The AI frontier tilted geopolitical this week. Xi Jinping used the World AI Conference in Shanghai to stand up a 29-nation cooperation bloc that deliberately leaves the West out, while two of the most-watched labs made open weights the currency of the moment. Underneath the diplomacy, money kept pooling in the unglamorous inference layer that actually runs the models — and two human stories, a factory strike and a researcher's resignation, served notice that automation and defense-AI timelines will meet resistance from the people they touch.
Xi builds an AI bloc without the West
In his first in-person appearance at the World AI Conference since 2018, Xi Jinping formally launched the World AI Cooperation Organization, a body of 29 founding members — Russia, Pakistan, Brazil, Indonesia and South Africa among them — headquartered in Shanghai and pointedly excluding the United States, the EU, the UK, Japan and Korea. It is the clearest institutional answer yet to US chip-export controls and the containment strategy some call "Pax Silica." The message is structural: global AI is bifurcating into two governance-and-supply spheres, and China is positioning itself as the default AI partner for the Global South, pitching cooperation against Washington's restrictions. Standards, data governance and compute alliances now have a real chance of fracturing along bloc lines.
Mira Murati finally shows her hand
After more than fourteen months of near-total silence and a $12 billion valuation, Mira Murati's Thinking Machines Lab has released its first model, Inkling — an open-weight, 975-billion-parameter mixture-of-experts system with roughly 41 billion active parameters, natively multimodal, trained on about 45 trillion tokens, with the weights posted openly on Hugging Face. What is striking is the framing: the lab says plainly that Inkling is not meant to be the strongest model. The whole bet is on customization — its Tinker platform for adapting the model to specific tasks — over the leaderboard chase. It is a wager that the value in AI is migrating from raw frontier capability toward fit-for-purpose deployment, and a direct challenge to the idea that one giant closed model should do everything.
The money is in running the models, not just building them
The week's largest AI round went not to a model lab but to the layer that serves models at scale. Fireworks AI, an inference "neocloud," raised $1.5 billion in a Series D at a $17.5 billion valuation, backed by Nvidia, Index Ventures, TCV and Atreides, on more than $1 billion in annual recurring revenue and roughly 40 trillion tokens served every day. The number is a marker of where AI-infrastructure margin is settling: the business of running inference at high throughput has become a mega-cap venture category in its own right, sitting between raw GPUs and the applications on top — and Nvidia is quietly funding the companies that make its chips useful.
The first strike against the robots
In Korea, more than 34,000 Hyundai workers launched a rolling strike with an unusually specific goal: to stop Boston Dynamics' Atlas humanoids from being deployed on the production line. It is being described as the first humanoid-robot strike in history, the first time organized labor has walked out explicitly to block humanoid automation rather than to protest pay or hours. The automation debate, long treated as an engineering timeline, just became an industrial-relations one. A powerful union is drawing its line before the robots arrive — and in doing so it reframes humanoid deployment as a political problem that the aggressive rollout schedules priced into Boston Dynamics, Tesla and others may not fully account for.
A safety researcher walks out over "killer robots"
DeepMind safety researcher Alex Turner has resigned, and explained why in a public essay alleging that Google sold AI to the military with no ban on lethal autonomous weapons or mass surveillance. He says a proposal backed by roughly 250 employees, which he escalated directly to CEO Demis Hassabis, went nowhere. The specifics will be contested, but the shape of the story matters on its own: one of the field's own alignment researchers is publicly accusing a frontier lab of walking back its stated ethics commitments as it pursues defense contracts. It crystallizes a fault line the AI boom has been widening all year — between the defense-AI gold rush and the safety community that built the underlying systems.
The money view
Read the momentum and it runs geopolitical first. WAICO formalizes a non-Western AI bloc just as open weights — Inkling this week, Kimi K3 last — become the standard export of that sphere, while capital keeps flowing to the inference layer that turns models into products, with Fireworks' $17.5 billion mark as the tell. But the two quietest stories may be the most instructive: a humanoid-robot strike and a defense-AI resignation both say that automation and military-AI timelines face human friction the market has not fully absorbed. What to watch: which countries formally join WAICO and what standards it sets, whether Thinking Machines' customization-first approach gains developer traction, and how the Hyundai strike resolves — because it may set the precedent for every humanoid rollout that follows. Signal, not advice.