The AI news that matters, explained for business and IT professionals.
A CIO column argues that open-weight models from Meta, Google and others are challenging reliance on a few closed providers, and that leaders should stop debating and start testing. It cites Vercel data showing open-weight models at 36% of AI Gateway token volume in July 2026, up from 11% in April.
The author frames the shift as a way to reduce cost, lock-in and opacity, starting from personal experiments and moving to team use. It is an opinion piece based on one practitioner’s experience, not a market study.
Pick one low-risk internal workload and run it on an open-weight model in parallel with your current provider, comparing quality and cost.
Pete Steinberger, creator of the OpenClaw assistant and now at OpenAI, told Cloudflare’s CEO at Cloudflare Connect 2026 that he spent $10 million on tokens last month, according to InfoWorld citing The Register. He said not all of that usage was useful.
Google has launched Agent Compute Engine, a service that suspends inactive AI agents and restores them on demand while keeping their execution state, according to InfoWorld. The company says it runs agents in hardware-isolated microVMs, which boot faster than conventional virtual machines and isolate better than containers.
A CIO column argues that when AI workloads underperform, adding GPUs may not help if the network cannot move data quickly and consistently. It notes that distributed workloads, especially training, require GPUs to exchange large volumes of data constantly.
According to Siècle Digital, Microsoft presented several OneDrive updates at a storage-focused event, including new Copilot capabilities in OneDrive and SharePoint. Some are available now and others will roll out over the coming months. One example cited is Copilot analyzing several spreadsheets at once.