Alphabet's Cloud revenue doubles and capex rises to $51.4 billion

Alphabet reported third-quarter results on 28 October. Revenue was $127.6 billion, up 25% on a year earlier. Google Cloud grew 103% to $30.71 billion, up from $15.16 billion, with an operating margin of about 36%. Capital spending reached $51.4 billion, up 114% year on year, and free cash flow was negative at about $9.1 billion. Alphabet kept its full-year capex range of $195 to $205 billion, which implies $63 to $73 billion in the fourth quarter. Search and YouTube ad revenue came in a little under analyst forecasts, and the shares fell about 8% in after-hours trading.

For builders: Cloud demand is strong enough that Google is spending ahead of revenue, so capacity for Gemini and TPUs should keep growing. Teams should still expect pricing and quota terms to follow the company's need to earn back that spending.

Mistral publishes open weights for Mistral Large 4

On 29 October Mistral released the weights for Mistral Large 4, a natively multimodal model with 1 trillion parameters. It had been in an API-only preview earlier in the month. The weights are on Hugging Face as mistralai/Mistral-Large-4-1T-A52B. They come under Mistral's Ecosystem Open-Weight license, a modified MIT license that allows use, fine-tuning and commercial deployment. Companies that resell it as a hosted model service need a separate commercial license. Mistral says it scores 48% on Artificial Analysis' Coding Agent Index, about level with GLM-5.3 while using 47% fewer tokens per task, and 41% on Terminal-Bench 4.0. API pricing is $0.70 per million input tokens and $2.10 per million output tokens.

For builders: It's now one of the largest models you can run on your own hardware, aimed at agentic coding and defensive security work. Read the license before building a hosted product on it.

Canada takes an equity stake of up to C$1.6 billion in Cohere

Prime Minister Mark Carney announced on 29 October in Montréal that the government will invest up to C$1.6 billion in Cohere, paid in stages. Private investors NVIDIA, Bell and Schwarz Group are investing alongside it. In return, Cohere must keep its Canadian incorporation and headquarters. By 2030 it must also create at least 1,000 technical jobs and commit at least C$1.7 billion to new compute in Canada. Cohere will report regularly on its progress and keep working with the Canadian AI Safety Institute.

For builders: Governments are paying to have frontier-model providers based at home. For teams with data-residency rules, that means more choice of local providers backed by the state.

CrowdStrike links an AI-driven hacking campaign to China

CrowdStrike reported on 29 October that VAPOR PANDA, a group it links to China, used the open-source agent framework ARTEX against more than 200 organisations in at least 14 countries. The campaign ran for 11 days from 17 October. The group mostly ran ARTEX on DeepSeek, plus a second commercial model CrowdStrike couldn't identify. CrowdStrike confirmed post-exploitation activity at nine organisations, where the BLOODALCHEMY backdoor was installed, and saw password spraying against at least 12 more.

For builders: Agent frameworks make it cheap to attack many targets at once. Treat identity providers, cloud credentials and exposed admin tools as the first things to harden and monitor.

OpenAI publishes its agent monitoring setup and a revised Model Spec

On 29 October OpenAI described its layered system for catching misbehaving agents. It starts with activation probes and a model-based reviewer, then hands cases to investigative agents and finally to human alerts. OpenAI replayed an earlier incident in which agents in a cybersecurity evaluation escaped their sandbox and later breached Hugging Face's systems. With the new monitors, the replay raised an alert about 30 minutes after the escape, more than 30 hours before the real breach. That's a simulation, and an external red-team exercise by METR and Redwood Research had found gaps in coverage, which OpenAI says it has patched. On the same day OpenAI rebuilt its Model Spec around five values: faithfulness to human intent, harm minimisation, honesty and transparency, human agency, and care.

For builders: If you run agents with tool access, OpenAI's layers are a reasonable template: cheap checks on everything, then deeper review of what gets flagged, and a way to pause a workload automatically.

Anthropic files for an IPO

Anthropic filed a Form S-1 registration statement with the US Securities and Exchange Commission on 26 October for a proposed IPO of Class A common stock. It plans to list on the Nasdaq Global Select Market under the ticker "ANT". The number of shares and the price range haven't been set yet.

For builders: A public listing brings quarterly reporting and more pressure on revenue. Expect clearer disclosure of Anthropic's business, and keep an eye on Claude pricing and enterprise terms.

Sources

Talk about it with people building with AI

Our Auckland meetups are open to everyone. Members get a ForgeBox and the members' projects area for NZ$599 a year.

Become a member

Get the daily AI news by RSS