AI Sell Off Hits US Stocks as AMD and Nvidia Slip

AI-related selling jolted U.S. markets, with the Nasdaq recording its largest one-day drop since June 2022 in one account and entering correction territory. The Dow and S&P 500 also crossed the 10% decline threshold from recent highs, while another report said the Nasdaq’s Friday loss erased its 2026 gains. Investors cited worries about AI valuations alongside inflation, interest rates and the global economy. The sell-off also reached precious metals: gold fell 3% Friday and was down about 9.4% for the week, while markets priced in a 96% chance of a 25-basis-point December rate hike. In after-hours trading, Oracle rose 2.3%, AMD gained 1.3% and Nvidia advanced 1.1%.

Alibaba is training next-generation AI models on its own chips as it seeks to reduce reliance on advanced U.S. processors. For the quarter ended September 30, revenue rose 10% to about 271.5 billion yuan and net income increased 213% to 64.4 billion yuan. AI Cloud and Compute Services revenue climbed 53% to 61 billion yuan, while external AI cloud and computing product revenue rose 67%. The company reported 50.48 billion yuan in capital spending and a 34.47 billion-yuan free-cash-flow outflow; adjusted earnings per ADS missed estimates. Management targets positive free cash flow in the second half of 2028, expects cloud revenue growth of 30% to 45% next year and aims for $100 billion in external cloud revenue by 2030. Baidu’s AI Cloud Infra revenue rose 94% to 8.2 billion yuan.

Baseten introduced bison-1 and groundhog-1, open-weight decision models built on Google’s Gemma 4. They return probabilities for yes-or-no, choice and scoring questions, support images and handle contexts up to 128,000 tokens. Bison-1 ranked first among 19 models on the Artificial Analysis Decision Index v2.1, scoring 76 and achieving the lowest Brier score, 0.14. Baseten lists prices of $0.10 per million tokens for bison-1 and $0.035 for groundhog-1. Separately, Google’s Chrome security team developed Mirror Dimension, a simulated internet for repeatable testing of browsing agents. It reproduces activities such as shopping and banking and lets researchers test prompt injection attacks without affecting real sites. Google says its automated system generates more than 1,000 test candidates per security engineer daily, over 30 times the manual rate.

OpenAI made GPT-Live-1.1 available through its API as a multilingual voice model that can listen while speaking. The company says it offers more consistent accents and more reliable task completion, but has not published scores supporting those improvements. A standby billing option cuts idle WebRTC session costs from $0.05 to $0.02 per minute, with idle timeouts adjustable from 5 to 60 seconds. The model is API-only, and switching an existing live session requires starting a new one; backend reasoning and tools carry separate charges. The articles also refer to ChatGPT in the broader AI context, but provide no separate ChatGPT product update.

In Austin, school trustees adopted the district’s first local AI policy and gave the superintendent up to 18 months to draft detailed rules for staff and students. Proposed topics include classroom and administrative use, training and vendor requirements. The policy says students should know when they are interacting with AI and establishes an advisory committee that includes parents and a conflict-of-interest rule. Classroom use remains limited to a small pilot, though the board president said teachers and students report that students use AI; staff are due to report in January on provisions that could take effect sooner.

AI is also prompting new approaches to company data and auto retail. Carbon Arc, led by former Point72 market-intelligence head Kirk McKeown, says its platform holds about 600 alternative datasets, including credit-card spending, retail sales and healthcare claims. Customers can query data with AI agents and pay for what they use, with providers receiving a share. McKeown sees potential for recurring revenue from company data, but privacy, ownership and valuation remain obstacles. Spyne’s report says most U.S. dealerships use AI, yet few can show it boosts sales. It recommends tracking qualified appointments, conversion and inventory turnover. In Q3 2026, new-car financed payments averaged a record $787 monthly and the average amount financed was $44,664; loans of at least 84 months accounted for 25.5% of purchases. Cox Automotive found 63% of surveyed shoppers would probably or definitely use AI to buy their next vehicle, while 29% of dealers had begun adapting to AI-powered search.

