Top AI stories · Saturday, Nov 7, 2026
The Nasdaq Composite rose 1.09% on Friday as dip buyers stepped into semiconductor stocks, though the index still ended with a fourth straight weekly loss and the S&P 500 posted a 2.7% weekly drop, its worst since April 2025. The Philadelphia Semiconductor Index climbed 4.21% on the day, led by AMD jumping 11% to $449.04, Lam Research gaining 8%, Intel rising 7.7%, Micron advancing 5.1% to $690 and Nvidia adding 2.8% to $186.64. For the session, the S&P 500 edged up 0.31%, while the Dow Jones Industrial Average fell 237 points, or 0.46%. Friday's gains followed a stronger than expected October payrolls report showing the US added 165,000 jobs versus 65,000 expected and unemployment fell to 4.1%, which pushed the 10 year Treasury yield to 4.87% as traders priced a 77% chance of a December Fed rate hike. The bounce came a day after a post-midterm rout sent the Nasdaq Composite into correction territory and drove the Philadelphia Semiconductor Index down 12.25% on fears of tighter AI regulation and data center construction pauses. Voters approved restrictions or moratoriums in nearly all of the more than 40 local data center ballot measures on Tuesday, while Democrats who campaigned on pausing new projects won governorships in key buildout states including Texas, Georgia and Ohio. Goldman Sachs warned that short-term state and local moratoriums could affect 45% of planned US data center capacity, though Texas Governor-elect Gina Hinojosa said her proposed pause would apply to new permits rather than projects already under construction. By Wednesday morning, senior OpenAI leaders were discussing whether to reset 2027 data center capacity expectations, the Wall Street Journal reported, while Sen. Bernie Sanders reiterated calls for a national AI data center moratorium and a ban on superintelligence.
Anthropic plans to proceed with its initial public offering roadshow next week and still targets a Nasdaq listing during the week of Nov. 16, before Thanksgiving, The Information reported. The Claude developer is moving forward after potential investors raised doubts earlier on Nov. 6 to Reuters and the Wall Street Journal about whether a post-election rout in artificial intelligence stocks would delay next week's marketing kickoff or shrink the offering toward $80 billion. The startup expects to raise $80 billion to $100 billion at a valuation of $1.25 trillion to $1.75 trillion, centered on $1.5 trillion, the Financial Times reported. That target follows a 6.4% drop in the Nasdaq 100 on Nov. 5 that erased $2.3 trillion in market value, after which prospective IPO buyers discussed valuations of $1.2 trillion to $1.3 trillion and as low as $1.1 trillion, down from roughly $2 trillion in early October, Bloomberg reported, with one major investor telling the Wall Street Journal the mark could land closer to $1 trillion. Anthropic's annualized revenue run rate topped $80 billion at the end of October, the Wall Street Journal reported, after an amended S-1 filing on Nov. 2 showed third-quarter revenue grew 47% from the prior quarter to $16.96 billion, slowing from 144% growth in the second quarter. The filing also disclosed a $567 million third-quarter operating loss and $895.4 billion in compute and infrastructure commitments stretching to 2045, including $69.8 billion due within 12 months, against $131.9 billion in cash and short-term investments as of Sept. 30.
Google will allow K-12 school administrators to disable specific artificial intelligence tools, such as Gemini and NotebookLM, for student accounts without entirely losing access to the company's web search engine. Under the new policy, which takes effect early 2027, Google will set its search and AI services to be off by default for students, requiring schools to manually enable the features they want to provide. The change comes after early reports indicated that schools blocking AI features entirely would be barred from using Google Search. The company continues to state that search and AI are deeply integrated, meaning that students will still encounter AI-generated overviews and guided learning tools even if standalone chatbots are disabled. This technical link prompted criticism from educators and politicians, including Rep. Greg Casar and former FTC chair Lina Khan, who suggested the remaining integrated features effectively force schools to adopt Google's AI ecosystem.
The U.S. government granted export licenses for the H200 processor to approximately 24 companies in China to assemble the chips into artificial intelligence systems, though some of those authorized appear to be shell companies. A New York Times investigation found that the listed offices for firms such as New Matrix and Maginfra were empty. New Matrix's owners were linked to the blacklisted Wise Road Capital, while Maginfra shares personnel and patents with the blacklisted server maker Inspur. The reporting coincides with a request from Republican chairs of three House committees for a Nov 18 briefing with Nvidia CEO Jensen Huang concerning his connections to an alleged Chinese intelligence asset. While Washington has maintained strict export controls to prevent China's military modernization, the company's chief executive previously lobbied the administration to allow the sale of these advanced chips. Nvidia has stated that its dealings with Chinese entities were routine business and involved no exchange of sensitive information.
