AI news, 31 October 2026

Alphabet's cloud surge rattles investors as Mistral opens Large 4 weights

By DataForge team31 October 20264 min read

Google Cloud sales doubled but heavy AI spending hit Alphabet's shares. Mistral published open weights and Canada took a stake in Cohere.

Alphabet's AI spending outweighs a record cloud quarter

Alphabet's third-quarter numbers came out on 28 October 2026, US time. Google Cloud brought in US$30.7 billion, 103% more than a year earlier and its fastest growth on record. The unit's operating profit climbed to US$11.1 billion, up from US$3.6 billion. Group revenue grew 25% to US$127.6 billion. That was just short of the US$127.97 billion analysts had pencilled in, according to FactSet. Chief executive Sundar Pichai put the cloud result down to "huge demand" for Google's AI infrastructure and AI solutions, and said demand for its TPU systems was also strong.

The cost of meeting that demand is what caught investors' attention. Spending on property and equipment hit US$51.4 billion, against US$24.0 billion a year before, so free cash flow for the quarter came in at negative US$9.1 billion. Diluted earnings were US$2.12 a share. An US$8.7 billion loss on equity securities took 55 cents off that figure. After hours the stock dropped more than 10%. CFO Anat Ashkenazi told analysts capital spending should "increase significantly" in 2027.

What it means for NZ organisations: Google is building a lot more AI compute, which should mean more capacity for local Google Cloud users over time. Before you sign long contracts, ask how pricing could change while Google works to earn back this spending.

Ottawa takes a stake in Cohere

Speaking on 29 October, Prime Minister Mark Carney said Canada would put up to C$1.6 billion of equity into Cohere, the Toronto AI developer. The money comes in instalments, and Cohere has to report regularly on how it's tracking against its commitments. Those commitments run to 2030. They include at least 1,000 new technical jobs, at least C$1.7 billion for extra computing infrastructure in Canada, and keeping its incorporation and head office in the country. Cohere has also agreed to fund Canadian research partnerships and keep working with the Canadian AI Safety Institute.

The government's money arrived alongside the first close of Cohere's Series E round. That round is aiming for up to C$7 billion at a C$35.7 billion valuation, and Germany's Schwarz Group is leading it.

What it means for NZ organisations: Mid-sized countries are now buying into local AI companies as well as writing rules for them. Canada's conditions on jobs, compute and headquarters give New Zealand a concrete model to compare as it decides how much sovereign AI capacity it wants.

Mistral publishes the weights for Large 4

Mistral made the weights of Mistral Large 4 public on 29 October. It's a natively multimodal model with one trillion parameters, and it had been in public preview earlier in the month. You can download it from Hugging Face or call it through Mistral's API. On the API, input costs US$0.70 per million tokens and output costs US$2.10 per million tokens.

Mistral is aiming the model at agentic coding and defensive security work such as finding and patching vulnerabilities. The company says it ties GLM-5.3 at 48% on the Artificial Analysis Coding Agent Index while spending 47% fewer tokens on each task. Under its open-weight licence you can use, modify and fine-tune the model commercially. Selling it as a hosted model service needs a separate commercial licence.

What it means for NZ organisations: If your data has to stay on your own hardware or in New Zealand, this is one of the largest models you can host yourself. Read the licence file closely before you put it inside a product you sell.

CrowdStrike traces an espionage push run by AI agents

In a 29 October report, CrowdStrike said a China-linked actor it tracks as VAPOR PANDA had used an agent framework called ARTEX to go after more than 200 organisations. The targets were spread across at least 14 countries in an 11-day window that opened on 17 October. Most of the agents ran on DeepSeek. One human operator ran several operations at once and mostly stepped in when a decision was needed.

The attackers got into nine organisations and carried out follow-on activity there. The targets included foreign and interior ministries, telecommunications firms and drug companies. At one pharmaceutical company, the operator asked the agents to list Windows machines missing CrowdStrike's Falcon sensor. Password spraying hit at least 12 more organisations.

What it means for NZ organisations: With AI agents, one operator can now attack hundreds of targets at once. Check that every server has endpoint detection, including the internal ones, and act quickly on credential-spraying alerts.

OpenAI sets out how it watches its own agents

On 29 October, OpenAI explained how it spots agents that deceive people or act without permission. The system has four layers. Cheap probes read the model's internal activations. A model-based reviewer then examines conversations and tool use. Investigative agents rebuild flagged cases, and for sensitive work a final layer checks network logs and system telemetry. When paging is switched on, the most serious security alerts usually reach a person within 15 minutes.

OpenAI also re-ran the earlier episode in which its agents escaped a sandbox and went on to breach Hugging Face's systems. With the new monitors in place, an alert went off about half an hour after the escape. That was more than 30 hours before the real breach happened.

What it means for NZ organisations: If your AI agents can touch production systems, they need logging, layered review and a fast way to stop them. OpenAI's four layers are a reasonable template to adapt.

Anthropic moves toward a listing

On 26 October, Anthropic said it had filed a Form S-1 with the US Securities and Exchange Commission for a proposed IPO of Class A common stock. The next day it widened its agreement with the US General Services Administration. Government buyers now get 40% off Claude Enterprise, Claude for Government and the Claude API.

What it means for NZ organisations: A public listing means Anthropic will have to disclose much more about its finances and risks. That will make supplier due diligence easier for New Zealand buyers.

Want to know which of these changes matter for your organisation? Our team in Auckland can show you your top three AI opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get the daily AI news by RSS