Skip to content
Back to The Briefing

Anthropic prices record IPO and Google launches Gemini 4 Neon

The Briefing··7 min read

This morning’s briefing leads with Anthropic’s record flotation and Google’s new Gemini model. There’s also live dictation in ChatGPT and new reports on how the AI build-out is financed and powered.

Anthropic prices record US$81bn share sale

What happened. Anthropic priced its flotation at US$83 a share on 18 November and increased the offering from 825 million to 980 million shares. The sale totals US$81.34bn before fees, including shares sold by existing shareholders. Trading is scheduled to start on Nasdaq on 19 November under the symbol ANT. Anthropic had set an initial range of US$70–78.

Associated Press puts Anthropic’s valuation at about US$1.5tn at the offer price. The offering overtakes the US$75bn SpaceX raised at its initial pricing in June.

Why it matters. Anthropic says the offering will increase its capital and financial flexibility. Its filings set out large commitments for the computing capacity it needs to train and run Claude. The Nasdaq listing gives investors a daily price against which to judge that spending. Associated Press describes the debut as a test of demand for AI stocks.

The filing also gives organisations using Claude a view of how Anthropic makes release decisions. The company says progress it expects from its models has triggered further assessment under its Responsible Scaling Policy. It says safety measures can include delaying a model release or limiting who gets it first.

US$bn

Anthropic edges past SpaceX at initial pricing

The five largest IPOs by value of shares offered

Anthropic (2026)
81.34
SpaceX (2026)
75
Saudi Aramco (2019)
25.6
Alibaba (2014)
21.8
SoftBank Corp (2018)
21.3
Gross IPO proceeds at initial pricing before options to buy additional shares, 2014 to 2026. Sources: Anthropic’s final offer price and share count · Dealroom’s historical IPO proceeds ranking · Associated Press on the size of the offering

People are asking

**What’s the difference between US$81bn and US$1.5tn?

** US$81.34bn is the value of the shares on offer before fees. About US$1.5tn is the implied value of the whole company at the same share price. Associated Press explains the pricing.

**When do the shares start trading?

** Anthropic has scheduled its Nasdaq debut for 19 November under the symbol ANT.

Google brings Gemini 4 Neon to Pro and Ultra

What happened. Google launched Gemini 4 Neon on 18 November for work including document analysis, coding and research. It is rolling out to Google AI Pro and Ultra subscribers in the Gemini app and Antigravity, and to Gemini for business through its model picker and API. Google made it generally available on the paid Gemini API.

Artificial Analysis scores Neon at 60 on High, level with Gemini 4 Carbon after rounding. Neon cost US$3.74 per benchmark task at High, against US$7.76 for Carbon at standard rates. Neon at Extra High scored 61 and cost US$6.44 per task.

Why it matters. The Artificial Analysis results give teams a starting point for deciding how much reasoning to pay for. Moving Neon from High to Extra High raised its rounded index score from 60 to 61 and its cost per benchmark task from US$3.74 to US$6.44. Teams can test both settings on a recurring job and see what the extra spend buys.

Google’s API price list also gives buyers a date to budget for. Carbon’s introductory input and output rates of US$2 and US$10 per million tokens rise to US$4 and US$20 on 1 December. Neon’s Standard rates are US$3 and US$15.

scatter

Neon closes in on Carbon at a lower benchmark cost

Each point is one reasoning effort setting

Intelligence Index2030405060708002,5005,0007,50010,000Cost to run the index, US$lowmediumhighx highGemini 4 NeonGPT-6.2 SolGemini 4 Carbon
Artificial Analysis Intelligence Index scores and full benchmark costs by reasoning effort, 19 November 2026. Sources: Model Picker table of benchmark scores and full-index costs · Artificial Analysis launch comparison

People are asking

**Who can choose Neon?

** Google is rolling it out in the Gemini app to Google AI Pro and Ultra subscribers. Gemini for business users get it through their model picker and API.

**Which effort setting should I try first in the API?

** Use High, the default. For a difficult task, run the same example at Extra High and compare both answers against the source material.

IEA projects 70% rise in data centre emissions by 2030

What happened. In its 19 November World Energy Outlook, the International Energy Agency projects that global data centre electricity use will double to 1,000 terawatt-hours by 2030 under its Stated Policies Scenario. The IEA attributes three quarters of the increase to AI and projects a 70% rise in the sector’s carbon emissions, The Guardian reports.

The agency expects different sources of power across regions. Its 2030 figures put gas at about 60% of the electricity used by US data centres and renewables at about 70% of the electricity used by EU data centres.

Why it matters. The IEA identifies grid connections and equipment supply as constraints on data centre growth. Its work on connecting new data centres shows why location matters. It affects how quickly a provider can add capacity and which power sources it can draw on. Organisations buying AI services can use that information when they ask where a provider will run their workloads.

