AI news, with the context that matters.

Latest articles

Anthropic's IPO: Japan's retail order deadlines and pricing rules, six-times demand and the revenue forecasts

In Anthropic's IPO, online retail orders in Japan close at 10:00 a.m. on November 18 at Mizuho Securities, 10:59 a.m. at SBI Securities and 2:00 a.m. on November 19 at Rakuten Securities (all Japan time, all planned). The offer price is published 7 to 23 hours later and can fall outside the price range. We checked the Japanese, EEA, UK and Australian filings with the SEC and the brokers' notices for deadlines, price limits, currency and the share counts by region, and worked back from reported demand and revenue forecasts to an implied October–December quarter.

Business & Markets
Read article →

Jensen Huang resigns and Colette Kress becomes Nvidia's interim CEO: the share reaction, six cases of a CFO as interim CEO, and the first board chair

On November 12 Nvidia announced that Jensen Huang had resigned as president, CEO and director, and that CFO Colette Kress would serve as interim CEO while remaining CFO. Director Harvey Jones became the company's first-ever board chair. The 8-K gives no reason for the resignation and no findings from the board's inquiry. In six past cases where a CFO served as interim CEO, a permanent CEO was named 48 to 224 days later, with a median of 110 days. We go through the 8-K, the share reaction, why Nvidia had no chair, and Huang's stake and exit terms.

Business & Markets
Read article →

OpenAI and Anthropic in the third quarter: $10 billion versus $17 billion of revenue, compared on matched accounting and per dollar of revenue

The Information and the WSJ reported on November 10 that OpenAI's July–September 2026 revenue was about $10 billion, with an operating loss excluding stock pay of about $10.9 billion. Anthropic's second amended S-1, filed the same day, shows revenue of $16.956 billion and an operating loss of $567 million for the same quarter. We compare the two after aligning how cloud-channel sales are counted in both directions, and calculate results per dollar of revenue, run-rate and valuation multiples, the IPO terms and placement, and OpenAI's 2026 plan.

Business & Markets
Read article →

Nvidia's board inquiry into Jensen Huang and the ouster reports: a 19-point gap with chip stocks, past CEO exits and the terms for H200 sales to China

Nvidia's board opened an independent inquiry on November 8 after reports that CEO Jensen Huang had been the target of a foreign intelligence effort. On November 11 the WSJ reported that the board was preparing to remove him, and the shares fell 8.6% after hours. From the November 6 close to November 11, the SOX semiconductor index rose 12.5% while Nvidia fell 6.9%. We calculate the loss of market value, compare past CEO exits, and look at Huang's stake, the terms for H200 sales to China and Nvidia's stakes in AI companies.

Business & Markets
Read article →

Reported intrusions by AI agents: what Operation Snowball confirms, and the gap between organizations targeted and organizations breached

On November 8, Malbear Labs published Operation Snowball, in which an agent running on DeepSeek Flash attacked a European telecom operator for more than two weeks. Sorted by strength of evidence, the claims of 13 breached organizations and millions of patient records rest on the agent's own records, with no confirmation yet from victims. With the reports around ARTEX, we look at how old the vulnerabilities were, the gap between targets and breaches, the time attacks and detection took, alerts in Japan, and the defenses the reports share.

Society & Policy
Read article →

Anthropic's retracted IPO terms and its amended S-1: a third quarter back in the red and $895.4 billion of compute commitments

On November 9, Anthropic sent banks IPO terms to raise up to $69.3 billion at $70 to $80 a share, then retracted them the same day as inaccurate. Its November 2 amended S-1 shows third-quarter revenue of $16.956 billion and an operating loss of $567 million. We calculate the valuation implied by the retracted range and the per-share value the filing implies for September 30, check LATENT's October 30 revenue estimate, break down $895.4 billion of commitments, and compare the seven placement buyers with what Anthropic pays them.

Business & Markets
Read article →

Claude Code's "Don't ask again" and secrets in public repositories: inside 1,122 files, and the 19% two scanners missed

Security firm Apiiro examined 104,185 Claude Code settings files on public GitHub and found 1,379 secrets in 1,122 of them. Choosing "Don't ask again" at an approval prompt saves the command, password included, as a permission rule. We checked the mechanism against the original report, Claude Code's documentation and changelog, and GitHub issues, and calculated the kinds of secrets, how the share rises with file size, what the scanners missed and how many secrets remain in history. We also include steps to check your own repositories.

