Nvidia falls 22% as AI stocks lose $4 trillion in a week

A sharp sell-off in AI-linked stocks sent markets lower during the week of November 16–20. On Friday, November 20, the S&P 500 fell 3.7% and entered a correction after dropping more than 10% from its October 6 record; the Nasdaq Composite lost 5.28% and turned negative for 2026. Nasdaq companies shed more than $4 trillion in market value during the week, while chip stocks lost about $1 trillion in one week. The iShares Semiconductor ETF fell 10% on Friday and 34.5% since September 30, its steepest two-month slide in 25 years. Nvidia fell 13% Friday and 22% for the week, and Advanced Micro Devices, Micron Technology and Broadcom also dropped sharply. Apple was down 4% for the week. New US restrictions on advanced chip sales to China, reports of an SEC investigation involving Nvidia and doubts about the sustainability of AI spending added to investor concerns. Losses also hit Asian and European markets, with Japan’s Nikkei and South Korea’s Kospi each down more than 5%; gold and bitcoin fell as well.

Anthropic’s $81.34 billion IPO, priced at $83 a share, came amid the market turmoil. Its shares rose as high as $111 after opening Thursday, then fell on Friday: accounts in the reports put the close at $88.30 or $88.90, still above the offer price but giving back nearly all debut gains. A separate report described the IPO as a record and said the stock had fallen back toward its offering price. Federal Reserve meeting minutes also showed that almost all officials viewed further rate increases as likely appropriate. Market commentators Thomas Martin of Globalt Investments and Scott Bauer of Prosper Trading Academy discussed rising volatility and defensive trading strategies on “The Claman Countdown.”

South Korea is pushing to expand AI infrastructure. President Lee Jae Myung urged faster investment during a visit to an SK Group and Amazon Web Services data centre under construction in Ulsan. The project carries an estimated cost of 7 trillion won, or $5.2 billion, through 2029 and is expected to begin operating in the second half of 2027. SK Chairman Chey Tae-won said it initially plans 103 megawatts of capacity by 2029, with possible expansion of nearly 900 megawatts. Lee said the centre could attract investment and support growth beyond Seoul, and called for plans to train workers to build data centres and AI infrastructure.

Companies are also putting AI to work in operations and software development. NextEra Energy says its Grid Composer tool has saved Florida Power & Light customers more than $20 million so far this year by analyzing about half a trillion data points daily to recommend generation, fuel and maintenance decisions. Its Optos software suite, developed with Google Cloud’s Gemini Enterprise Agent Platform, combines six utility functions, and NextEra says it has more than 100 AI initiatives underway. Google Cloud has added AI agents to Managed Service for Apache Airflow: agents can investigate failed tasks, suggest environment improvements and check DAG readiness for Airflow upgrades, while coding agents can write, migrate and deploy DAGs.

Other AI tools target specialized workflows. Hebbia says GPT-6 Astra met 97.9% of presentation requirements in its tests, with 99.5% of cited excerpts deemed relevant; within Hebbia’s Matrix platform, it can find sources, draft answers and create cited PowerPoint slides. The company says Astra beat the best alternative by 13.6 percentage points on citing required source types, while noting bankers still need to check decks. Phemex introduced an AI tool for drafting trading orders, presenting entry price, take-profit and stop-loss levels, position size and leverage in an editable card; users must confirm before submission.

Nvidia said 20,000 post-training examples created with Factory’s Droid harness helped improve Nemotron models’ planning, tool use and error recovery. Nemotron 3 Ultra’s Terminal-Bench 2.0 score rose from 38.5% to 49%, while a preliminary Nemotron 3.5 Super VL checkpoint improved from 52.11% to 60.4% on Terminal-Bench 2.1. Separately, cybersecurity experts warned that AI can help identify software vulnerabilities but can also help attackers exploit them. They cautioned that rushed patches may create new problems or outages, recommended testing updates on a smaller scale first, and stressed understanding dependencies between systems.

In India, Karnataka’s Higher Education department is testing AI facial-recognition attendance at a few government colleges. Teachers would record attendance through a mobile app, with updates sent to parents’ registered numbers after each entry. Officials say the proposal aims to reduce absences and missed exams linked to attendance shortages. The state chose software over classroom biometric equipment because it requires a one-time design investment, and plans to expand the system to government degree colleges and state-run universities from the next semester.

