The headline number today is $11.6 billion — the size of Anthropic's new cloud deal with Akamai over seven years. That's a staggering infrastructure commitment for a company that has positioned itself as the thoughtful, safety-first alternative to OpenAI, and it signals that the race for frontier AI isn't slowing down regardless of what the rhetoric around governance says. Meanwhile, the Pentagon's supply-chain designation of Anthropic — upheld by an appeals court — adds a genuinely unusual wrinkle: the company is simultaneously signing nine-figure cloud contracts and being flagged as a national security concern by the same government it's increasingly doing business with.

Crusoe's quiet abandonment of its $1.25 billion plan to power AI data centers with Boom Supersonic turbines rounds out the infrastructure picture. The turbine scheme was always a creative workaround for the energy bottleneck — and its collapse suggests that novel power solutions are dying faster than the AI compute demand they were meant to serve.

Meta used Connect 2026 to put Muse everywhere — smart glasses, hardware accessories, a new Muse Image and Muse Video generative platform — and the coverage this week reflects exactly what happens when a company ships faster than its security team can keep up. The AI assistant that Meta is betting its hardware future on launched at Connect and is already generating controversy on multiple fronts: Amazon has cut off its shopping data access in a platform dispute that will define the agentic economy's access politics, and separate reporting found unsecured OpenAI agents (not Meta's) posted user images online — a timely reminder that agent deployment is still running ahead of containment. Wired's design critique — that an adults-only AI biometric keychain looks like a children's toy — is the kind of detail that tends to stick.

Two concrete regulatory developments landed today, which is two more than most days produce. An appeals court has upheld the Pentagon's ability to designate Anthropic a supply-chain risk — an opaque national security designation that the company has been fighting and which now has legal standing behind it. And Stanford HAI's new analysis documents that data brokers subject to California's strict privacy laws are systematically non-compliant, a finding that matters for AI training pipelines that depend on purchased data and for any enterprise whose AI governance posture assumes third-party data vendors are playing by the rules they signed up for.