The half-year statement. About ten minutes. Watch it or read the summary below.

The shift

For three years the AI story was acceleration. ChatGPT, Claude, Gemini, Nvidia at the center of the market, hyperscalers spending at a scale the world had never seen. AI was the gas pedal.

In the past few weeks the story changed. Washington put a speed bump in front of the frontier: OpenAI delayed the broad rollout of its most advanced model at government request; Anthropic's most advanced models were restricted, then brought back this week under tighter guardrails. That is an earthquake. The story is no longer speed. The story is control.

The question every board is now asking

The old question was "which model is best?" The new question is:

What happens if I cannot access it tomorrow?

The stranger twist

OpenAI has reportedly proposed giving the US government a 5% stake in the company and suggested other leading labs do the same, through a public fund. Not regulation. Not procurement. Not defense contracts. Equity. The state moving from referee to shareholder.

One reading is public dividend: if AI creates trillions of dollars of value, citizens should share the gains. Another reading is political insurance: if government can slow your release or shape your IPO path, a stake starts to look less like generosity and more like a seatbelt. Either way, the question is unavoidable for the other 95% of the world. Do you want your critical AI infrastructure tied to a company where one government may own the upside, shape access, or influence release?

Not just an American story

China is treating AI like strategic infrastructure. Reports say China has restricted overseas travel for top AI talent at companies like Alibaba and DeepSeek. Chinese courts have ruled companies cannot simply fire workers because AI can replace them. Different systems. Same signal. AI is no longer just software. It is power.

And Chinese models are closing the gap: DeepSeek, Qwen, Kimi, GLM. Powerful, cheaper, easier to deploy. If the US can restrict frontier models for security reasons, what happens when Chinese models become equally powerful? Will Europe restrict them? Will companies restrict them internally? Will boards allow employees to use them on sensitive work? Who decides?

As INSEAD Professor Theos Evgeniou recently asked at Station F in Paris: which AI do you want your kids to learn history from? AI is not only writing code. It is answering questions, shaping memory, explaining politics, summarizing history, and telling children, employees, voters, and boards what happened and what matters.

The circular economy underneath

The labs need compute. Hyperscalers need customers. Nvidia sells the chips. Data centers need power. Public markets need the AI story to keep working. Companies need the tools to keep functioning. Everyone depends on everyone else. That interdependence is now a governance surface.

What I would tell a board this week

Do not ask only "are we using AI." Ask:

  • What are we dependent on?
  • Which models are mission-critical?
  • Which vendors could cut us off?
  • Where do we need a second model?
  • Where do we need an open-weight or sovereign alternative?
  • Where is the kill switch?
  • Who owns the decision if access changes overnight?

The risk is no longer only that AI gives the wrong answer. The risk is that the AI you built around is no longer available. That is the question for the second half of the year. Not "how fast can we go?" but "what is our plan B if the road closes?"

Go deeper