Key Takeaways

  • The November 16–20 AI-linked sell-off pushed the S&P 500 into correction territory and the Nasdaq into negative territory for 2026; Nasdaq companies lost more than $4 trillion in market value during the week.
  • The iShares Semiconductor ETF fell 10% Friday and 34.5% since September 30; Nvidia dropped 13% Friday and 22% for the week, while AMD and other chipmakers also declined.
  • Anthropic raised $81.34 billion in an IPO priced at $83 per share; its Friday close was reported at $88.30 or $88.90 after an early rise to $111.
  • South Korea’s SK Group and Amazon Web Services are building a $5.2 billion data centre in Ulsan, targeting operations in the second half of 2027 and 103 megawatts of initial capacity by 2029.
  • NextEra says its Grid Composer AI has saved Florida Power & Light customers more than $20 million this year, processing about half a trillion data points daily.
  • Google Cloud added AI-agent support to Managed Service for Apache Airflow for troubleshooting, environment recommendations and DAG development, migration and deployment.
  • Hebbia says GPT-6 Astra met 97.9% of presentation requirements in its tests and produced citations judged relevant 99.5% of the time; users still need to verify presentations.
  • Phemex’s AI order-drafting tool assembles key trade parameters in an editable card, but requires user confirmation before submission.
  • Factory’s 20,000 training examples helped Nvidia raise Nemotron 3 Ultra’s Terminal-Bench 2.0 score from 38.5% to 49%; a preliminary Nemotron 3.5 Super VL checkpoint reached 60.4% on Terminal-Bench 2.1.
  • Experts warn AI-enabled vulnerability discovery can increase pressure to patch quickly, while rushed updates may cause outages; Karnataka is separately testing AI facial-recognition attendance at government colleges.

Semiconductor stocks suffer worst slump since 2001 as AI trade falters

The iShares Semiconductor ETF fell 10% on Friday and 34.5% since September 30, nearing its worst two month decline in 25 years. Nvidia dropped 13% on Friday and lost 22% for the week, while other major chipmakers also fell sharply. The wider sell-off pushed the S&P 500 into a correction and erased about $1 trillion in market value from chip stocks in one week. New US limits on advanced chip sales to China, investigations involving Nvidia and concerns about AI spending added to investor worries. Anthropic, which raised $81.3 billion in a record IPO, also fell back toward its $83 offering price after its debut.

AI stock sell-off pushes Nasdaq into the red for 2026

A sharp sell-off in AI stocks on Friday dragged the S&P 500 down 3.7% and sent the Nasdaq Composite into negative territory for 2026. The S&P 500 entered a correction after falling more than 10% from its October 6 record, and its seven-day losing streak was the longest since 2020. Chipmakers including Nvidia, Micron Technology, Broadcom and Advanced Micro Devices led the declines, while Corning and Bloom Energy also fell. Japan’s Nikkei and South Korea’s Kospi each lost more than 5%, and European markets also dropped. Investors questioned whether the high values and heavy spending tied to AI were sustainable, while Anthropic shares fell 9.1% to $88.30, still above their $83 IPO price.

AI-linked stocks drive one of Wall Street’s worst weeks in years

US stocks fell sharply on Friday, November 20, as investors sold companies linked to artificial intelligence. The S&P 500 dropped 3.7% that day and more than 10% over the month, entering a correction, while the Nasdaq Composite fell 5.28% and turned negative for the year. Nasdaq companies lost more than $4 trillion in market value during the week, and all of the Magnificent Seven stocks declined, with Apple down 4% for the week. Anthropic shares, which opened at $83 on Thursday and rose to $111 before closing at $97, fell to $88.90 on Friday, erasing nearly all their debut gains.

AI sell-off hits global markets after Anthropic’s record IPO

Stocks fell across global markets during the week of November 16 to 20, as investors sold AI-linked companies. Anthropic raised $81.34 billion in an IPO priced at $83 per share, but its shares closed Friday at $88.30 after dropping 9.1% that day. Nvidia also continued to slide, despite reporting results, as its leadership change and reports of an SEC investigation drew attention. Losses spread from Asia to Europe, while gold and bitcoin also fell. The report also said Federal Reserve meeting minutes showed that almost all officials viewed further rate increases as likely appropriate.

South Korea’s Lee urges speed in AI race during data centre visit

South Korean President Lee Jae Myung urged the country to move quickly in the global AI race during a Saturday visit to a data centre being built in Ulsan. SK Group is developing the centre with Amazon Web Services, with an estimated cost of 7 trillion won, or $5.2 billion, through 2029. The facility is scheduled to begin operating in the second half of 2027. SK Group Chairman Chey Tae-won said the project initially plans 103 megawatts of capacity by 2029, with a possible expansion of nearly 900 megawatts. Lee said the project could help bring investment and growth to regions outside Seoul, and called for plans to train workers to build data centres and AI infrastructure.