The world's largest contract chipmaker is examining ways to expand into advanced memory production, including the possible acquisition and conversion of a manufacturing facility. TSMC is currently drafting plans to produce 3D DRAM, a type of high-density memory that stacks cells vertically, at the request of its largest customer, Nvidia, to mitigate AI-driven hardware shortages and diversify its revenue streams beyond logic chips. The move would introduce a fourth major source of advanced memory into a market currently dominated by Samsung, SK Hynix, and Micron, potentially giving Nvidia more bargaining power over pricing in future supply contracts. Industry analysts suggest that because TSMC lacks existing DRAM intellectual property, mass production would be a multiyear undertaking and unlikely to solve current inventory gaps. TSMC shares rose 6.9% in premarket trading on Nov. 6 following the report.
The developer of ChatGPT has updated its career site to identify its data processing facilities as "AI compute campuses" to reduce friction with the communities where they are located. To counter public opposition to the environmental and infrastructure costs of its data centers, OpenAI is proposing new state legislation and community compacts intended to give local residents a role in determining project benefits and construction terms. This rebranding is part of a strategy to influence U.S. regulation during the lame-duck session starting Nov. 9, where OpenAI will lobby for mandatory speed limits on the most powerful AI models. As the company prepares for a potentially more restrictive House of Representatives, it has also begun outreach to 10 prospective 2028 presidential candidates to steer the political framing of artificial intelligence issues.
Britain's government will proceed with its plans to expand computing infrastructure and artificial intelligence capabilities regardless of the political climate in the United States. The growth minister stated that the country remains committed to accelerating its progress in the sector, even as voters across the Atlantic increasingly oppose the construction of data centers. The government is currently targeting 6GW of computing capacity by 2030. To meet this goal, the UK plans to provide priority grid connections for AI facilities and implement measures to reduce energy costs for operators.
The credit risk for technology firms expanding their artificial intelligence capacity is climbing as a new analysis reveals a vast majority of their infrastructure obligations remain hidden from corporate balance sheets. J.P. Morgan estimates that the seven largest AI companies have pledged $4.5 trillion for future data center leases and hardware, with just 15% of those total liabilities recorded as debt. This disparity has caused the cost to insure the companies' debt against default to rise, while hedge funds have reduced their exposure to the sector by 8%, the steepest drop since 2016. The largest of these off-book obligations belong to Google, at $1.08 trillion, followed by Meta at $810 billion and Microsoft at $575 billion. For some smaller players, the gap is more extreme; Anthropic's off-balance-sheet commitments are more than 100 times its recorded debt. These multiyear agreements secure scarce chips and power capacity, including data center leases that can extend as long as 30 years. Among the five largest operators, roughly 70% of their $1.3 trillion in future lease commitments were added within the last year.
The involvement of Brian Chesky in a privately funded artificial intelligence venture will not alter his responsibilities as chief executive of Airbnb, the company announced. Airbnb also stated that the duties of former Meta generative AI leader Al-Dahle will remain the same, despite his ownership of equity in the separate startup. The stealth lab previously paid Airbnb $7.1 million in cash for expense reimbursements and the use of intellectual property. While Chesky retains a controlling interest and has self-funded the operation to date, Airbnb has secured a future option to license the venture's technology or invest capital into the business.
The research team at Hugging Face has demonstrated that reasoning models can be adapted into autonomous agents using specialized reinforcement learning and lightweight LoRA adapters. In a newly published study, the team used webAI's TwIL-8B model, finding that it outperformed Liquid AI's LFM3-1.7B across three different agent harnesses during preliminary baseline evaluations, even though TwIL-8B was not explicitly trained for agentic tool use. The findings build on the Nov. 4 release of TwIL-8B, an open-weight model with 8.8B parameters built on IBM's Granite 4.2 8B. Designed to run on consumer CPUs and graphics cards via a 4.98 GiB quantized file, the model was intended to reduce reliance on expensive proprietary APIs by matching 93% of frontier performance on coding and 90% on contest mathematics. David Stout of webAI said the immediate collaboration with Hugging Face illustrates the potential for open research to create adaptable, efficient AI without the need for ever-larger models.
Thom Wolf has released a comprehensive guide titled "The Ultimate MoE Training Handbook," which provides free and open instructions for developing Mixture-of-Experts (MoE) AI models. The resource, which is approximately 70,000 words long, includes 55 interactive figures and runnable code derived from 800,000 GPU hours of experimentation by Wolf and his team. The handbook aims to standardize training methods for an architecture that currently powers most of the top 20 open AI models, though Wolf noted that the specific technical expertise has remained concentrated within a small number of organizations. The guide addresses common bottlenecks in model training, including the efficiency of all-to-all communication and the selection of router functions to prevent processing delays.
Fireworks has appointed a new chief financial officer to oversee its operations as it builds a platform designed for enterprise artificial intelligence. David Wehner, who is resigning as chief strategy officer at Meta, takes the role to help the startup develop tools that enable companies to create specialized AI based on their own proprietary data. The hire follows an announcement from Wehner that he is leaving the social media company to join a founder-led growth firm. Fireworks' Lin Qiao said Wehner's experience scaling Meta during its key growth phases would be an asset as the company focuses on enhancing the economics of AI training and inference.