Geopolitics & Sovereign Positioning

Following eight hours of talks at JPMorgan Chase's New York headquarters on Sunday, US Treasury Secretary Scott Bessent and USTR Jamieson Greer confirmed with Chinese Vice-Premier He Lifeng an agreement to establish a formal AI dialogue, including a proposed threat-notification hotline. The talks were explicitly framed as preparatory groundwork for a Trump-Xi summit later this week. Bessent described the engagement as 'very successful'; Chinese negotiator Li Chenggang offered the more guarded assessment of 'not bad.' This asymmetry in framing is itself a signal: Beijing is treating the dialogue as a trade-adjacent confidence measure, not a strategic concession on technology access. The agreed channel is a diplomatic commitment without binding enforcement mechanisms — there is no text yet governing scope, escalation thresholds, or verification.

Analysts cited by the South China Morning Post and South China Morning Post describe the hotline as a pragmatic floor for crisis prevention rather than a resolution of structural competition. The deep-seated disputes — chip export controls, model distillation restrictions, and market access asymmetry — remain entirely unresolved and are not on the declared agenda for the dialogue. A Bank of America analyst noted ahead of the summit that neither country can win the AI race without third-party dependencies, which creates a rational basis for limited cooperation even amid systemic rivalry. The question is whether the dialogue channel survives the next round of US chip control tightening, which is the more likely test of its durability than anything agreed in New York.

At its annual Apsara Conference in Hangzhou, Alibaba unveiled what it called China's most powerful domestic AI chip and announced plans to train a model with up to 10 trillion parameters — an order of magnitude beyond current frontier public benchmarks. Chairman Joe Tsai explicitly framed this as part of an end-to-end, full-stack capability build, encompassing chips, infrastructure, models, and applications, with a stated ambition toward artificial superintelligence. The chip announcement is particularly significant: it represents a direct response to US export controls on Nvidia H100 and A100 series hardware, and confirms that the controls have functioned more as an accelerant for domestic semiconductor development than as a ceiling on Chinese AI ambition. Alibaba's chip remains almost certainly behind Nvidia's current generation in raw performance, but the trajectory matters more than the snapshot. As reported by South China Morning Post, the company is reaffirming investment in capabilities that were explicitly targeted for denial by US policy.

The 10-trillion-parameter model announcement should be assessed cautiously — parameter count is not a reliable proxy for capability, and teasing a training target is not the same as completing it. But the strategic signal is deliberate: Alibaba is communicating to domestic regulators, international partners, and capital markets that China's largest tech company is competing at the frontier on its own terms. Combined with Moonshot AI's Kimi K3 model now available on Amazon Web Services' Bedrock platform — as reported by South China Morning Post — Chinese AI firms are simultaneously building domestic infrastructure and penetrating Western distribution channels, a dual-track strategy that complicates US containment logic.

A Foreign Policy analysis published ahead of UNGA makes the case that China's AI gains in the Global South are not primarily about model superiority or price — they are about an explicit strategy of openness. Beijing is winning influence by enabling other countries and developers to build on Chinese AI tools rather than locking them into proprietary platforms with restrictive licensing, data-sharing requirements, or political conditionality visible in the terms of service. This is a structurally different proposition from what US hyperscalers offer: dependency embedded in API contracts versus capability transfer that survives the bilateral relationship. The article identifies this as the 'real reason' for Chinese traction — not dumping or state subsidy, but a permissive ecosystem model that allows local adaptation.

The strategic implication for Washington and its allies is significant. The US approach to the Global South has largely been framed around AI governance standards, responsible use frameworks, and access to American cloud infrastructure — all of which implicitly require alignment with US regulatory and foreign policy preferences. China's open-tool model imposes no such alignment cost, making it more attractive to governments that want AI capability without geopolitical entanglement. This dynamic is already visible in UNGA discussions, where, per Foreign Policy, AI governance fragmentation and 'runaway AI fears' are on the agenda without any emerging consensus on a multilateral framework. Countries that build on Chinese AI tools today are setting technical and regulatory path dependencies that will be difficult to reverse.

A Rest of World analysis frames the UAE's AI position with unusual precision: the Emirates owns the compute — billions invested in Nvidia GPU clusters — but rents the intelligence, dependent on proprietary models controlled by US companies. Nvidia's forthcoming open-source model release is being interpreted in Abu Dhabi as a potential inflection point. If the UAE can run frontier-quality open models on its own hardware without ongoing API dependency or licensing exposure to US policy decisions, its sovereign AI posture changes materially. This is not about capability parity with the US or China — it is about the UAE's ability to offer AI infrastructure services to third parties, including governments that cannot or will not use American cloud platforms, without those services being a vector for US leverage.

The UAE's positioning is a microcosm of a broader dynamic among middle powers with capital and infrastructure but limited domestic model development capacity. The terms under which they access frontier AI — who controls the model, under what licensing conditions, with what data visibility — are now explicit foreign policy variables. Abu Dhabi's decision to invest heavily in Nvidia hardware was itself a strategic bet on US alignment; Nvidia's open-source move, if it materialises at sufficient capability levels, partially decouples that hardware investment from US model dependency. This matters for how the UAE positions itself as an AI hub for the broader region, including for customers in the Gulf, Africa, and South Asia who need a non-US, non-Chinese option.