AI Next WaveOne Day Into the Roadshow, the Price Talk Rose. The Small Buyer Decides Next Week.
Anthropic asked the public to help fund a $900 billion spending plan on Tuesday, and set aside up to a fifth of the sale for the households being asked.
The Short Version
Anthropic, the company behind the Claude chatbot, launched the roadshow for its first public share sale on Tuesday, November 10. Its amended filing with the Securities and Exchange Commission offers 825 million shares at an expected $70 to $78 each, under the ticker ANT on the Nasdaq Global Select Market, and the registration has yet to take effect, so no shares can be sold yet. Late Tuesday, the Financial Times reported that the bankers running the sale expect to price above that range after heavy first day demand, which could lift the company's value to $1.5 trillion or more on a fully diluted count, the count that includes shares promised to employees and other holders. CNBC reported that pricing is expected on November 18 and trading on November 19, citing people familiar with the plans, and Anthropic has confirmed neither date. The same company launched spreadsheets, whiteboards and a data tool inside Claude on Tuesday, with a meeting assistant listed as coming soon, in a direct push into the office work Microsoft 365 holds today. Meta said on Monday it will expand AI age checks on Instagram worldwide, moving millions more teenagers into restricted accounts even when the birthday on the account says adult. And Notion spent the week telling customers that a vendor behind its AI assistant had exposed chat transcripts from workplaces that never chose that vendor.
The filing asks a household to believe a $900 billion spending plan will pay off
Anthropic set an expected price of $70 to $78 on 825 million shares, set aside up to 20 percent for retail buyers, and carries about $900.9 billion in signed spending commitments
Start with the buyer this sale is built to include. Most large share sales reserve a small slice for individual investors, often 5 to 10 percent of the deal. Anthropic's amended filing expects up to 20 percent of its Class A shares to go to retail investors in the United States and abroad, according to the Motley Fool's reading of the filing, with the final share depending on demand. Twenty percent is a ceiling the filing sets, and it means the company wants households in this sale at a scale most first time listings skip. A household that takes part is buying a company whose own filing describes a business growing at a speed few public companies have matched, and spending at a speed that matches it.
The growth figures sit in the filing itself. Revenue for the first nine months of 2026 came to $33.2 billion, up from $2.5 billion in the same nine months of 2025. The second quarter of 2026 produced an operating profit, $974 million of operating income by the filing summaries, the first quarter in the series to show one, and the third quarter went back to an operating loss. Operating profit here means the core business earned more than it spent in that quarter before interest, taxes and accounting charges, and the nine month bottom line stayed deep in loss, so a buyer is paying for growth and a spending plan, with sustained profit still ahead of the company rather than behind it. The risk section of the filing runs from page 29 to page 122, about 94 pages of named ways this can go wrong, and a reader with a week to decide can start there, because that section is the company describing its own weak points in its own words.
The spending plan is the largest number in the document. As of September 30 and in agreements signed after that date, Anthropic has entered purchase commitments and lease agreements with an aggregate committed spend of about $900.9 billion, substantially all of it related to computing and infrastructure, by the filing's own wording. That figure updates the $895.4 billion the company's November 2 amendment reported, and it is the bill the share sale exists to help pay. Alongside the public sale, seven companies that sell Anthropic its chips, cloud capacity and components have agreed to buy $26 billion of a separate class of stock at the same price the public pays, in a private placement that closes right after the public sale. Google and Nvidia account for about $10 billion each at the midpoint price, AMD about $2.5 billion, and Broadcom, Micron and Microsoft about $1 billion each, with TSMC Partners at about $500 million. The suppliers funding the customer is a fact a small buyer should sit with. It can read as confidence from the firms closest to the work. It also means a large share of the new money arrives from companies whose own sales rise when Anthropic spends.
By Tuesday night the price itself was moving. The Financial Times reported late Tuesday, citing five people familiar with the matter, that the bankers expect to raise the price above the $70 to $78 range after first day orders covered the whole offering, lifting the value target to $1.5 trillion or more on the fully diluted count, and pushing the total raised past $100 billion when the placement and the banks' extra share option are counted. Anthropic has filed no new range, and any change has to appear in a new filing before it counts. CNBC reported the order book was already covered at the filed range on Tuesday. For the household buyer, the practical shape of the week looks like this. The price you read on Tuesday may sit below the price you are offered next week, the dates in circulation, pricing on November 18 and trading on November 19, come from CNBC's sources and carry no confirmation from the company, and the Australian retail offer that opened Tuesday evening New York time closes on November 17, which shows how fast the retail window can shut. The decision window is short, the document is long, and the honest starting point is the filing's own risk pages and its spending commitments, read before the price talk.
