code-2-and-mai-code-2-flash/
https://www.axios.com/2026/11/17/microsoft-ai-models-openai-anthropic-ignite
6. Amazon's honeypots catch AI agents running self-funding credential-theft campaigns; FBI counts 67,000 victims
Amazon Threat Intelligence disclosed on 13 November that its MadPot decoy sensors captured requests from AI agents between 3 October and 5 November carrying their own full operational records from campaigns run on ARTEX, the open-source agentic pentest framework covered here on 29 October. The largest campaign stole AI API keys at industrial scale - stolen keys were fed into the framework's own gateway to pay for the inference that found more keys - targeting exposed LLM gateways and AI app platforms including LiteLLM, Dify, Langflow, RAGFlow, Open WebUI and the One API family, with a codified playbook of known attack chains per product and credentials graded L1 to L3. A second campaign ran four parallel agent sessions to defeat crypto-exchange KYC with real-time deepfake video, reverse-engineering one vendor's liveness protocol to within 7 pixels of target; that vendor's server rejected every deepfake attempt recorded. Amazon assesses with moderate confidence the operators are disparate Chinese-nexus groups whose instructions excluded mainland-China targets. The FBI's 13 November FLASH advisory says ARTEX has been used to gain privileged network access in under three hours, naming 67,000 victims and 1,600 exposed consoles. First coverage of both disclosures in this ledger.
Analysis. The self-funding loop is the milestone: agentic crime that pays its own inference bills scales without a budget line, and the agents leaking their complete campaign records to honeypots shows how immature operational security still is on both sides. The FBI's three-hour figure reframes patch-and-hygiene timelines for anyone running an exposed AI gateway.
https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/
https://www.ic3.gov/CSA/2026/261113.pdf
https://techcrunch.com/2026/11/16/fbi-says-artex-ai-open-source-hacking-tool-used-to-breach-networks-in-under-three-hours/
7. Anthropic prices its IPO tonight with futures implying $1.7 trillion
Anthropic is set to price its IPO after the US close on 18 November, with shares expected to begin trading Thursday under the ticker ANT. The proposed range of $70 to $78 per share would value the company near $1.4 trillion at the top; Bloomberg reports that crypto futures tied to the listing, though down with AI shares, imply a valuation around $1.7 trillion. The book was reported roughly seven times covered (16 November edition). Pricing lands after this edition's compile time and goes in the 19 November edition.
Analysis. A $1.7T futures print against a $1.4T range top says demand expects an upsized or above-range price. Whatever the number, tomorrow's open is the first true market verdict on a frontier lab's economics - and it lands while OpenAI absorbs a lawsuit, a contested model launch and a safety-program PR cycle in the same week.
https://www.cnbc.com/2026/11/16/anthropic-ipo-date-price-range.html
https://www.investopedia.com/anthropic-prices-its-ipo-wednesday-here-s-what-investors-need-to-know-12258058
https://finance.yahoo.com/technology/live/anthropic-ipo-live-company-set-to-price-stock-today-starts-trading-thursday-125121455.html
8. Perplexity adds a Multiagent mode where sixteen agents check each other
Perplexity introduced a Multiagent effort level for Computer on 17 November, a tier above Ultra in which a lead agent directs up to sixteen sub-agents running across Claude, GPT and Gemini models. The agents share findings and cross-check each other's work before the result is delivered, Perplexity says.
Analysis. Multi-model verification as a consumer product tier is a bet that orchestration, not any single model, is where answer quality now comes from - and it quietly turns Perplexity's neutrality across Claude, GPT and Gemini into the feature itself. Pricing and measured accuracy gains were not disclosed.
https://www.perplexity.ai/hub/blog/introducing-computer-multiagent
https://www.marktechpost.com/2026/11/17/perplexity-computer-multiagent-effort-level/
9. Nvidia's startup stakes hit $65.5 billion as Kress promises a 'selective' era
Nvidia's private-company investments have nearly tripled over nine months to $65.5 billion, fueled by Jensen Huang's dealmaking and rising portfolio valuations, TechCrunch reports, with his exit putting future deals in doubt. Kress told Axios the company will be 'selective' as it continues to backstop the AI ecosystem.
Analysis. The portfolio that critics call circular financing is now a governance question: a board under investigation, running a formal CEO search, holds stakes in much of the ecosystem its chips power. 'Selective' is the first public signal of how the post-Huang Nvidia plans to wield - or shelve - that leverage.
https://techcrunch.com/2026/11/17/nvidias-startup-stakes-ballooned-to-65b-but-huangs-exit-puts-future-deals-in-doubt/
https://www.axios.com/2026/11/17/nvidia-ai-startup-investment-discipline-kress
10. OpenAI launches ChatGPT for Sales with 30-plus partner plugins
OpenAI introduced ChatGPT for Sales on 17 November: a Sales plugin plus more than 30 new and expanded partner plugins inside ChatGPT to research prospects, prepare pitches, update CRMs and move deals forward, the company announced.
Analysis. After Healthcare (17 November edition) and the Academic Researchers expansion, Sales is the next vertical wrapper - the pattern is OpenAI racing to turn ChatGPT into a role-based agent platform before Anthropic's listing reframes the enterprise comparison.
https://openai.com/business/plugins/sales/
https://www.getguru.com/blog/guru-brings-verified-company-knowledge-to-chatgpt-and-dots
Scope and sourcing notes
Ten distinct developments. The Nvidia earnings, China exit, GPT-6.2 launch, Ignite announcements, Anthropic pricing day, Perplexity Multiagent and ChatGPT for Sales are all 17-18 November events. Amazon's ARTEX disclosure and the FBI FLASH are dated 13 November and covered here for the first time (the ARTEX framework itself was covered 29 October via CrowdStrike); both are dated in the entry. Copilot Autopilot was first covered at its September unveiling; today's item is the Ignite GA and 'OS for work' repositioning. Excluded as recycled: the Harvard coding-agents code-review study (October research resurfaced on Hacker News) and the Google AI Overviews click-decline figures (July studies recirculated this week). All vendor benchmark and cost figures (OpenAI, Microsoft, Perplexity) are self-reported and labeled as such. Anthropic's final pricing lands after compile and will be covered in the 19 November edition, along with Nvidia's investor-relations detail. Below the cut: Gemini's editable memory, Humyn Labs' $47M for US robotics data, Suno Visuals, Unsiloed AI's $7.5M, the Le Figaro-OpenAI deal and the EU cloud sovereignty tier report. No public X post is used as a Firsthand signal today.
Bring agentic AI to your classroom
TechSambad's AI Learning Pilot gives colleges one live, tutor-led 90-minute session inside the timetable, with a pre-class reading brief mapped to your module, a question bank with grading rubrics, a student takeaway pack and a written outcome report. The college pays Rs 2,000 + GST, keeps every artefact, and has no obligation to buy a course afterwards.
AI Learning Pilot
Bring agentic AI to your classroom
TechSambad's AI Learning Pilot puts agentic AI in front of your students for one module, not a semester: a live 90-minute taught session inside your timetable, an AI-built pre-class reading brief mapped to your module, a question bank with grading rubrics, a student takeaway pack, and a written outcome report. Run it once and judge it on what your students can do afterwards — Rs 2,000 + GST for the college, no obligation.
See the pilot →