AI model providers

Cybersecurity experts interviewed by CSO say that press reports and AI model providers use terms like 'rogue agents' or 'misalignment' when describing incidents involving AI systems, and argue this language obscures a simpler truth about what actually happened in those cases. The debate is drawing attention from technology and security commentators.

News coverage in the semiconductor sector focuses on how companies supplying chips for artificial intelligence are shifting toward strategic partnerships to develop infrastructure. The discussion centres on collaborations between chipmakers, cloud providers and data centre operators, alongside ongoing innovation in AI accelerator hardware, as demand for computing power continues to reshape the industry's business models.

Chinese lidar maker Hesai has launched Kosmo, a 3D data platform designed to support robot training and the development of world models. The platform aims to provide spatial data infrastructure for robotics and embodied AI applications, extending Hesai's business beyond automotive lidar sensing into the growing robotics sector.

FAICON, a startup working on sovereign AI, has received a 50,000 Swiss franc grant (about $49,000) from Venture Kick, the Swiss early-stage funding program. The grant supports the company's efforts to build AI technology that keeps data and model control within national or organisational borders. The news circulated via startup funding trackers and technology press.

Google has released Gemini 4 Argon, its newest flagship AI model, making it available to developers and businesses. The launch gives companies access to Google's most advanced AI capabilities for building products and services. The announcement is drawing attention from the tech industry as competition among major AI providers continues to intensify.

A writer and self-described LLM critic is calling for artificial intelligence models that individuals can train themselves on their own machines, with no data centers required. The alternative proposed is server-based models that are highly specialized and trained only on open data. The appeal reflects a broader debate among AI watchers about user control, privacy, and concentration of AI infrastructure in large companies, resonating with people who want more transparency and independence from big tech providers.

The search engine Ecosia has switched its AI features from French startup Mistral to open-weight models including Alibaba's Qwen, GLM and Moonshot's Kimi. The move highlights growing adoption of openly available Chinese AI models by European companies seeking cheaper and more flexible alternatives to proprietary providers.

TrustedRouter has raised a $6 million seed round to build an alternative to OpenRouter, the popular gateway for accessing multiple AI models, with a focus on data security. The startup is betting that companies want stronger guarantees about how their prompts and data are handled when routed between AI providers, positioning privacy as its main selling point against the incumbent.

AI cloud company Prime Intellect has acquired a startup focused on making model inference faster, according to a report by The Information. The deal signals continued consolidation in the AI infrastructure space, as providers race to cut the cost and latency of running large models. Details of the target and price were not disclosed.

Google has launched Gemini 4 Argon, its newest flagship artificial intelligence model, making it available to developers and business customers. The release continues the company's push to compete in the enterprise AI market, with the model expected to power applications and services built on Google's AI infrastructure. Further details on pricing, capabilities and availability were not immediately provided.

Alibaba chairman Joe Tsai said open-source AI is Europe's only realistic path to technological independence, speaking at a technology event in Italy. He warned European manufacturers against relying on closed American AI systems, arguing companies should retain control of the underlying technology rather than depend on US providers.

French cloud provider Scaleway is set to build a new AI supercomputer in Sweden powered by AMD chips. The announcement puts AMD in direct competition with Nvidia in the AI infrastructure market and adds to Europe's growing capacity for training large AI models.

Microsoft has introduced new Maia artificial intelligence chips, which the company says are designed to deliver greater token efficiency. The claim points to hardware that can process more AI work per unit of compute, reducing costs for running large language models. The announcement lands as major cloud providers race to build custom AI silicon and lessen reliance on Nvidia, though detailed performance figures were not provided in the initial report.

Meta's MTIA is the company's in-house AI chip family, designed to run the ranking and recommendation models behind its ads and news feed. New coverage explains how the chips work, their development roadmap, and their limits. Discussion centres on how far Meta can go in replacing external chip suppliers with its own silicon for AI workloads.

Microsoft cloud chief Scott Guthrie discussed the company's new custom AI chip for Azure in an interview, saying demand for AI computing power is 'insatiable'. He also addressed the ongoing price war among AI model providers, as cloud and chip rivals compete aggressively on cost and capacity to serve booming generative AI workloads.

Accenture, one of the world's largest consulting and IT services firms, faces questions about whether artificial intelligence will boost or undermine its business. The debate centres on whether AI automation will erode demand for the human-driven consulting and outsourcing work that underpins Accenture's revenues, or whether the company can reposition itself as a leading AI services provider.

Analysts argue Australia can strengthen its AI sovereignty by hosting and adapting foreign open-weight models domestically rather than relying on them as-is. Running the models on local infrastructure would make them more secure and better suited to Australian needs, even though their parameters were trained abroad. The discussion centres on reducing dependence on foreign providers and mitigating risks tied to overseas-controlled AI systems.

A new report asks whether artificial intelligence can foresee internet outages before they occur. The piece explores how AI models might analyse network traffic patterns to detect early warning signs of disruptions, potentially allowing providers to fix problems proactively. Interest reflects growing concern about network reliability as connectivity becomes essential infrastructure for work, commerce and daily life.