The spreadsheet moved inside the chatbot, and the meeting assistant is on its way
Claude Sheets and Whiteboards launched in beta on paid plans, Claude Data became generally available, and Claude Meetings is listed as coming soon
On the same morning it set that expected price, Anthropic moved Claude into the middle of office work. The company's announcement, titled More ways to work with Claude, introduces Claude Sheets and Claude Whiteboards as new places where a team works with the assistant in the room, makes Claude Data generally available, and upgrades interactive connectors so partner apps such as Adobe and Atlassian open in full inside a Claude conversation, with Claude making changes in them directly. A separate product page states the availability plainly: Sheets and Whiteboards launched in beta, which means paying customers can use them while the company still labels them works in progress, and Claude Data, the renamed Dashboards tool from an October beta, is now generally available. All three sit on paid plans, on web, desktop and mobile. On Team plans all three switch on by default. On Enterprise plans Sheets and Whiteboards start switched off, and Claude Data switches on by default on November 19, 2026, a date the administrator at a large employer controls and can change in the organization settings.
Claude Meetings carries a different status, and the difference matters for anyone planning around it. Anthropic's own announcement lists Meetings as coming soon, in the company's words it will let you add Claude to a call to take notes and update the documents your team is working on during the conversation. TechCrunch reports a beta on Team and Enterprise plans with Google Meet first, a visible participant that drafts the plan being discussed and answers questions in the meeting chat, and a rule that it leaves when an external guest joins unless an admin has authorized its presence. Readers should hold both labels at once: coming soon in the company's own post, a limited beta in the trade reporting. What has yet to be published in either source is a general release date.
The person at the center of this launch works in an office, on a team, in meetings that produce notes somebody has to write up. The pitch Anthropic is making runs against the suite that person already uses. Microsoft has spent two years putting its AI inside Word, Excel and Teams, the apps offices already pay for. Anthropic is pulling the spreadsheet, the whiteboard and the dashboard inside the chatbot instead, and asking the team to come to Claude. Microsoft shares fell about 3 percent in Tuesday trading following the announcement. Microsoft also sits on the other side of this same week. It sells Claude inside its own products, and it has committed about $1 billion to Anthropic's private placement at the public sale price, per the filing figures above. Partner, supplier and rival describe the same pair of companies in the same seven days, and the office worker in the middle will feel the result as a choice of where the meeting notes live, who can search them, and which employer controls, the admin defaults above, decide whether the new tools arrive switched on.
Millions more teenagers will wake up in restricted Instagram accounts, including teenagers who lied about their age
Meta expanded its AI age checks worldwide on Monday, and the system reads profiles for clues such as school grades and birthday posts
A teenager who opened Instagram with an adult birthday on the account has been able to browse as an adult, with adult contact settings and adult content rules. On Monday, Meta said it is expanding its AI age estimation technology globally on Instagram, and the system will place people it suspects are teenagers into Teen Account protections even when the birthday on file says adult. Meta says millions more teenagers worldwide will land in those protections, and the company gives no exact count, no accuracy rate and no per country start date in its announcement. Teen Accounts limit who can contact the teenager, limit the content the account can see, and limit time on the app, on Instagram, Facebook and Messenger, by Meta's and Reuters' shared description. The expansion covers Instagram now. The same technology is still rolling out on Facebook, and Meta says a global launch on Threads comes later this year.
How the system decides is the part a parent needs in plain terms. In Meta's May 2026 account of the technology, which the November expansion builds on, the AI reads an entire profile for context clues, such as birthday celebration posts and mentions of school grades, across posts, comments, bios and captions, and the company has added visual analysis of photos and videos. Meta corrected that May account on July 31 to state plainly that the system uses no biometric data, creates no facial fingerprint and identifies nobody in a photo, and that it looks for general visual patterns suggesting a person in an image may be underage. TechCrunch described the visual side as including height and bone structure, wording Meta itself avoids, and families should treat the company's corrected description as the governing one until Meta publishes test results. In Australia, Meta has begun using Apple's age range signal, a range the phone passes to an app each time a teenager downloads one, as one more input.
The legal ground under this move is visible in both sources. Meta's post ties the expansion to its recent agreement with 52 bipartisan attorneys general, and Reuters reports the August settlements with nearly every state and territory were worth up to $18 billion, with an agreement to improve how Meta identifies teenagers and children under 13. That $18 billion figure is Reuters' reporting, carried in its November 9 story, and Meta's own post gives the agreement without the dollar figure. For a household, the live effects arrive without any announcement at the kitchen table. A teenager whose account gets moved will see contact and content limits change, and a young adult whose profile reads young, a college student with school grade posts still on the account, could get moved too, with an appeal route Meta's November post leaves undescribed. The setting worth checking this week sits in the account itself: the birthday on file, the account type the app reports, and whether the protections Meta describes match the account your teenager actually holds.
