October 27

Today’s Note,

The AI race is becoming a contest of scale, trust and staying power. Anthropic has filed to go public, SpaceX has secured a $40 billion commitment to finance AI chips, and OpenAI is pushing its technology into always-on assistants and scientific research.

The question is no longer just what AI can do, but who can build a business around it, and how reliably the technology can deliver.

In today’s Noted:

Anthropic files for an IPO as Claude expands across businesses

We noted:

Anthropic, the company behind Claude, filed a registration statement with the US Securities and Exchange Commission on October 26 for a proposed initial public offering. It intends to list its Class A shares on Nasdaq under the ticker ANT. The number of shares and the proposed price range have not yet been announced.

Anthropic’s filing gives investors a more detailed look at how the company plans to grow. Its strategy extends beyond selling access to Claude models: it is encouraging businesses to use Claude across its developer platform, Claude Code for software development and Claude Cowork for broader knowledge work.

Why it matters

A public listing would give investors a way to buy shares in Anthropic once the offering is approved and completed. For businesses using Claude, the more immediate development is the company’s push to make its AI useful across more of their daily workflows, not just through a chatbot.

SpaceX commits $40 billion to finance AI chips

We noted:

SpaceX announced a binding commitment for approximately $40 billion in GPU financing through a master finance lease with an entity managed by Apollo Global Management. The arrangement is intended to finance Nvidia’s next-generation Vera Rubin chips and related data-centre equipment for SpaceX’s computing infrastructure.

Under the arrangement, the Apollo-managed entity will purchase the chips and related equipment and lease them to SpaceX. SpaceX says the infrastructure will support Grok, its AI services and cloud-computing customers.

Why it matters

The financing is not a new AI product that consumers can immediately use. It is a bet on the infrastructure required to serve more AI workloads, and the actual return will depend on how quickly SpaceX deploys the equipment and generates enough business to justify the cost. Companies with substantial revenue from other businesses may have more options for financing infrastructure than companies whose income depends almost entirely on renting out computing capacity.

OpenAI says its AI models have produced three major scientific results

We noted:

OpenAI announced three research results involving mathematics, computer science and theoretical physics, and is trying different approaches to involve independent academic experts in understanding and verifying the work.

The results concern the local smoothing conjecture in harmonic analysis, a proposed disproof of the Exponential Time Hypothesis in computer science, and mathematical existence results related to quantum field theory.

Why it matters

Formal verification is not the same as establishing the full significance of a discovery. Researchers still need to examine the assumptions, understand the implications and determine what new algorithms or scientific insights follow. If the claims hold up to independent scrutiny, AI could become more than a tool for summarising research or helping scientists write code, and could contribute to generating new mathematical results. The practical consequences, particularly for the computer science result, remain to be established.

OpenAI is turning coding tools into Dots, an always-on personal assistant

We noted:

OpenAI is extending ideas from its coding tools into Dots, its always-on personal assistant designed to work towards user-defined goals in the background rather than wait for a new prompt each time.

Traditional chatbots generally respond when a user sends a message. Dots is designed to keep working over time, using connected tools and context to pursue tasks with less repeated instruction.

Why it matters

The potential use cases include monitoring ongoing work, preparing updates, organising information and following up on tasks. The value will depend on how reliably Dots can maintain context, handle interruptions and ask for help when it reaches a decision it should not make alone. If AI assistants can work continuously, the competitive difference may shift from how well a chatbot answers a question to how much useful work it completes without constant supervision.

AI coding agents are learning to test whether software actually works

We noted:

OpenAI described how Greptile is using GPT-6 Astra to test applications, identify potential security issues and help developers catch bugs beyond what a code review alone can reveal. Separately, CodeRabbit is using GPT-6.1 Sol to investigate security risks across complex codebases.

Reading code can reveal logical problems, but some bugs only appear when an application is running. A button might disappear at a particular screen size, for example, or a change in one component might break another part of the product.

Why it matters

AI coding tools may increasingly check not only whether code looks reasonable, but whether the application behaves as expected. These systems can help developers catch more problems earlier, although they do not eliminate the need for human review and established testing tools. AI is making it faster to produce code, but more code does not automatically mean more reliable software. Testing and verification are becoming essential parts of the AI-assisted development workflow.

Products worth trying

 

Everything else in AI

Anthropic faces a bigger question than its IPO: what kind of company is it?

Anthropic’s public messaging has long emphasised AI safety and its public-benefit structure. Its IPO filing now puts that identity alongside the realities of running a rapidly growing commercial AI company, including the need for compute, revenue and sustained investment.

MIT Technology Review examines the tension between Anthropic’s safety-oriented narrative and the influence it is gaining over how AI is developed and discussed. The question is not simply whether the company says the right things, but how its incentives, decisions and responsibilities align as its influence grows.

 

OpenAI releases data on how people use ChatGPT for work

OpenAI released aggregated data covering 165 work activities across nearly 100 countries, giving researchers a way to examine how ChatGPT use at work changes over time. The dataset spans the first quarter of 2025 through the third quarter of 2026.

Across 74 countries observed in every quarter, the ten most common activities accounted for an average of 51.1% of classified work messages in Q3 2026, down from 56% in Q1 2025. That suggests work-related use is spreading across a wider range of activities, although the data covers individual ChatGPT plans, not enterprise conversations or Codex usage.

 

Fanatics uses AI to recommend complete outfits for sports fans

Fanatics and OpenAI launched an updated Shop the Look experience that recommends coordinated merchandise when customers browse products such as team jerseys. The system considers product information, team affiliation, stock availability and merchandising rules.

In testing across roughly 30,000 MLB products, Fanatics reported 10% higher engagement than its previous experience. That is a company-reported result, but it illustrates a practical use for AI in ecommerce: helping customers find complementary products rather than simply showing them more items.

 

OpenAI says Codex has cut incident investigations from hours to minutes

OpenAI engineers are using Ultrafast, a high-speed mode in Codex, to investigate incidents by searching logs, Slack discussions and other data sources in parallel. One security architect reported that work previously taking an hour or two to gather and contextualise now takes 10 to 15 minutes.

The example shows how AI agents can support incident response by bringing information together quickly, but the reported improvements come from OpenAI’s own internal experience. Teams still need engineers to interpret the findings and decide what action to take.

See you tomorrow
same time, different note.

~ Stackari