Anthropic Raises $81.34 Billion in Record IPO

Anthropic’s IPO draws attention to both the scale of demand for AI and the costs behind it. The company raised $81.34 billion, the largest IPO on record, pricing shares at $83 and opening at $111.50—a 34% increase. Its third-quarter sales reached $17 billion, more than 12 times the year-earlier level; a separate report says revenue for the nine months through September was $33 billion, up about 1,200%, largely on demand for Claude Code. Its prospectus lists $518 billion in future cloud, computing and infrastructure bills. Amazon owns 14.7% of Anthropic and is due to receive about $110 billion in computing payments through April 2036. The debut is being watched as a test of investor confidence in AI amid concerns about infrastructure spending and returns.

The spending boom is also reshaping markets and investment plans. Alphabet, Amazon, Meta, Microsoft and Oracle issued about $160 billion in US-dollar debt in the first three quarters of 2026. A study of 14 bond issues from January 2023 to October 2026 found 10-year Treasury yields rose an average of 3.8 basis points more on announcement days than on other days, with the research pointing mainly to increased long-term debt supply and investment needs. Fitch says US investment-grade corporate bond issuance by non-financial companies reached $940 billion in the first 10 months of 2026, up 33% year over year, largely due to AI borrowing. It warns that higher Treasury yields or weaker hyperscaler credit profiles and share prices could widen spreads and make refinancing harder for companies in health care, technology and utilities.

AI infrastructure suppliers offer another way to track the buildout, but their results and risks vary. Vertiv reported third-quarter sales of $4.31 billion, up 61%, while Bloom Energy’s revenue climbed 138% to a record $1.24 billion. Both supply data-center infrastructure. IREN, which is shifting from crypto mining toward AI infrastructure and reportedly supplies Anthropic with extra GPU capacity, saw its shares fall nearly 2% despite Anthropic’s debut. IREN’s AI Cloud revenue nearly tripled to $200 million in the September quarter, but it pushed its target for more than $4 billion in annualized revenue to January 2027, with a possible slip into February. Its shares are down about 67% over 12 months; it reports a $20.3 billion revenue backlog and expects $15 billion in remaining fiscal 2027 capital spending, with $6 billion still to raise.

Analysts also point to possible beneficiaries among chipmakers. Lynx Equity’s KC Rajkumar names Micron as his top pick, with a $1,325 price target, and says it trades at about 2.3 times his estimated calendar 2028 earnings. He sets targets of $2,450 for SanDisk and $400 for Lam Research, and is more positive on Advanced Micro Devices and Intel. He argues that data-center permitting concerns have weighed on semiconductor stocks without changing long-term demand for chips and equipment. Rajkumar says a successful Anthropic IPO could renew interest in AI stocks; seven technology companies reportedly invested $26 billion in a private placement tied to the IPO price.

Companies are also adding AI tools to specific workflows. Snowflake now offers TwelveLabs video search through SQL: Marengo turns visuals, audio and speech into 512-dimensional vectors, while Pegasus generates summaries, scene descriptions and structured metadata. Users can create embeddings with SELECT AI_MULTI_EMBED and request text outputs with SELECT AI_COMPLETE; video stays on Snowflake stages, and access controls and audit logs cover derived data stored in tables. Code Effects Version 7 lets AI agents work with business rules, versions, tests and evaluation history through an MCP server, but agents cannot approve, delete or publish rules. Human approval remains necessary before deployment, and approved rules execute in the host application’s .NET environment without an LLM.

Other releases focus on coordinating agents and keeping their use under control. Researchers’ Learning to Coordinate framework adjusts how AI agents collaborate during automated safety testing. Across six configurations, it improved attack success rates by 1.4 to 4.5 percentage points over EvoSynth while cutting average token use by 35.6% to 37.3%. Reply.io’s open-source Jason Harness, released as an alpha on November 17, assigns agents tasks such as prospect research, follow-up planning and message writing. It works with agents including Claude Code and OpenAI Codex, stores campaign information locally across sessions and requires human approval before messages are sent. It has nine roles, includes no AI model of its own, and its launch materials provide no independently verified performance results.