Workers pasted their work into an AI chat at their desks. A vendor they had never heard of held the transcripts.
Notion says under 1 percent of its workspaces were affected in the Fireworks incident, the Insights feature is disabled, and anyone who pasted a password or a key into a chat should rotate it
Workers used the AI assistant inside Notion, a workplace notes app their employers chose, and asked it questions in the course of a normal week. In October, an AI infrastructure company called Fireworks AI, which ran some of the models behind that assistant, was breached through a third party cloud provider that hosted its research machines. On November 9 and 10, Notion published its account of what that breach carried away from its own customers. A dataset the attackers reached included Notion AI chat transcripts used for the Insights feature, a feature that summarized workspace activity. Notion says the incident affects under 1 percent of its workspaces, a share with no customer count attached in its advisory, and that for the vast majority of affected workspaces the transcripts were unattributable, meaning the chats carried no label naming who wrote them or which workspace they came from, and held no tool call results. The company adds the sentence that decides how a worker should read that comfort: beyond the transcript itself there is no information tying a chat back to a workspace, a user or a company, and the transcript itself is a record of whatever a person typed, which can name a project, a client, a salary band or a health detail in the person's own words.
Notion's timeline runs across a week of changing knowledge. Fireworks confirmed the final dataset on November 8. Notion says it first heard on November 2 that the exposed data was synthetic, test data with no customer content, disconnected Fireworks on November 3, learned on November 6 that customer data might be involved after all, and reached affected customers on November 9. The Insights feature is now disabled, and Notion says it no longer serves open weight models through Fireworks. Its advice to affected workspaces is direct and worth repeating exactly: treat the chat data as exposed, and if anyone pasted sensitive information, passwords, API keys or other credentials into Notion AI during the incident window, rotate them, which means replacing each password and key with a fresh one so the old value stops working. If you have heard nothing from Notion, the company says it has identified no effect on your workspace and no action is needed right now. Render, a cloud company whose founder shared his notification publicly, stands as the named example of an affected workspace in The Register's November 10 report.
Fireworks' own bulletin describes a wider incident at the vendor itself, and the two scopes belong in separate columns. Fireworks says the attackers also reached account and billing information for its own customers, and for a limited set of accounts, secrets, datasets and a subset of incomplete model weights, while payment card numbers were never stored by Fireworks and were unaffected. The company says it has detected no effect on its inference and training services, which run with zero data retention by default, and its investigation continues with law enforcement notified and an outside forensics firm engaged. Fireworks describes the attackers as a financially motivated group selling stolen access to AI models through illicit gateways, a practice the bulletin names LLMJacking. None of Fireworks' wider categories should be read onto Notion's customers: Notion's confirmed exposure is the Insights transcripts, and its advisory claims nothing broader. The household lesson travels past both companies. Any AI chat inside a work tool can sit on a chain of vendors the worker never picked, and the safe habit is the one Notion's advice assumes: keep passwords, keys and personal records out of chat boxes entirely, because a transcript can outlive the conversation and travel further than the person who typed it ever intended.
Opportunity Radar
The plain English prospectus companion. Anthropic's filing runs hundreds of pages with about 94 pages of risk factors, and the retail slice of this sale is up to 20 percent, which means a large group of first time buyers will face a document written for lawyers and fund managers. A small service could sell a plain language companion for a named public offering: the risk pages retold in everyday words, the spending commitments laid out as a dated list, the supplier placement explained in one diagram, and a checklist of what changes when a new filing lands. The buyer is a financial educator, a community investing group or a brokerage content team that wants its retail customers reading past the price headline. Validation means publishing the companion for this offering and counting how many readers finish the risk section summary, measured by the educator's own page data. Success is a second offering's companion commissioned before its roadshow ends. The idea fails if readers only want a buy or skip verdict, which the first edition's completion data will show quickly, because a companion that sells verdicts becomes advice and leaves the education ground it stands on.