TWh

Data centre electricity use rises in all four IEA scenarios

2025 use compared with the IEA’s April 2026 scenarios for 2030

2025
485
2030 Headwinds
833
2030 High Efficiency
868
2030 Base Case
945
2030 Lift-Off
1,008
Global data centre electricity use in 2025 and under four IEA scenarios for 2030. Sources: IEA Key Questions on Energy and AI, Table A.1

People are asking

**Does the 70% figure refer to AI alone?

** It covers carbon emissions associated with all data centres. The IEA separately estimates that AI accounts for three quarters of the growth in electricity use.

**What can a business ask its AI supplier?

** Ask which regions will run the service and what the provider reports about electricity use and sources in each. The IEA’s analysis of data centre power explains why these questions matter.

ECB warns AI borrowing could spread losses across markets

What happened. In its 18 November Financial Stability Review, the European Central Bank describes how big technology companies sign long-term contracts for data centre and chip capacity. Infrastructure owners can borrow against the promised payments. The ECB’s analysis of company filings puts the firms’ total off-balance-sheet commitments at more than US$4tn in the second quarter of 2026.

Why it matters. The ECB says weaker demand or faster chip obsolescence could undermine the returns that companies expect from new capacity. Losses could then reach bondholders and private lenders financing the build-out. The ECB finds limited direct lending by euro-area banks to the AI supply chain and identifies their links with other financial firms as a route for losses to spread.

table

Where the ECB found AI exposure in the euro area

Lending commitments and holdings in US technology shares

ExposureECB measureScope
Direct bank lending commitments€24.3bn, including €11bn undrawn165 major AI supply-chain firms and their subsidiaries
Broader bank lending commitmentsAround €75bn, or 1.1% of corporate commitmentsData centre, software and semiconductor sectors
Non-bank equity holdings16–26% of equity investmentsUS technology and AI-related firms across non-bank sectors
Selected measures of euro-area exposure to AI-related companies in the ECB’s November 2026 review. Sources: ECB Financial Stability Review, November 2026

People are asking

**What’s an off-balance-sheet commitment?

** A cloud provider can promise to pay for future data centre capacity. The developer can use those payments to secure a loan. The ECB explains how the future payments can sit outside the cloud provider’s reported liabilities.

**Why does the ECB bring up subprime mortgages?

** Its comparison with the subprime boom concerns how hard it is to see who ultimately carries the risk when commitments and loans pass through several firms.

New releases

ChatGPT. Live Dictation displays text as you speak and removes the 10-minute recording cap, with a global rollout to all plans on web, desktop, iOS and Android. More

ChatGPT Work. The Accounting plugin is free to install on eligible Business and Enterprise desktops and helps reconcile transactions, draft invoice entries and prepare month-end reports, with administrator approval required on Enterprise. More

Claude Data. Anthropic has moved the date Claude Data switches on by default for Enterprise workspaces without a saved setting from 19 to 20 November. More

Gemini 4 Neon. Gemini 4 Neon is rolling out to Google AI Pro and Ultra subscribers and is available on the paid Gemini API at Standard rates of US$3 per million input tokens, US$0.15 per million cached input tokens and US$15 per million output tokens. More

Gemini Notebook. Google confirmed that interactive sites built from notebook sources are rolling out on the web to users aged 18 and over, including Workspace and Workspace Individual accounts. More

Gemini API. The Gemini API now accepts Word, Excel, PowerPoint and OpenDocument files directly and can process selected document pages, worksheet tabs, cell ranges or sections of audio and video. More

Project Genie. Project Genie is rolling out Genie 4 to eligible Google AI Ultra subscribers aged 18 and over, with live edits and sound in generated worlds, a creation agent and a gallery for sharing and remixing. More

Perplexity Computer. Perplexity Max subscribers in the US now have automatic personal skills on by default, letting Brain turn lessons from previous tasks into reusable instructions. More

Perplexity Computer. Perplexity Max subscribers can now call their AI teammates in the iOS and Android apps to assign work, ask for updates or change priorities. More

GitHub Copilot. Gemini 4 Neon is rolling out to GitHub Copilot Pro+, Max, Business and Enterprise, with usage-based rates of US$3 input, US$0.15 cached input and US$15 output per million tokens. More

GitHub Copilot. GitHub Copilot Business and Enterprise team maintainers and organisation owners can now make approved Agent Plugins the default for their team’s direct members. More

GitHub Copilot. GitHub Copilot’s public preview adds eight connectors, including Box, Calendly and Supabase, on all plans, with Business and Enterprise access controlled by administrators. More

Elsewhere in AI

UK financial regulators issued new directions on cyber defences for City firms after a rise in hacking attempts raised concerns about weaknesses exposed by AI, the Financial Times reports. Read

TikTok owner ByteDance is in talks to spend about US$500m more on Microsoft’s Azure cloud services, primarily to use OpenAI models, The Information reports. Read

The incoming chair of the US Senate Homeland Security Committee plans hearings after federal watchdog tests found that four major chatbots generated plans for violent attacks, Associated Press reports. Read

Four current and former Maryland high school students sued their school board, local police and AI vendor Omnilert after officers detained them over an alert that mistook a crisp packet for a gun. Read

Two US senators from opposing parties asked the director of national intelligence to identify and fix weaknesses in critical infrastructure after outages hit major companies and federal agencies. Read

Get The Distillation