Anthropic’s IPO paperwork puts its September-quarter revenue at $16.96 billion, ahead of OpenAI’s reported figure of more than $10 billion. The comparison is imperfect because the companies account differently for sales through cloud partners, and their reported losses use different measures: Anthropic recorded a $567 million operating loss, while OpenAI reported an adjusted operating loss of $10.9 billion. Anthropic’s filing also attributes costs to $5 billion in short-term compute and a $2 billion non-cash donation-matching charge. The company is expected to price its IPO on November 18 and begin trading November 19, with a suggested range of $70 to $78 per share.
AI token prices have fallen even as infrastructure costs remain a concern. Yardeni Research says the usage-weighted Silicon Data LLM Token Expenditure Index dropped more than 40% from June 30 to early October, potentially making more AI uses affordable. Software improvements and more efficient models can reduce the cost of customer tasks, while GPU and memory costs have risen. Separately, Goldman Sachs forecasts 2.8% global growth in 2027, citing AI investment and lower energy prices as supports, but flags geopolitical tensions, fiscal pressures and uncertain technology spending.
OpenRouter’s usage rankings for the week of November 9 show DeepSeek V4.1 Flash 1017 at the top, with 55.4 trillion tokens—27.8% of the platform’s 199 trillion total. DeepSeek’s two Flash models together handled 70 trillion tokens, or 35%. Tencent’s Hy4 ranked second with 17.5 trillion tokens after its November 4 release. Six of the nine listed models came from Chinese labs; OpenAI had two in the top nine and Anthropic one. The figures track tokens processed, not spending or user counts.
Analysts and investors are also weighing the cost and financing risks of AI build-outs. One article estimates planned infrastructure could require more than $10 trillion in company cash, public and private debt, and other funding, warning that competition among seven or eight firms may leave only two or three winners. Rapidly aging data centers could need at least $3 trillion in annual revenue to generate adequate returns. The author calls for stronger oversight, clearer disclosure and a review of accounting practices. Hana Securities’ Park Seungjin expects high rates and heavy AI spending to pressure cash flows into early 2027, and recommends cash-generative ETFs including COWZ, CALF, TIGER US Cash Cow 100, QUAL and SPHQ; he names MAGS and MGK for large-tech exposure if cash-flow pressure eases.
RAND’s November 12, 2026 survey found that 57% of teens in households earning at least $150,000 used AI for homework in spring 2026, versus 43% in households earning under $50,000; both groups were at 39% a year earlier. Of 533 teens matched with parent responses, 81% of those who felt encouraged by a parent used AI for homework, compared with 38% who felt discouraged. RAND says the effect on achievement gaps remains unclear and cautions against unrestricted AI use on students’ first attempts at schoolwork.
Microsoft has lost machine-learning researcher John Langford after shutting down its 50-person AI Frontiers team, founded in 2023. The group worked on AutoGen, Fara and projects including Next Latent, Dion3 and Full Bandwidth Transformers. Langford said Microsoft AI prioritized larger models, computing power and training data over research into efficient algorithms, optimization and model designs; other senior researchers linked to the team have reportedly left. Uber, meanwhile, has built a framework to check AI agents’ actions against user requests, agent purpose, permissions and context. It adds behavioral authorization to traditional controls through an AI Registry for policies and limits and AI Guards that review interactions and proposed actions. The article also names Agent Guard runtime hooks but does not explain them in detail.
AI is also reshaping creative work: South African creator Coral Hart says she has published about 370 romance books since February 2025 using several language models, roughly four books a week. She says newer tools can draft a whole book in one sequence and has discussed differences between ChatGPT and Claude. Hart reports receiving tens of thousands of hostile messages but says she believes the technology will not be switched off because some authors oppose it.
A New York Times report raises questions about possible conflicts involving Commerce Secretary Howard Lutnick and AI-chip policy. It says Lutnick pushed to loosen restrictions on advanced chips while Cantor Fitzgerald, which he once led and his sons now run, helped arrange financing for Asian data centers. A Malaysian project expected to open next year is set to house hundreds of millions of dollars in Nvidia servers intended to provide computing for a Chinese technology giant. The report found no evidence that Cantor’s dealings influenced Lutnick’s decisions, and Commerce says he fully divested his financial interests. Other reports say he delayed efforts to restrict suspected chip diversions to China; the department has defended its enforcement. Earlier scrutiny of a Tether loan benefiting his children and Cantor’s offer to buy tariff refund claims also did not establish wrongdoing.