South Korea’s Lee pushes faster AI investment at $5.2 billion data center

South Korean President Lee Jae Myung called for faster investment in AI infrastructure during a visit to a $5.2 billion data center project in Ulsan on Saturday. SK Group and Amazon Web Services are jointly developing the facility, with about 7 trillion won in investment planned through 2029 and initial operations expected in the second half of 2027. SK Group Chairman Chey Tae-won has outlined plans for 103 megawatts of computing capacity by 2029, with possible later expansion of nearly 900 megawatts. Lee said the project could attract more global AI investment and support regional growth, and asked government ministries to study training for data center workers.

Investors examine the worsening AI stock sell-off

On November 20, 2026, Globalt Investments senior portfolio manager Thomas Martin and Prosper Trading Academy CEO Scott Bauer discussed the AI-driven stock sell-off on “The Claman Countdown.” They analyzed rising market volatility and shared defensive trading strategies for navigating the downturn.

AI finds more software flaws, raising concerns about outages

Cybersecurity experts say AI is helping uncover more software vulnerabilities, creating pressure on companies to fix flaws quickly. Experts cautioned that there is no clear evidence connecting recent technology disruptions. AI can help find weaknesses, but attackers can also use it to exploit them, while rushed software patches may cause new problems. Darrell West recommended testing updates on a smaller scale before applying them widely, and experts also stressed understanding how systems depend on one another. Cybersecurity expert Powers said faster efforts to patch flaws could potentially lead to more outages.

NextEra says AI has saved Florida Power & Light customers $20 million

NextEra Energy says its Grid Composer AI tool has saved Florida Power & Light customers more than $20 million so far this year. The system analyzes about half a trillion data points each day, including equipment readings, customer demand and plant output, to recommend power generation, fuel and maintenance choices. NextEra built the first version in less than 12 weeks and later introduced the technology to other utilities through its Optos software suite, developed with Google Cloud’s Gemini Enterprise Agent Platform. Optos Composer combines six functions, including running plants, buying fuel, scheduling maintenance, trading energy, managing backup capacity and overseeing battery storage. NextEra says it has more than 100 AI initiatives underway.

Google Cloud adds AI agents to help manage Airflow pipelines

Managed Service for Apache Airflow, formerly Cloud Composer, now offers AI agent support across a pipeline’s lifecycle. In the Google Cloud console, agents can investigate failed tasks, suggest environment improvements and check whether DAGs are ready for a newer Airflow version. In an IDE or coding agent, they can write, migrate and deploy DAGs. The article explains the features through a fictional company called Cymbal Shops.

Hebbia says GPT-6 Astra reduces presentation rework

Hebbia says GPT-6 Astra met 97.9% of presentation requirements and 99.5% of its cited excerpts were relevant in the company’s tests. Hebbia tested a prerelease build in August and September 2026, using financial research cases and the same grading criteria for each model. Inside Hebbia’s Matrix platform, the system can find sources, draft answers and create PowerPoint slides with citations linked to source passages. Hebbia says Astra led the best alternative by 13.6 percentage points on citing required source types. Bankers still need to check details before presenting a deck.

Phemex adds AI tool to draft trading orders

Phemex introduced an AI-powered tool that helps users draft trading orders. An editable card brings together the entry price, take-profit and stop-loss levels, position size, and leverage. The user must confirm the choices before the order is submitted. Phemex did not provide further details about the tool’s availability or other features.

Nvidia boosts Nemotron coding scores with 20,000 training examples

Nvidia used Factory’s Droid harness to create 20,000 rows of post-training data for Nemotron 3 Ultra. Factory found issues with tool calls, argument formatting, and cases where the model continued working without an error. Nvidia said the work helped improve the model’s planning, tool use, and error recovery. Nemotron 3 Ultra’s Terminal-Bench 2.0 score rose from 38.5% to 49%. A preliminary Nemotron 3.5 Super VL checkpoint rose from 52.11% to 60.4% on Terminal-Bench 2.1.

Karnataka plans AI attendance updates for college students’ parents

Karnataka’s Higher Education department has proposed AI-based facial recognition attendance for undergraduate and postgraduate students at government colleges. Teachers will record attendance with a mobile app, and updates will go to parents’ registered mobile numbers after each entry. Officials say the system aims to reduce absences and missed exams linked to attendance shortages. The department chose the AI system over biometric equipment because it requires a one-time investment to design software rather than equipment in every classroom. The system is being tested at a few colleges, and the state plans to expand it to government degree colleges and state-run universities from the next semester, according to collegiate education commissioner H N Gopalakrishna.

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 stocks stock market volatility semiconductor industry Nvidia AI investment Anthropic IPO AI infrastructure data centers South Korea NextEra Energy Google Cloud Apache Airflow AI agents Hebbia Phemex Nvidia Nemotron AI cybersecurity facial recognition

Comments

Loading...