Products & Services
Read article →

Data center pause pledges after the US midterms and the chip selloff: 256GW that could stop, and the roughly 11GW expected to stop

In the November 3 US midterms, governor candidates who backed pauses or conditions on data center construction won in Texas, Georgia, Ohio and elsewhere, and the Philadelphia Semiconductor Index fell 12.25% on November 5. The pause states are said to hold 40% of US capacity, but most of it is proposals that have not broken ground. We converted an analytics firm's figures into GPU counts and investment dollars, and calculated how much is actually expected to stop, how household power prices moved by state, and where OpenAI's new commitments overlap with governors' conditions.

Chips & Infrastructure
Read article →

What Revolut pays to run GLM-5.3 on its own GPUs: unit prices against commercial AI, and the engineer count where self-hosting pays off

UK fintech Revolut moved coding assistance for about 2,000 engineers to GLM-5.3 running on 56 B200 GPUs. It puts the blended cost at $0.30 per million tokens and says its bill is one-tenth of what it paid Anthropic and OpenAI. We checked the figures against the talk transcript, compared them with today's list prices using the token mix from Bun's rewrite, estimated the monthly GPU cost, and calculated how many engineers a company needs before self-hosting pays off.

Chips & Infrastructure
Read article →

Where the Money for AI Comes From: Free Cash Flow at Four Big Tech Companies, OpenAI's Smaller Funding Target and Anthropic's Rating Outlook

In July–September 2026, capital spending at Microsoft, Alphabet, Amazon and Meta exceeded their operating cash flow by about $26 billion. A year earlier the same four had roughly $62 billion left over. We put the filings on a common footing and calculated the new borrowing and share sales, the size of AI company contracts within cloud backlogs, and the funding position of OpenAI and Anthropic.

Business & Markets
Read article →

What Anthropic's S-1 reveals about its operating profit: estimated third-quarter revenue, a $3.1 billion monthly SpaceXAI contract, and four unauthorized-access incidents

The only operating profit in Anthropic's S-1 quarterly table is April to June 2026, and the filing says it buys compute 12 to 18 months ahead. From the table and the contract terms we calculate why the profit appeared, the third-quarter revenue left blank, and the fixed costs that rise from December. We also compare the incident the filing uses as its example with the one Anthropic said in September concerned it most.

Business & Markets
Read article →

Gemini 4 Carbon goes to cyber defenders first. Its gap over Argon in finding, exploiting, and patching vulnerabilities, and an introductory price that doubles when it ends

Google is giving Gemini 4 Carbon to a subset of cyber defenders first and has not said when developers get it. We read the gap over Argon in three stages, finding, exploiting, and patching vulnerabilities, from DeepMind's performance table and evaluation methodology. We also sort out the introductory price, which matches Argon's, the price after it ends, the Fairwind Program's conditions, and how many days Argon took to reach developers.

Products & Services
Read article →

Can Genie 4 replace a game engine? How Gemini watches over a generated world, and where it falls short

Google DeepMind's Genie 4 is a world model that generates video and sound together in response to key presses and text instructions. We read the official description of how Gemini 3.8 Flash watches the scene and steers it, then compare Genie 4 with game engines, which keep world state as numbers, and sort out what game developers can try with it.

Technology & Research
Read article →

What does one computer-use task cost on GPT-6.1 Terra, and what do you pay for the gap to Astra?

GPT-6.1 Terra's API rates are one tenth of GPT-6 Astra's. Using every data point in OpenAI's launch charts, we read the cost of one computer-use task at each reasoning effort, then estimate what you pay for Astra's higher scores and what screenshots cost.

Products & Services
Read article →

Can Nemotron 3.5 Bolt E4B share a GPU with your game as a conversational NPC?

NVIDIA's roughly 4GB small model for game characters. We read the published 3.5GB figure on an RTX 5070 against its test conditions, then look at the VRAM left for the game, response time, and use in Japanese.