Key Takeaways
- Anthropic reported $16.96 billion in September-quarter revenue, versus OpenAI’s reported more than $10 billion; their revenue accounting and loss measures are not directly comparable.
- Anthropic is expected to price its IPO on November 18 and begin trading November 19, with a suggested $70–$78 per-share range.
- The Silicon Data LLM Token Expenditure Index fell more than 40% from June 30 to early October, while GPU and memory costs rose and software efficiencies lowered some task costs.
- Goldman Sachs forecasts 2.8% global growth in 2027, citing AI investment and lower energy prices while noting geopolitical, fiscal and technology-spending risks.
- DeepSeek V4.1 Flash 1017 led OpenRouter’s week-of-November 9 rankings with 55.4 trillion tokens; its two Flash models together accounted for 35% of platform token traffic.
- RAND found homework AI use rose to 57% among teens in households earning at least $150,000 and 43% among those under $50,000; parental encouragement was associated with higher reported use.
- An article estimates AI infrastructure plans could require more than $10 trillion in financing; a separate investment outlook advises prioritizing cash-generative ETFs amid high rates and AI spending.
- John Langford left Microsoft after its AI Frontiers team shut down; the team worked on AutoGen, Fara and other research projects.
- Uber’s agent-security framework combines an AI Registry and AI Guards to assess agent actions against policies, permissions and context.
- Coral Hart says she has published about 370 AI-assisted romance books since February 2025; a report on Howard Lutnick and chip policy found no evidence Cantor Fitzgerald’s business influenced his decisions.
Anthropic reports $17 billion quarter, ahead of OpenAI
Anthropic reported about $17 billion in revenue for the September quarter, compared with OpenAI’s more than $10 billion. Anthropic’s figure comes from IPO paperwork, while OpenAI’s was reported and later confirmed by the company to investors. The companies count some sales through cloud partners differently, which affects the comparison. Anthropic reported a $567 million operating loss, while OpenAI’s adjusted operating loss was $10.9 billion, though the figures are not directly comparable. Anthropic is expected to price its IPO on November 18 and begin trading on November 19, with a suggested price range of $70 to $78 per share.
Cheaper AI tokens may help providers attract more users
Yardeni Research said falling AI token prices may benefit AI providers because lower costs can make more uses worth paying for. The usage-weighted Silicon Data LLM Token Expenditure Index fell more than 40% between June 30 and early October. Anthropic’s third-quarter revenue reached $16.96 billion, up 47% from the second quarter, but the company swung from a $974 million operating profit to a $567 million loss. It cited $5 billion in costly short-term compute and a $2.0 billion non-cash donation-matching charge. Meanwhile, GPU and memory costs rose, while software improvements and more efficient models helped lower the cost of customer tasks.
RAND finds a growing income gap in teens’ AI homework use
A RAND survey published on November 12, 2026, found that 57% of teens in families earning at least $150,000 used AI for homework in spring 2026, compared with 43% in families earning under $50,000. A year earlier, both groups reported 39% use. Higher-income parents were also more likely to use AI themselves and encourage their children to use it for schoolwork. Among 533 teens matched with their parents’ responses, 81% of teens who felt encouraged by a parent used AI for homework, compared with 38% who felt discouraged. RAND said it is not yet clear whether the gap will widen or narrow achievement differences, and cautioned against unrestricted AI use on students’ first attempts at schoolwork.
Goldman Sachs sees global growth reaching 2.8% in 2027
Goldman Sachs expects the global economy to grow 2.8% in 2027. Its research report points to continued investment in artificial intelligence and lower energy prices as support for growth. The outlook also faces risks from geopolitical tensions, fiscal pressures and uncertainty about technology spending.
DeepSeek leads OpenRouter’s most-used AI models in November
OpenRouter’s usage data for the week of November 9, 2026, ranked DeepSeek V4.1 Flash 1017 first, with 55.4 trillion tokens, or 27.8% of the platform’s 199 trillion-token total. Tencent’s Hy4 placed second with 17.5 trillion tokens after its November 4 release. Six of the nine listed models came from Chinese labs, while OpenAI had two models in the top nine and Anthropic had one. The rankings measure tokens processed, not spending or the number of users. DeepSeek’s two Flash models together handled 70 trillion tokens, or 35% of platform traffic.
AI Infrastructure Spending Could Set Up a Future Financial Crisis
Plans to build AI infrastructure could require more than $10 trillion from company cash, public and private debt, and other financing. The article warns that the race may lead to overinvestment, since seven or eight companies are competing but only two or three may succeed. Rapidly aging data centers could require at least $3 trillion in yearly revenue to earn an adequate return. Long-term funding may make a fire sale less likely than in the 2008 financial crisis, but complex financing and hidden obligations create risks. The author calls for stronger financial oversight, clearer disclosures, and a review of accounting practices.