Key Takeaways

  • U.S. stocks suffered a sharp AI-linked sell-off: the Dow and S&P 500 entered correction territory, and reports said the Nasdaq also entered correction territory and erased its 2026 gains.
  • Gold fell 3% Friday and about 9.4% for the week as AI-stock losses prompted some investors to sell gold; markets put the odds of a December 25-basis-point rate hike at 96%.
  • Alibaba is training next-generation AI models on its own chips. AI Cloud and Compute Services revenue rose 53% to 61 billion yuan, while the company reported a 34.47 billion-yuan free-cash-flow outflow and targets positive free cash flow in the second half of 2028.
  • Alibaba aims for $100 billion in external cloud revenue by 2030; Baidu reported 94% growth in AI Cloud Infra revenue to 8.2 billion yuan.
  • Baseten’s Gemma 4-based bison-1 ranked first among 19 models on the Artificial Analysis Decision Index v2.1; bison-1 and groundhog-1 support images and up to 128,000 tokens of context.
  • Google’s Mirror Dimension simulates websites to test browsing agents and prompt-injection defenses; Google says it creates more than 1,000 test candidates per security engineer each day.
  • OpenAI’s API-only GPT-Live-1.1 adds multilingual voice capabilities; standby idle WebRTC billing falls from $0.05 to $0.02 per minute.
  • Austin ISD set an 18-month deadline for detailed AI rules and adopted a policy requiring students to know when they are interacting with AI.
  • Carbon Arc’s platform offers about 600 alternative datasets for AI-agent queries, but privacy, data ownership and valuation remain unresolved hurdles.
  • Spyne says dealerships need to measure AI against sales outcomes: Q3 2026 average financed new-car payments reached $787 monthly, and 63% of surveyed shoppers said they would probably or definitely use AI to buy their next vehicle.

AI concerns send US stocks into a sharp decline

U.S. stocks fell sharply as investors worried that the boom in artificial intelligence shares could be fading. The Dow Jones Industrial Average dropped 949.73 points, or 1.93%, to 48,147.33, moving into correction territory after falling more than 10% from its August record. The S&P 500 fell 265.44 points, or 3.70%, to 6,910.39, also entering a correction after its October peak. The Nasdaq Composite lost 1,261.47 points, or 5.28%, and came close to a bear market, which means a fall of at least 20% from a peak.

Nasdaq plunges as AI shares and major stocks fall

The Nasdaq Composite fell 1,261.64 points, or 4.8%, to 24,841.51, while the S&P 500 fell 3.4% and the Dow Jones Industrial Average dropped 3.2%. Anthropic shares fell 45.6% to $1.23, and other AI stocks also dropped sharply. Concerns about inflation, interest rates and the global economy contributed to the sell-off, which affected all 11 major S&P 500 sectors. The Nasdaq recorded its largest one-day drop since June 2022, extending a decline that began Friday and pushing the index into correction territory.

AI stock sell-off wipes out Nasdaq gains for 2026

A sharp sell-off in AI stocks pushed the Nasdaq Composite down almost 4% Friday, erasing its gains for 2026. The index had been up almost 19% for the year after reaching a record close on October 6. The Dow Jones Industrial Average and S&P 500 also entered correction territory, meaning they had fallen more than 10% from recent highs. MarketEdge said the S&P Oscillator, a momentum measure, reached minus 8% during Friday's trading, showing the strongest selling pressure since March.

Alibaba trains AI models on its own chips as cloud sales grow

Alibaba CEO Eddie Wu Yongming said the company is already training its next-generation AI models on its own chips. The move aims to reduce reliance on advanced U.S. processors as export restrictions limit access, and supports Alibaba's AI and cloud businesses. For the quarter ended September 30, revenue rose 10% to 271.5 billion yuan, while net income climbed 213% to 64.4 billion yuan. Revenue from its AI Cloud and Compute Services segment increased 53% to 61 billion yuan, and external revenue from AI cloud and computing products rose 67%. Wu said Alibaba expects cloud revenue growth of 30% to 45% next year and aims to reach $100 billion in external cloud revenue by 2030.

Alibaba cloud grows 53% as spending weighs on cash flow

Alibaba's AI cloud revenue rose 53% in the September quarter, but its free cash flow showed a 34.47 billion yuan outflow as the company increased spending on cloud infrastructure. Revenue rose 10% to 271.52 billion yuan, and AI-related product revenue grew by triple digits for the 13th quarter in a row. Alibaba reported 50.48 billion yuan in capital spending, while adjusted earnings of 8.10 yuan per ADS missed estimates of about 11 yuan. Chief Financial Officer Toby Xu said the company is targeting positive free cash flow, and management expects to reach it in the second half of 2028. Baidu's AI Cloud Infra revenue grew 94% to 8.2 billion yuan, while GDS Holdings was due to report on November 25.

Baseten’s New Decision Models Lead the Benchmark

Baseten released bison-1 and groundhog-1, two open-weight decision models built on Google’s Gemma 4. They return probabilities for yes or no, choice, and score questions instead of writing text, and both support images and context up to 128,000 tokens. Bison-1 scored 76 on the Artificial Analysis Decision Index v2.1, ranking first among 19 models, and had the index’s lowest Brier score at 0.14. The models are available through Baseten, with bison-1 priced at $0.10 per 1 million tokens and groundhog-1 at $0.035 per 1 million tokens.