The meeting tool default audit. Anthropic's launch puts three AI tools on by default for Team plans, holds two of them off by default for Enterprise plans, and switches Claude Data on by default for Enterprise plans on November 19, 2026, unless an administrator changes the setting. That spread of defaults is a small, sellable job for an IT consultant serving small and mid size firms: inventory which AI features sit on in the tools a firm already pays for, record who can search meeting notes and chat transcripts, set retention and credential rules in writing, and date the next review. The buyer is a managing partner or operations lead at a firm with no full time IT staff. Validation means completing the audit at two firms and finding at least one default or permission nobody at the firm had chosen on purpose, a result the Notion incident, where a transcript sat with a vendor the workers never picked, suggests is common. Success is a quarterly repeat audit on the calendar. The idea fails if the admin consoles already hand a small firm this inventory in one screen, which the first audit will reveal within an hour.
What You Can Do With This
If you are weighing the Anthropic sale
Read the filing's risk pages and its commitments figure before the price talk, and treat November 18 pricing and November 19 trading as reported dates awaiting confirmation, with the filed range of $70 to $78 as the only price the company itself has published. A price above the range would arrive by a new filing, and a household buying at a first day price rise is paying for other buyers' enthusiasm, a cost the filing cannot protect you from. Size any purchase so a bad first year is survivable, because the company's own nine month record shows growth with a bottom line still in loss.
If your teenager uses Instagram
Open the account together and check the birthday on file and the account type the app reports. Meta's worldwide expansion means a suspected teenager lands in Teen Account protections even with an adult birthday entered, and Meta's November post describes no appeal route and publishes no accuracy rate, so a wrong move is a family discovery before it is a company announcement. Parents in the United States will also see an in feed video this month explaining Meta's approach, by the company's own account.
If you use an AI chat at work
Keep passwords, access keys and personal records out of the chat box, starting today. Notion's advice to affected customers was to rotate any credential pasted into its AI during the incident window, which means issuing a fresh password or key so the old one dies. A transcript can sit with vendors your employer hired and you never chose, and the Fireworks incident shows a transcript traveling two companies away from the person who typed it.
The Bigger Picture
One company spent Tuesday at two counters at once. At the investor counter, Anthropic asked the public to help pay for about $900.9 billion in signed computing commitments, with its suppliers buying $26 billion of the stock at the public price. At the office counter, it moved the spreadsheet and the whiteboard inside its chatbot and aimed at the suite its own $1 billion placement partner sells. Each counter runs on a promise a reader can check against a record: dollars of revenue growth in the filing, hours of office work in the launch. That checking habit is the through line of today's edition. Meta's age system promises protection and publishes no accuracy rate. Notion's comfort word, unattributable, holds only until a transcript's own words name their author.
The cost side of the build out keeps arriving on its own schedule. The Energy Information Administration said on Tuesday that United States electricity consumption will rise to record highs in 2026 and again in 2027, driven largely by AI data centers and electrification, with residential sales rising alongside commercial and industrial demand, Reuters reported. That is the same demand Anthropic's commitments are built to serve, and the same demand that arrives in household power bills. A household meets this industry as an investor, a worker, a subscriber, a parent and a ratepayer, often in the same week, and each meeting deserves the same plain question, stated with its unit and its date attached: what exactly was promised, what does the record show, and who carries the difference.
References
Anthropic, Anthropic launches roadshow for proposed initial public offering, November 10, 2026 Anthropic, PBC, Amendment No. 2 to Form S-1, filed November 10, 2026, SEC EDGAR MLQ News, Anthropic's bankers weigh pricing its IPO above the $70 to $78 range, lifting its valuation to $1.5 trillion or more, FT reports, November 11, 2026 Motley Fool, Anthropic is reserving up to 20 percent of its IPO for retail investors, November 10, 2026 Anthropic, More ways to work with Claude: Sheets, Whiteboards, and Interactive Connectors, November 10, 2026 Anthropic, Introducing Claude Whiteboards, Claude Sheets, and Claude Data, November 10, 2026 TechCrunch, Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks, November 10, 2026 PYMNTS, Anthropic unveils office suite to challenge Microsoft ahead of landmark IPO, November 10, 2026 Meta, Expanding Teen Account protections to teens around the world, November 9, 2026 Reuters, Meta expands AI age checks for Instagram teens as lawsuits mount, November 9, 2026 TechCrunch, Meta expands Instagram's AI age-guessing tech globally, November 9, 2026 Meta, AI age assurance for teens, May 5, 2026, corrected July 31, 2026 Notion, Security incident involving a third party AI vendor, updated November 10, 2026 Fireworks AI, October 2026 security incident update, updated November 9, 2026 The Register, Fireworks AI gets burned twice as customer data exposure widens, November 10, 2026 Reuters, US power demand to beat record highs in 2026 and 2027 as AI use surges, EIA says, November 10, 2026