Intel and Microsoft describe a security partnership for business PCs as agents begin interacting with apps, files and data. Windows 11 Pro and Intel vPro combine protections including vPro’s hardware root of trust and Windows safeguards for startup, credentials and memory. BitLocker encrypts drives, while Intel Total Storage Encryption shifts encryption work to a dedicated hardware engine. Microsoft Pluton runs its firmware in an isolated part of the chip through Intel Partner Security Engine; the companies say security must protect agents while they run as well as sensitive data.

AI is also entering education and news distribution. Bodhan AI, an education-focused centre of excellence incubated at IIT Madras, plans to train more than one million Indian school teachers over the next 12 to 18 months, covering AI literacy, responsible use, classroom applications and risks including inaccurate outputs, bias and privacy. Learning Links Foundation and Vidya Shakti will support the rollout, with the Centre for Teacher Accreditation leading implementation in Andhra Pradesh. Vidya Shakti has already trained around 4,000 teachers from 1,500 schools across six states; participants reported saving at least 30 minutes a day on repetitive work. Separately, Le Groupe Figaro and OpenAI will make current and archived journalism available for discovery in ChatGPT, which will provide summaries and excerpts linked to original stories. Le Figaro retains editorial control and independence, plans to introduce ChatGPT Enterprise across several teams, and will explore further uses under its AI charter and editorial standards.

Key Takeaways

  • Anthropic raised $81.34 billion in a record IPO; shares opened at $111.50, 34% above the $83 offer price.
  • Anthropic reported $17 billion in third-quarter sales and lists $518 billion in future cloud, computing and infrastructure bills; Amazon holds 14.7% and is due about $110 billion in computing payments through April 2036.
  • Alphabet, Amazon, Meta, Microsoft and Oracle issued about $160 billion in US-dollar debt in the first three quarters of 2026; research links bond announcement days to a larger average rise in 10-year Treasury yields.
  • Fitch reports $940 billion in US investment-grade non-financial corporate issuance in the first 10 months of 2026, up 33% year over year, and warns AI-related borrowing could increase bond-market volatility and refinancing costs.
  • Vertiv’s third-quarter sales rose 61% to $4.31 billion and Bloom Energy revenue rose 138% to $1.24 billion; IREN’s AI Cloud revenue nearly tripled to $200 million, but it delayed its annualized revenue target and still needs to raise $6 billion for planned spending.
  • Analyst KC Rajkumar identifies Micron as his top semiconductor pick with a $1,325 target; he also sets targets of $2,450 for SanDisk and $400 for Lam Research, and is more positive on AMD and Intel.
  • Snowflake added TwelveLabs video search through SQL, while Code Effects Version 7 lets AI agents manage rule-related tasks but reserves approvals and production publishing for humans.
  • Learning to Coordinate improved automated red-team attack success rates by 1.4–4.5 percentage points over EvoSynth while reducing token use by 35.6%–37.3% across six configurations.
  • Reply.io’s open-source Jason Harness supports agents including Claude Code and OpenAI Codex, stores campaign context locally, and requires human approval before sending messages; it has no built-in model.
  • Bodhan AI plans to train over one million Indian teachers in 12–18 months; Le Figaro and OpenAI will bring linked summaries and excerpts of current and archived journalism to ChatGPT while Le Figaro retains editorial control.

Anthropic’s record IPO puts AI investor confidence to the test

Anthropic priced its IPO at $83 a share, above its earlier range of $70 to $78, before its Nasdaq debut. Gray market trading suggested the shares could open 20% to 26% higher, valuing the company at about $1.5 trillion. Investors will watch the debut as a test of confidence in AI companies, which face concerns about high interest rates, infrastructure costs, and when large investments will pay off. Anthropic reported $33 billion in revenue for the nine months through September, up about 1,200% from a year earlier, largely due to demand for Claude Code. A strong or weak debut could affect sentiment across the wider AI market.

Anthropic’s record IPO highlights AI demand and its risks

Anthropic raised $81.34 billion in the largest IPO in history, pricing shares at $83 and opening at $111.50, a 34% rise. Its third-quarter sales reached $17 billion, more than 12 times the level a year earlier, while its prospectus lists $518 billion in future cloud, computing, and infrastructure bills. Amazon holds a 14.7% stake in Anthropic and is due to receive about $110 billion in computing payments through April 2036. The article also highlights Vertiv, which sells data center power and cooling equipment, and Bloom Energy, which makes onsite fuel cells; both supply parts of the AI infrastructure buildout. It warns that record IPOs can signal peak enthusiasm, while AI stocks are already under pressure.

