Archive

Due 2026-10-15 We expect another software or cloud provider – not Microsoft, Oracle, Google, AWS or OpenAI – to build a strong AI model from another maker into its own product by mid-October.

Due 2026-10-31 We expect Microsoft, AWS or Google to expand an offering for companies by the end of October that lets them run strong AI models from other makers both in the cloud and in their own data centres – with control over rules and costs.

Due 2026-10-31 We expect Mistral to release the weights of its Mistral Large 4 model by the end of October, as announced, so that companies can run it themselves.

Due 2026-11-30 We expect Microsoft to do two things by the end of November: offer another AI model from a maker other than OpenAI in Copilot or Foundry – and move one of its own products onto an in-house AI model.

Due 2026-11-30 We expect one of two things by the end of November: OpenAI, Anthropic or Google bring one of their models to another company's cloud – or Meta, NVIDIA or Mistral release a freely downloadable model that is explicitly made to run on PCs or laptops.

Due 2026-11-30 We expect another major provider – not Microsoft, AWS or Snowflake – to introduce by the end of November either a switchboard that spreads requests across AI models and caps spending, or a way for AI agents to buy services themselves under supervision.

Due 2026-12-15 We expect Microsoft, Google, AWS, OpenAI or Anthropic to release a finished offering for companies by mid-December that brings cloud AI and on-device AI together under one product – with central control over who may use what.

Due 2026-12-15 We expect another maker of strong AI models – not OpenAI or Anthropic – to give business customers their own tools for monitoring how its models are used by mid-December, or to tie access to conditions, for example when a customer changes owner.

Due 2026-12-31 We expect Google, OpenAI, AWS or Salesforce to let administrators choose between at least two AI models for the same task in one of their business AI products by the end of the year – inside the product, not just for developers.

Due 2027-01-31 We expect a maker of strong AI models – not OpenAI – to offer its newest flagship model through AWS, Microsoft or Google within 14 days of its launch, by the end of January.

Due 2027-01-31 We expect another major provider – not OpenAI or Microsoft – to tie cheaper AI use to fixed rules by the end of January, for example billing by consumption or limits on who may use how much.

Due 2027-01-31 We expect Microsoft, OpenAI or Anthropic to announce a product for companies by the end of January that automatically hands tasks to a cheaper or a stronger AI model.

Due 2027-01-31 We expect Microsoft, Apple or Google to extend their operating system by the end of January so that any app can use an AI model directly on the device.