Products & Services
Read article →

What can you build with OpenAI Decisions API? Turning everyday language into application actions

Follow a player’s request from conversation to an NPC’s action, then explore responsive voice interfaces, work routing, and document and image review. When is a separate decision API worth using alongside a conventional LLM?

Technology & Research
Read article →

What GoTo’s regional-language AI reveals about language costs and reasoning strengths

Equivalent meaning can take different numbers of tokens across languages and tokenizers. Starting with GoTo’s Gemma 3 derivative, we separate billing, inference computation, reasoning text, and internal representations, and examine when English translation helps and when the original matters.

Technology & Research
Read article →

DeepSeek’s new model demonstrates game development in Godot

Starting from DeepSeek’s Godot demo, we follow a one-room game through Godot, Unity 6 and UE 5.8: requirements, scene edits, compilation, input and logs, diagnosis and saving, with concrete roles for agents and developers.

Products & Services
Read article →

Muse Spark Open releases its weights: can you own a powerful AI at home?

An 816B model now runs across two Mac Studios. Its 42B active parameters, roughly 493GB of quantized weights, and distributed inference measurements show what changes for a local-LLM budget of one or two million yen.

Products & Services
Read article →

Nemotron 3.5 Bolt versus 3 Super: compression and the conditions behind 24GB

Nemotron 3.5 Bolt brings a large model into a local memory budget. Comparing it with Nemotron 3 Super reveals how structural compression, quantization and speculative decoding work—and what the memory and speed figures actually measure.

Products & Services
Read article →

Why Anthropic moved from Git to Scion

Supporting more than 30,000 daily merges takes more than faster coding. Scion changes how code changes are recorded, history is queried and files are fetched, reducing the work done by development infrastructure.

Technology & Research
Read article →

Anthropic’s distillation defenses and the limits on reusing model reasoning

Anthropic combines restrictions on reasoning reuse, classifiers, and account controls. We examine what the reported 90–99% reduction measures, the full-trace versus summary experiment, and the implications for legitimate agent developers.

Technology & Research
Read article →

Muse Wonder and Dots: different ways to delegate work

Separate Muse Wonder models from Muse Code and the Muse app, then compare tools, computers, memory, integrations, approvals, prices and access from Japan with Dots.

Products & Services
Read article →

ColGrep-Agent delegates code search to a local 2B model

How a local 2B model searches for a larger coding agent, what the token savings trade off against time and accuracy, and how this relates to local LLM configurations.

Technology & Research
Read article →

How much easier can Qwen-Image-2.1-Turbo make creative iteration?

Eight-step image generation and editing can shorten waits, but inspection remains. Background art, character consistency and Japanese lettering reveal the tradeoffs and hardware questions.

Products & Services
Read article →

Gemini agent, Dots, Muse, Copilot and Cue: how far can we entrust work to AI?

Gemini, Dots, Muse, Copilot and Manus Cue compared on execution, memory, apps, approval, identity, audit, costs and availability, as of October 9, 2026.

Products & Services
Read article →

Can Liquid AI d1 run on the same PC as your game?

d1-3B and the experimental 600M return decisions in one pass. What the published 8 ms really measures, and how to budget memory and integrate decisions without stalling a game.

Technology & Research
Read article →

Search photos and audio by meaning: what EmbeddingGemma 2 brings to local retrieval

EmbeddingGemma 2 puts text, photos, audio and video into a shared representation for search. A guide to its role alongside generative AI, the steps in a local index, its model footprint and the trade-offs in shorter vectors.

Columns
Read article →

When popularity raises the AI bill: what happened to Teach My Little Sister How to Drive

A surge in demo players brought AI costs, service limits and conversation quality into the same problem. What the developer’s account tells us about funding ongoing AI in a game sold once, and the alternatives offered by user plans and local processing.

Columns
Read article →

How practical is a local LLM you can actually rely on?

What can today’s Mac, DGX Spark, and Windows PCs offer someone who wants to own their AI? A practical look at Japanese quality, published performance evidence, VRAM and RAM, Strata, and what ¥1 million or ¥2 million buys.

Columns
Read article →