ETF Experts Favor Cash Generators as AI Costs and Rates Pressure Companies
Hana Securities researcher Park Seungjin expects high interest rates and heavy AI investment to strain company cash flows into early 2027. He said supply-driven inflation, including pressure from oil prices and tariffs, could keep rates high, while AI adoption may slow demand for workers. Park expects big tech companies to keep spending on AI infrastructure before those investments produce returns. He advised ETF investors to focus early in the year on companies with strong cash flow, recommending COWZ, CALF, TIGER US Cash Cow 100, QUAL, and SPHQ. For exposure to major technology companies, he named MAGS and MGK as options if cash-flow pressure eases.
John Langford Leaves Microsoft After AI Research Team Shuts Down
Machine learning researcher John Langford left Microsoft after more than 14 years, following the shutdown of its AI Frontiers team. The 50-member group, founded in 2023, worked on projects including Microsoft's AutoGen multi-agent framework and Fara, a computer-use agent. Langford said Microsoft AI favored larger models with more computing power and training data over research into more efficient algorithms, optimization methods, and model designs. AI Frontiers also explored projects called Next Latent, Dion3, and Full Bandwidth Transformers. Other senior researchers linked to the team have reportedly left, while ByteDance is testing similar ideas.
Uber Framework Checks AI Agents’ Actions Against Policies and Context
Uber developed a security framework to assess whether AI agents act appropriately based on a user's request, the agent's purpose, its permissions, and the situation. The company says traditional identity and API controls do not fully protect agents that reason over outside content and take real-world actions. Its framework adds behavioral authorization to the usual policy administration, decision, and enforcement steps. It uses an AI Registry to manage agent policies and limits, and AI Guards to check interactions and proposed actions against those policies. The article also names Agent Guard runtime hooks as a framework component, but the source text ends before explaining them in detail.
Coral Hart Says AI Helps Her Publish Hundreds of Romance Novels
South African content creator Coral Hart has published about 370 books since February 2025, using several large language models to generate prose. That output equals roughly one book every 1.75 days, or about four books a week. Hart says recent AI tools let her draft a whole book in one sequence, and she has described different strengths and weaknesses in tools such as ChatGPT and Claude. She says she has received tens of thousands of hostile messages about her process but sees herself as an ambassador for it. In comments to AFP, she said the technology will not be switched off because some authors object.
Report raises conflict questions over Lutnick and AI chip rules
A New York Times report says Commerce Secretary Howard Lutnick pushed to loosen restrictions on advanced AI chips while Cantor Fitzgerald, the firm he once led and his sons now run, helped arrange financing for Asian data centers. One Malaysian project, A201, is expected to open next year with hundreds of millions of dollars in Nvidia servers intended to provide computing power to a Chinese technology giant. The report found no evidence that Cantor’s business dealings influenced Lutnick’s decisions, and the Commerce Department says he fully divested his financial interests in the firm. Other reports say Lutnick delayed efforts to restrict suspected chip diversions to China, while the department has defended its enforcement record. Earlier scrutiny involved a loan from Tether to a trust benefiting Lutnick’s children and Cantor’s offer to buy tariff refund claims, but neither account established wrongdoing.
Sources
- Anthropic’s Quarterly Revenue Is Now $17 Billion, Far Ahead Of OpenAI’s $10 Billion
- Why falling AI token prices aren’t the negative for AI providers one might assume
- RAND survey: income gap in teens’ AI homework use opens up
- Global economy to grow 2.8% in 2027 as AI spending, lower oil prices support outlook: Goldman Sachs
- These Are The Top AI Models On OpenRouter By Number Of Tokens [November 2026]
- Column: The AI boom may be sowing seeds of another financial crisis
- 2027 High Interest Rates, AI Risks... "ETFs: Invest in Companies That Generate Steady Cash" [Jumal Money]
- John Langford Quits Microsoft After 14 Years, Says AI Research That Could Cut Costs Was Ignored
- Uber's Secure AI Framework Governs Autonomous AI Agents Across Their Lifecycle
- Author Says the Hundreds of Full-Length Romance Novels She Publishes Every Year Are Very High Quality in Spite of AI Usage
- NYT: Lutnick Eased AI Chip Restrictions While Firm Helped Finance Asian Data Centers
Comments
Please log in to post a comment.