Geopolitics & Sovereign Positioning

Zhipu AI's confirmation that GLM-5.3-Flash was trained on a 100,000-chip cluster of domestically produced processors is the most strategically significant chip-related disclosure from China in months. The model reportedly processed 62 trillion tokens in stealth trials before formal release, achieving performance competitive enough to top OpenRouter's rankings. This is not a proof-of-concept — it is a production deployment at scale on indigenous silicon. South China Morning Post reports the announcement drove a surge in Zhipu's share price, reflecting market confidence that domestic chip dependency is no longer a binding constraint for at least some tiers of frontier AI work.

This development intersects directly with the broader US policy debate flagged by Foreign Policy, which argues that Washington's export control strategy was predicated on assumptions about the hardware dependency of frontier AI that the open-weights ecosystem — and now domestic Chinese silicon — is actively undermining. Separately, China's three leading chip packaging firms — JCET, Tongfu, and Huatian — have launched multibillion-dollar capacity expansions, signalling that Beijing's push for full-stack semiconductor self-sufficiency is accelerating through the advanced packaging layer that US controls have not effectively targeted. South China Morning Post notes these investments are driven by both AI demand and explicit Beijing policy directives on tech self-sufficiency.

Alibaba's launch of two data centres in Brazil — its first in South America — is a direct play for infrastructure dependency in the world's seventh-largest economy and Latin America's dominant digital market. The centres offer cloud infrastructure and agentic AI services to local enterprises, positioning Alibaba Cloud as an alternative to AWS, Google Cloud, and Microsoft Azure in a market where US hyperscalers have established but not uncontested footholds. South China Morning Post frames this explicitly as part of a broader Chinese tech expansion into emerging markets amid intensifying US-China rivalry.

The strategic logic is straightforward: whichever cloud and AI platform hosts a country's enterprise data and runs its AI workloads creates long-term switching costs and, critically, potential intelligence access. Brazil's government has been navigating between US and Chinese technology relationships for years — its data localisation rules create both a barrier and an incentive for in-country infrastructure investment. Alibaba's move forces Brazilian policymakers to make more explicit choices about which foreign AI infrastructure they are willing to host, and on what terms. This dynamic is examined more broadly in Rest of World's analysis of how Europe, India, Brazil, and China are each attempting — with limited success — to break free from Big Tech dependency, with each country's approach reflecting distinct sovereignty concerns.

The July 6 Zaporizhzhia incident — in which a Russian drone apparently guided entirely by onboard AI, with no human in the loop for the terminal strike, killed three civilians — represents a potential first in the operational use of fully autonomous lethal force. Lawfare's Rational Security flagged this as a threshold event that experts believe may constitute the first confirmed autonomous AI kill in a live conflict. The qualification 'appears to have' reflects ongoing attribution uncertainty, but the expert consensus cited suggests high confidence.

The strategic implications extend well beyond Ukraine. If autonomous terminal engagement has been operationalised by Russia in a live theatre, it compresses the timeline for other actors — including China, the US, and regional middle powers — to make explicit decisions about their own autonomous weapons postures. The Atlantic Council's concurrent report on AI and autonomy in contested operations addresses the US dimension directly, noting that American forces must expect AI and autonomous systems to operate in denied, degraded, intermittent, and limited (DDIL) conditions — jamming, cyber-attacks, and anti-satellite capabilities that sever connectivity and force greater onboard autonomy. This creates a structural pressure toward removing humans from the loop that is difficult to resist through policy alone once adversaries have already crossed the threshold.

A CSET analysis published via the Australian Strategic Policy Institute argues that NATO's Indo-Pacific Four partners — Australia, Japan, South Korea, and New Zealand — should adopt NATO's model for integrating AI-enabled decision-support systems into coalition military operations. CSET's analysis is notable because it is explicitly a coalition interoperability argument, not merely a national capability argument: the value of AI decision-support in the Indo-Pacific context depends on whether allied systems can share data, communicate in DDIL environments, and maintain operational coherence under adversary electronic and cyber disruption.

The IP4 framing is significant geopolitically. These four countries span a geography from the South Pacific to Northeast Asia and collectively represent the core US-aligned technology partners in the Indo-Pacific. Their AI military alignment — if it proceeds — would represent a meaningful extension of the NATO AI governance and interoperability model into a theatre where China is the principal pacing threat. However, this remains a recommendation and a diplomatic aspiration, not an enacted framework. South Korea and Japan maintain their own AI defence procurement cycles, and New Zealand's threat perception differs materially from Japan's. The gap between analytical recommendation and binding coalition commitment is substantial.