Squared: The Week AI Grew Up (and Got Political)

Per the FT (via The Verge and Ars Technica), Sam Altman is in active talks to give the US government a roughly 5% ownership stake in OpenAI to ease tensions with the Trump administration and blunt public backlash — lower than Senator Sanders' target.

When the largest AI provider starts negotiating equity with the state, your vendor's roadmap becomes subject to political weather. If you're building on OpenAI, factor regulatory and geopolitical exposure into your dependency risk — this is no longer a purely commercial relationship. Revisit your multi-provider stance now, not after a policy shock.

In an internal Meta meeting reported by TechCrunch, Zuckerberg told staff that AI development efforts — specifically agents — were not moving as quickly as anticipated.

This is the most useful sentence of the week. The company spending most on this is quietly resetting expectations. If your 2026 plan assumes fully autonomous agents replacing workflows this year, pressure-test it. Agents are real and useful in narrow, well-scoped tasks — but treat 'autonomous end-to-end' claims as aspiration, not delivery.

The US Department of Commerce lifted export controls on Claude Fable 5 and Mythos 5 after weeks of negotiation; Anthropic is restoring global access. Anthropic is also reportedly discussing a custom chip with Samsung — a week after OpenAI's Broadcom chip deal.

Two signals: frontier model availability can now be gated by government, and both major labs are chasing their own silicon to escape supply constraints. For buyers, that means capability access may vary by region — and the providers are betting big on owning the hardware stack. Don't assume uniform, uninterrupted access when you architect for scale.

Ars Technica reports Google's AI buildout drove a 37% increase in its electricity use in 2025, with the company scrambling to balance data-centre emissions against clean-energy commitments.

Compute isn't free and it isn't clean. If your organisation has sustainability commitments, AI adoption now has a measurable carbon and cost line that will surface in ESG reporting. And with NVIDIA openly courting capital partners to fund 'AI factories', the infrastructure economics are getting priced into the tools you buy. Ask your vendors about efficiency, not just capability.

OpenAI's new Signals data shows ChatGPT usage deepening across regions, languages and capabilities. Meanwhile TechCrunch noted a sandwich chain, Jersey Mike's, name-dropping AI in its IPO filing — a marker of how reflexive the hype has become.

Real usage is compounding, which validates investment — but 'AI' in a strategy deck now means nothing on its own. The discipline that matters is specificity: which task, what measurable outcome, what does it replace. If a plan can't answer those, it's theatre. Reward teams that ship narrow wins over those that promise transformation.