Pin this number above the AI spending debate: $33.2 billion. Microsoft has finally put a quarterly dollar figure on Azure, and it is the clearest answer in today’s AI news today brief. Azure revenue reached $33.165 billion in the September quarter, up 48% year over year—evidence that cloud demand is catching up with the industry’s spectacular infrastructure bill. Not all Azure revenue is AI, of course, and Microsoft still spent a record $45.2 billion on capital projects during the quarter. Elsewhere, Google paired a cyber-focused frontier model with a cloud business that more than doubled; Apple put touch on the MacBook Pro at last; Microsoft and GitHub standardized the machinery around coding agents; OpenAI moved deeper into ServiceNow workflows; and Anthropic expanded free Claude access for science. The market is no longer asking whether companies can build impressive AI. It is asking where the revenue, controls, and durable workflows show up.

AI news today: Microsoft reveals Azure’s $33.2 billion quarter

Microsoft reported $92.3 billion in fiscal first-quarter revenue, up 19%, with Azure revenue increasing 48% to $33.165 billion. This is the first time the company has disclosed Azure’s quarterly revenue in dollars rather than only a growth rate, making direct comparison with Amazon Web Services and Google Cloud much easier.

The number is substantial but needs a clean label: Azure includes ordinary cloud computing as well as AI workloads. TechCrunch noted that Azure accounted for roughly 36% of Microsoft’s quarterly revenue and exceeded analyst expectations. Microsoft Cloud overall reached $63.9 billion, while contracted commercial backlog rose 73% to $677 billion.

Why it matters: AI infrastructure is producing measurable cloud sales, but the cash race is not slowing. Microsoft’s quarterly capital expenditure reached $45.2 billion. The next test is whether growth stays ahead of depreciation, energy, and financing costs as new capacity comes online.

Google launches Gemini 4 Carbon as Cloud revenue doubles

Google introduced Gemini 4 Carbon, positioning it as its strongest model for cybersecurity and long-running engineering work. Google reports scores of 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym, and 71% on CWE-bench for patching. Those are company-selected benchmark results, not proof that Carbon can safely run an unsupervised security program.

Access is deliberately narrow: a subset of cyber defenders will receive it first through Google’s Fairwind Program, with broader developer and consumer availability promised later. Google says eventual introductory pricing will match Argon at $2 per million input tokens and $10 per million output tokens.

The launch landed beside Alphabet’s October 28 earnings filing: Google Cloud revenue rose 103% to $30.709 billion, and the company said the Gemini app reached one billion users. Alphabet also tied higher costs to AI infrastructure, talent, and TPU inventory. In other words, Carbon arrives inside a business already monetizing AI at scale—and spending accordingly.

Apple brings touch—and Dynamic Island—to MacBook Pro

Apple unveiled 14- and 16-inch MacBook Pro models with touch displays, its first touch-enabled Macs. The new PROLED XDR tandem-OLED panel supports up to 2,000 nits of HDR and SDR brightness, a 120Hz variable refresh rate, and Apple’s Dynamic Island for Spotlight, Live Activities, and Siri AI. Models use M5 Pro or M5 Max chips, support up to 128GB of unified memory and 16TB of storage, and start shipping November 6.

Apple’s performance and battery claims still need independent testing, particularly the promised 24-hour battery life. The strategic shift is clearer: Apple is no longer treating touch as an iPad-only interaction. For developers and creative-tool makers, the watch item is whether macOS apps gain genuinely touch-native interfaces or simply larger targets for fingers.

Microsoft and GitHub standardize the machinery around coding agents

Microsoft made its Copilot harness available through Foundry. The harness—originally built by GitHub—supplies the instructions, context, sandbox, tool access, routing, and controls around a model. Microsoft says it supports more than 20 frontier models and now powers Copilot Code, Copilot Studio, GitHub Copilot, the CLI, and VS Code.

The same day, GitHub put AI-application code scanning into public preview. It can flag indirect prompt injection, sensitive-data exposure, and excessive agency in pull requests, but requires GitHub Advanced Security and is not yet available on GitHub Enterprise Server. The useful connection is obvious: agents need both a runtime and a reviewer. The unresolved question is how often the scanner catches exploitable architecture without burying teams in false positives.

OpenAI and ServiceNow connect ChatGPT to governed workflows

OpenAI and ServiceNow expanded their partnership with a ServiceNow plugin that is available in the OpenAI marketplace today. Customers choose which workflow actions ChatGPT may use, while ServiceNow’s Action Fabric applies existing permissions and business rules. Planned capabilities include Codex inside ServiceNow’s AI Workflow Factory, a joint cyber-risk product, and voice-assisted risk assessments.

The dates matter. Enhanced plugin functions are expected by year-end; security tools enter early access in January 2027, with general availability planned for the second quarter; the Codex integration is slated for 2027. This is therefore a mix of shipped integration and roadmap, not one finished platform. It also follows yesterday’s ServiceNow–Google announcement, underscoring ServiceNow’s bid to become the neutral control layer across model providers. See our October 28 brief for that wider platform context.

Anthropic expands AI for Science to 20 organizations

Anthropic announced partnerships with 20 foundations, institutions, and research groups that will receive free Claude credits. Participants include the U.S. National Science Foundation, the Sloan and Simons foundations, the Space Telescope Science Institute, the Wellcome Sanger Institute, and EMBL.

The company points to early projects such as a research-preview standard for agents operating lab instruments and Allen Institute teams using agents to analyze public scientific datasets. These are program descriptions, not independently measured productivity results. What comes next should be public evidence: datasets, reproducible methods, open-source tools, and time saved without sacrificing scientific review. Anthropic’s capacity to fund efforts like this is also easier to assess after its recent IPO filing.

Watch & Learn

Editor’s note: Kevin Stratvert’s verified 12-minute tutorial, “How to Use Claude Artifacts,” walks through building documents, dashboards, small apps, persistent data, and connected workflows. It is useful for non-coders and product teams who want a practical view of what an artifact can do before building the mini-tool below.

AI, Translated: agent harness

An agent harness is the operating layer around an AI model. It supplies instructions, context, tools, a sandbox, permissions, memory, and checks that let the model complete a task. Think of the model as an engine and the harness as the vehicle: steering, brakes, dashboard, and road rules included. Microsoft’s new Copilot harness can route work among several models. You should care because reliable agents depend as much on this surrounding system as on raw model intelligence.

Try This Today: build an AI-spending calculator in Claude

Goal: turn headline cloud numbers into a simple, editable decision tool in about ten minutes.

1. Open Claude and ask for an interactive Artifact; enable Create and edit files under Settings → Capabilities if needed.
2. Paste three numbers: current quarterly revenue, year-over-year growth, and quarterly capital expenditure.
3. Test the calculator with deliberately bad inputs—blank fields, negative growth, and a currency symbol—then ask Claude to add validation and a clear assumptions panel.

Copy-ready prompt: “Create a private interactive Artifact that compares quarterly cloud revenue with capital expenditure. Include inputs for revenue, year-over-year growth, capex, and scenario growth; calculate annualized revenue, capex-to-revenue ratio, and next-year revenue under the scenario. Label every assumption, validate invalid inputs, and do not fetch external data.”

Anthropic’s current help page says Artifacts are available on Free, Pro, Max, Team, and Enterprise plans, but file creation must be enabled. Keep the first version private; connected apps and persistent storage have additional plan and administrator requirements.

One thing to remember

AI’s scoreboard is becoming more legible: revenue, capital intensity, permissions, and repeatable workflows. Model benchmarks still matter, but businesses ultimately pay for systems that work—and can be governed—after the demo ends.