Anthropic’s IPO spotlights spending opportunities and risks for AI stocks

Anthropic raised $81.34 billion in the largest IPO in history, with shares priced at $83 and opening at $111.50, up 34%. Sales reached $17 billion in the third quarter, more than 12 times the year-earlier level, and the company lists $518 billion in future cloud, computing, and infrastructure bills. Amazon has a 14.7% stake in Anthropic and is set to receive about $110 billion in computing payments through April 2036. Vertiv reported third-quarter sales of $4.31 billion, up 61%, while Bloom Energy’s revenue rose 138% to a record $1.24 billion; both companies provide infrastructure used by data centers. The article says investors should weigh these growth opportunities against the risk that a record IPO may signal fading enthusiasm, especially as AI stocks have already fallen sharply.

IREN shares fall despite Anthropic’s strong IPO debut

IREN shares fell nearly 2% on Thursday as Anthropic raised $81.3 billion in the largest IPO on record and opened 34% above its $83 offer price. Anthropic reportedly uses IREN for extra GPU capacity, while IREN shifts its business from crypto mining toward AI infrastructure. IREN’s AI Cloud revenue nearly tripled to $200 million in the September quarter, but the company delayed its target for more than $4 billion in annualized revenue to January 2027 and said it could slip into February. The stock is down about 67% over 12 months and trades well below its major moving averages. IREN reported a $20.3 billion revenue backlog and expects $15 billion in remaining fiscal 2027 capital spending, with $6 billion still to raise after other funding sources.

AI companies’ bond borrowing may be lifting Treasury yields

Alphabet, Amazon, Meta, Microsoft, and Oracle issued about $160 billion in US dollar debt during the first three quarters of 2026, helping fuel what is known as the AI bond boom. Researchers examined 14 bond issues from these companies between January 2023 and October 2026. On announcement days, the 10-year Treasury yield rose about 3.8 basis points more than on other days, on average. The findings point mainly to increased long-term debt supply and investment needs, while showing little evidence that inflation or expected interest rate changes drove the rise. Treasury yields influence borrowing costs beyond the United States, including across ASEAN+3.

AI borrowing could increase bond market volatility

Fitch Ratings says growing AI-related borrowing could make bond markets more volatile as they become more tied to hyperscalers. U.S. investment-grade corporate bond issuance by non-financial companies reached US$940 billion in the first 10 months of 2026, up 33% from the same period a year earlier, largely because of AI-related debt. Fitch expects AI-linked borrowing to continue next year and does not expect uncertainty over U.S. data centre rules to cause a major cut in 2027 spending. Higher Treasury yields or weaker hyperscaler credit profiles and stock values could widen spreads, raise borrowing costs, and make it harder for companies in health care, technology, and utilities to refinance upcoming debt.

Code Effects Version 7 adds secure AI agents and rule auditing

Code Effects Version 7 lets organizations use AI in business rule management while keeping rule execution deterministic and production operations under human control. Its Model Context Protocol server lets AI agents work with rules, versions, tests, and evaluation history, but agents cannot approve or delete rules or publish them to production. AI-generated changes need human approval before deployment. The release also adds natural-language rule tools, a web-based Rule Repository with versioning and testing, and an Audit component that records production evaluations for review and replay. Organizations can connect commercial or privately hosted language models, while approved rules continue to run in the host application's .NET environment without an LLM.

Bodhan AI plans to train more than one million teachers

Bodhan AI, an education-focused Artificial Intelligence centre of excellence incubated at IIT Madras, plans to train over one million school teachers in India over the next 12 to 18 months. The programme aims to help teachers use AI for teaching, assessment, and student engagement, while reducing repetitive work. Learning Links Foundation and Vidya Shakti will support the national rollout, and the Centre for Teacher Accreditation will lead implementation in Andhra Pradesh. The training will cover AI literacy, responsible use, classroom applications, and risks such as inaccurate results, bias, and privacy concerns. Vidya Shakti has already trained around 4,000 teachers from 1,500 schools in six states, and trained teachers reported saving at least 30 minutes a day on repetitive tasks.

