streugut.ch
Archive

The direction

Where the AI race is heading

AI competition is moving away from frontier models alone toward control of the stack that turns models into governed execution. That shift is being carried by agent environments tied to enterprise identity, data, workflow, and security, and by model access that is increasingly routed through managed clouds, compatibility layers, owned product surfaces, and now access controls. The infrastructure race underneath is widening into financing, datacenter capacity, and energy planning. What is particular now is that the same operating layer is being fixed at institution scale while eligibility to use advanced models is starting to narrow along regulatory and geopolitical lines.

The connection

How the moves fit together

Agent control planes and industrial AI build-outs

The biggest practical change is that the agent race is looking less like a battle of chatbots and more like a battle over the operating environment around them. This week, Microsoft moved Microsoft IQ into general availability and added closed-loop learning to agents connected to the Power Apps MCP server, while OpenAI said it will acquire Ona to bring customer-controlled cloud execution into Codex. Read together with Google Cloud, AWS, Snowflake, and Databricks adding their own context and control layers, the direction is clear: vendors are competing to become the place where agents run with the right identity, memory, business data, and policy checks. The non-obvious part is that this is not only a product design choice. It is also a way to make agent use governable enough for real work inside organizations.

A separate movement sits in infrastructure, and it is becoming more industrial and more financial at the same time. Google agreed to pay SpaceX about $920 million per month for compute, Apollo and Blackstone finalized a $35 billion financing package for Anthropic’s chip purchases and capacity expansion, and China announced a $295 billion plan for a national AI compute backbone. Those are not routine supplier deals. They show that competitive position now depends on lining up capital, power, sites, packaging, networking, and anchor customers as a single system. Crusoe’s Wyoming pause matters here too: large ambition is not enough if customers, costs, and delivery do not line up.

Distribution and regulation are also tightening together rather than pulling in opposite directions. Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9, AWS and Google Cloud made Fable 5 generally available, Microsoft added it to Foundry, and then Anthropic disabled access to Fable 5 and Mythos 5 for foreign nationals on June 13 after a US export-control directive. In parallel, Europe kept linking AI rules to real market conditions through measures on cloud capacity, energy planning, transparency, and launch constraints. The near-term meaning is not that model quality no longer matters. It is that where a model is placed, who can reach it, and under what controls are becoming part of the competition itself.