Artificial intelligence stepped into statecraft this week. The first G7 summit to seat all three major AI laboratory chief executives alongside heads of state opens in two days. Against that backdrop, three further developments bear on strategy, compliance, and competitive positioning.
G7 Évian: lab CEOs join heads of state for the first time
The 52nd G7 summit opens on 15 June at Évian-les-Bains under France's presidency. Sam Altman (OpenAI), Demis Hassabis (Google DeepMind), and Dario Amodei (Anthropic) are all confirmed — the first time the heads of all three major AI laboratories will be present at a G7, alongside executives from Mistral AI, Meta, Cohere, Sakana AI, Black Forest Labs, Synthesia, and Salesforce.
France has made AI a top priority of its presidency. The Wednesday working session brings political leaders and technology executives together to discuss AI infrastructure, frontier risk in cyber and biological domains, and youth safety online. The expected output is voluntary commitments, not binding rules — with youth safety at the top of Altman's personal agenda.
The harder signal is the rift running beneath the summit. The EU, the US, and Canada each arrived with competing national AI strategies designed to anchor domestic economic advantage. Tensions over AI sovereignty are simmering across the G7 membership. What language emerges from the communiqué will signal whether those strategies can coexist or whether AI governance is beginning to fracture along the same lines as trade. For any operator whose AI stack crosses borders, the direction of that signal matters considerably more than the communiqué's precise wording.
EU Digital Omnibus: the high-risk compliance clock resets by 16 months
On 7 May, EU institutions reached a provisional agreement on the Digital Omnibus that materially changes the AI Act compliance calendar for high-risk systems:
- Annex III systems (stand-alone high-risk use cases: HR screening, credit scoring, biometric categorisation, critical infrastructure, education access, law enforcement) move from 2 August 2026 to 2 December 2027 — a 16-month extension.
- Annex I systems (AI embedded in regulated products: medical devices, machinery, automotive, aviation) move to 2 August 2028.
The rationale is structural, not political. National competent authorities have not been designated on schedule, and the harmonised standards operators would use to demonstrate conformity remain incomplete. The deferral reflects the regulators' own readiness position, not a softening of intent.
What has not moved: the Article 50 transparency obligation for newly deployed AI systems remains active on 2 August 2026. Any system that makes AI-generated content visible to users — chatbots, synthetic media tools, recommendation interfaces with disclosed AI — must label it. Formal adoption of the Omnibus is expected before that date. Operators should treat the high-risk deferral as confirmed and run their Article 50 readiness assessment now, not after the summer.
Grok V9-Medium: xAI bets on distribution over parameters
xAI completed training Grok V9-Medium in early June at 1.5 trillion parameters — roughly three times the size of its current production model. The training dataset is consequential: rather than relying on public GitHub repositories, xAI used real Cursor developer workflow data, targeting Claude's software engineering benchmark lead directly.
The model is already rolling into Tesla's connected-vehicle fleet and X via over-the-air update, ahead of a full public API release expected mid-June. xAI also shipped Grok Voice and a Grok Imagine 1.5 image-generation preview to round out the platform before the full rollout.
The distribution point is the one that changes the competitive framing. Tesla's connected fleet and X's user base give xAI embedded reach into daily hardware and social context that no other AI laboratory can replicate organically. If V9-Medium performs at its benchmark targets, the question for competing providers is not whether the model is better. It is whether a user already inside that ecosystem will bother to leave. For the majority, the answer will be no.
NAVER and NVIDIA lay the sovereign AI factory template
On 7 June, NAVER and NVIDIA announced that NAVER will build AI factories at gigawatt scale using the NVIDIA DSX platform. The build begins at 55 megawatts at the GAK Sejong hyperscale facility in South Korea, scaling to 100 megawatts internationally within 2027 and 200 megawatts by 2028.
The scope extends well beyond raw compute. NAVER will use the full NVIDIA stack to develop next-generation HyperCLOVA X models, a Seoul World Model for Korean-language and regional contexts, and agentic AI services targeting enterprise and government clients. An AI Agent Platform launch is targeted for the second half of 2026.
The architecture is becoming a repeatable template: a national internet champion with an existing data moat, NVIDIA's DSX as the build-and-operate layer, and government alignment as the operating licence at scale. Operators sourcing AI capacity in markets with data-residency mandates or sovereignty requirements should treat this arrangement as the reference model, not a regional outlier.
The thread through all four items is the same: decisions about AI are migrating into rooms where the stakes are national rather than commercial. The G7 session at Évian will likely produce only voluntary language. But the people writing that language are now in the room. Operators who have been treating AI governance as a legal overhead item should consider upgrading it to a planning assumption.