Le Figaro journalism will appear in ChatGPT through OpenAI partnership

Le Groupe Figaro and OpenAI announced a partnership to help people discover Le Figaro journalism in ChatGPT. The agreement covers current and archived stories from Le Figaro and associated publications on news, business, culture, and lifestyle. ChatGPT will provide summaries and excerpts with links to the original stories. Le Figaro will keep full editorial independence and control, and plans to introduce ChatGPT Enterprise across several teams. The company will also explore other ways to use OpenAI technology in its work and reader services, guided by its AI charter and editorial standards.

Snowflake adds TwelveLabs video search through SQL

Snowflake has added TwelveLabs video search tools that let users find and analyze video through SQL. Marengo converts visuals, audio, and speech into 512-dimensional vectors, while Pegasus can create summaries, scene descriptions, and structured metadata from clips. Users can create embeddings with SELECT AI_MULTI_EMBED and request text outputs with SELECT AI_COMPLETE. Cortex Agents can combine video, audio, and transcription signals with custom ranking. Video files remain on Snowflake stages, while derived data can be queried alongside campaign, catalog, and spending data, with row-level policies, role-based access, and audit logs covering video-derived data stored in Snowflake tables.

L2C Cuts Token Use in Automated AI Red Teaming

Researchers introduced Learning to Coordinate (L2C), a framework that changes how AI agents work together during tests of large language model safety. It records agent roles, task links and information exchanged, then uses a trained controller to decide whether to keep or change the workflow based on test results and the remaining budget. The evaluation covered HarmBench, three execution backbones and two target models. Across six configurations, L2C raised attack success rates by 1.4 to 4.5 percentage points over EvoSynth while reducing average token use by 35.6% to 37.3%.

Analyst Sees Micron as a Value Play as Anthropic IPO May Lift AI Stocks

Lynx Equity analyst KC Rajkumar says worries about slow data center permitting have hurt semiconductor stocks without changing the long-term demand outlook for chips and equipment. He sees Micron Technology as his top pick and maintains a $1,325 price target, noting that it trades at about 2.3 times his estimated calendar 2028 earnings. Rajkumar also has price targets of $2,450 for SanDisk and $400 for Lam Research, and is more positive on Advanced Micro Devices and Intel. He says a successful Anthropic IPO could renew interest in AI stocks; seven technology companies reportedly invested $26 billion in a private placement tied to the IPO price.

Intel and Microsoft Build Security for AI PCs and Agents

In a November 19, 2026 article, Intel and Microsoft described how their hardware and software partnership aims to protect business PCs as AI agents begin interacting with apps, files and data. Windows 11 Pro and Intel vPro combine protections, including vPro’s hardware root of trust and Windows safeguards for startup, credentials and memory. Microsoft BitLocker encrypts drives, while Intel Total Storage Encryption moves encryption work to a dedicated hardware engine. Microsoft Pluton also works with Intel Partner Security Engine, which runs Pluton firmware in an isolated part of the chip. The article says agent security also requires protection while agents run, alongside safeguards for sensitive data.

Reply.io Releases Open Source AI Tool for Sales Outreach

Reply.io announced Jason Harness, an open-source alpha tool made public on November 17 that assigns AI agents tasks such as prospect research, follow-up planning and message writing. It works with existing agents such as Claude Code and OpenAI Codex, and stores campaign details, contacts and history locally so work can continue across sessions. The tool has nine roles and requires a person to approve messages before they are sent through Reply or another connected service. It does not include its own AI model, and local storage does not mean all processing happens offline. Reply says the code is available under the MIT license, but the launch materials provide no independently verified results for deliverability, replies or meetings.

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.

Anthropic IPO AI infrastructure spending AI investment AI debt financing cloud computing data centers AI market demand semiconductors Micron Vertiv Bloom Energy IREN AI agents agent coordination AI safety testing Snowflake TwelveLabs enterprise AI security AI in education AI journalism partnerships

Comments

Loading...