Austin ISD Sets 18 Months to Create Its First AI Rules

Austin ISD trustees adopted the district’s first local policy on artificial intelligence, giving the superintendent up to 18 months to write detailed rules for students and staff. Trustee Arati Singh proposed rules covering classroom and administrative use, training, and vendor requirements. The policy also says students should know when they are speaking with AI and creates a district AI advisory committee that will include parents and a conflict-of-interest rule. District staff said classroom AI use is limited to a small pilot, but Board President Lynn Boswell said teachers and students have told her that students use AI. Boswell asked staff to report in January on which parts of the policy can take effect quickly.

Google Creates a Fake Internet to Test Chrome AI Agents

Google’s Chrome security team built Mirror Dimension, a simulated internet for testing the safety and performance of AI agents that browse websites and complete tasks. It recreates activities such as shopping, banking, and social media use so researchers can repeat the same tests without affecting real websites or accounts. The system can test defenses against prompt injection attacks by placing malicious instructions in webpage text, images, accessibility labels, and tool responses. Google says its automated system produces more than 1,000 test candidates per security engineer each day, over 30 times what an engineer could create manually. The company plans to expand access to Mirror Dimension across its research teams.

AI Could Turn Company Data Into a New Wall Street Asset

Kirk McKeown, a former head of market intelligence at Point72 and now CEO of alternative data company Carbon Arc, says AI could change how investors use and buy data. Carbon Arc gathers information such as credit card spending, retail sales, and healthcare claims, and says its platform holds about 600 datasets. Customers can use AI agents to ask questions about company performance and pay for the data they use, while the businesses that provide it receive part of the revenue. McKeown believes companies could eventually earn recurring income from data they create and use that income to support financial deals. Privacy rules, ownership rights, and the challenge of valuing data remain major hurdles.

OpenAI Adds Cheaper Standby Billing for GPT-Live-1.1

OpenAI made GPT-Live-1.1 available through its API as a multilingual voice model that can listen while speaking. The company says the model has more natural, consistent accents across languages and completes tasks more reliably. A new standby billing mode lowers idle WebRTC session costs from $0.05 to $0.02 per minute, and users can set an idle timeout from 5 to 60 seconds. The model is API-only, and switching an existing Live session to gpt-live-1.1 requires starting a new session. OpenAI has not provided scores for its reported voice, transcription, and task improvements, and backend reasoning and tools have separate charges.

Gold Falls as AI Stock Sell-Off and Rate Hike Bets Weigh

Gold fell 3% to $3,514.08 per ounce on Friday, after reaching $3,502.49, its lowest level since September 2025. It has dropped about 9.4% this week, its biggest weekly decline since the week ending March 20. High Ridge Futures’ David Meger said losses in AI stocks were forcing some investors to sell gold to cover losses. Federal Reserve Board member Anna Paulson warned about inflation risks, and markets put the chance of a 25-basis-point U.S. rate hike in December at 96%. Silver, platinum and palladium also fell, leaving all three metals on track for a second straight weekly loss.

Oracle, AMD and Nvidia Rise in After-Hours Trading

AI-related stocks broadly rose in U.S. after-hours trading, according to ChainCatcher, citing RootData market data. Oracle gained 2.3%, AMD rose 1.3%, and Nvidia advanced 1.1%.

Spyne Report Finds AI Use at Most Dealerships but Limited Sales Growth

Spyne’s report, AI in U.S. Auto Retail: The Widening Measurement Gap, says most U.S. dealerships use AI, but few can show that it increases sales. It urges dealers to measure results such as qualified appointments, sales conversion and inventory turnover, rather than counting automated interactions. In Q3 2026, the average monthly payment on a financed new vehicle reached a record $787, while the average amount financed was $44,664; loans of 84 months or longer made up 25.5% of purchases. Cox Automotive research found that 63% of surveyed shoppers would probably or definitely use AI to buy their next vehicle, while 29% of dealers had started adapting to AI-powered search. Spyne recommends connecting customer conversations with sales follow-up and inventory decisions, while using accurate vehicle details, pricing and availability.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI market sell-off AI valuations Alibaba AI chips AI cloud computing Baidu AI cloud Baseten bison-1 groundhog-1 Google Gemma AI decision models Google Mirror Dimension AI agent security prompt injection OpenAI GPT-Live-1.1 multilingual voice AI AI education policy alternative data AI agents AI in automotive retail AI adoption measurement

Comments

Loading...