Geopolitics & Sovereign Positioning
Top Line
Trump has publicly rejected AI slowdown calls on the grounds that any deceleration cedes ground to China, a position that structurally forecloses US participation in the multilateral safety agreements being proposed ahead of the September 24 Trump-Xi summit.
Chinese AI firm Z.ai is pursuing a $5 billion combined equity and convertible-bond raise, signalling that Chinese frontier AI developers are capitalising aggressively despite US export controls — and that Hong Kong remains a viable capital-raising venue for the sector.
A DeepSeek engineer's public attack on Anthropic and OpenAI's 'pacing' proposals — framed explicitly as a US power-consolidation play — reveals a hardening Chinese domestic consensus that AI safety framing is a geopolitical instrument rather than a genuine risk-management framework.
China is outpacing the US on consumer AI adoption rates according to Morgan Stanley data, with 80 percent weekly usage versus 54 percent in the US, driven by super-app integration — a structural distribution advantage independent of frontier model capability gaps.
US Cyber Command has installed a new top AI official drawn from the National Geospatial-Intelligence Agency, marking a quiet but concrete step in institutionalising AI within combatant command operations.
Key Developments
Trump's China Framing Closes the Door on US Participation in AI Safety Governance
President Trump has explicitly dismissed AI safety warnings as interference from 'negative forces' and framed any regulatory constraint as an unacceptable concession to China. This is not a peripheral rhetorical position — it is the operating doctrine of the US executive branch going into the September 24 Trump-Xi meeting. As BBC and SCMP both report, Trump acknowledged some regulation may be necessary but offered zero specifics, suggesting the administration has no near-term regulatory architecture in preparation.
The consequence for international governance is severe. Multiple Atlantic Council analyses and a Foreign Policy assessment of the summit agenda (Foreign Policy) conclude that Trump and Xi hold fundamentally incompatible views of AI risk and regulation, making any binding bilateral safety agreement near-impossible. The Economist frames this as a two-level failure: American labs and US authorities are at odds domestically, while the US and China are at odds internationally — producing a governance vacuum at precisely the moment frontier capabilities are escalating. Atlantic Council commentary notes that while both leaders should not be counted on to slow AI, they should not be counted out entirely, but this is diplomatic optimism rather than a structural assessment of where incentives lie.
China's AI Capital Formation Continues Despite Export Controls — Z.ai's $5B Raise
Zhipu AI (Z.ai), one of China's leading frontier model developers, is executing a roughly $5 billion capital raise combining a Hong Kong H-share placement and a 20.14 billion yuan convertible bond, as SCMP reports. The timing — a second major raise after a July share sale — reflects an aggressive scaling posture. The company's stock is down 73 percent from its intraday high, indicating the market is pricing significant execution risk, but institutional demand for the placement suggests continued investor appetite for Chinese frontier AI exposure.
The Hong Kong listing route is strategically significant: it allows Chinese AI firms to access international capital without triggering US foreign investment screening mechanisms that apply to US-listed entities. This is an enacted workaround, not a proposal. Combined with Chinese researchers publishing a 100-fold improvement in ferroelectric memory endurance — relevant to reducing dependence on DRAM architectures where US export controls bite hardest (SCMP) — the picture is of a Chinese AI sector using every available non-US-controlled lever to close capability and infrastructure gaps.
Chinese AI Industry Rejects 'Pacing' as US Power Play — DeepSeek Engineer's Signal
A DeepSeek engineer's public broadside against Anthropic and OpenAI's calls to slow AI development — comparing the proposed concentration of AI resources in US proprietary labs to fascist-era centralization — is analytically significant not for its rhetorical excess but for what it reveals about Chinese industry's reading of US safety discourse. As SCMP reports, the comments were timed explicitly ahead of the Trump-Xi summit, suggesting coordinated messaging rather than individual opinion.
This dynamic has a structural implication for AI safety diplomacy: any US proposal to jointly govern AI development speed or capability thresholds will be received in Beijing — and publicly framed by Chinese industry — as an attempt to lock in US advantage by halting Chinese catch-up. Foreign Policy's analysis of Chinese firms potentially exposing data to US access (Foreign Policy) adds a further layer — suggesting AI competition creates mutual intelligence vulnerabilities that neither side has fully mapped. The combination of Chinese industry hostility to pacing proposals and potential data exposure risks creates a negotiating environment where trust is structurally absent.
US Cyber Command Institutionalises AI Through NGA Appointment
Ronzelle Green, an official from the National Geospatial-Intelligence Agency, has assumed the top AI role at US Cyber Command, succeeding Reid Novotny who drove the command's initial AI adoption push, according to Defense One. The NGA background is notable: geospatial intelligence has been among the earliest military domains to operationalise AI for pattern recognition and targeting support, and moving that expertise into Cyber Command suggests an intent to apply similar automation to offensive and defensive cyber operations.
Signals & Trends
Consumer AI Adoption as a Strategic Metric — China's Super-App Advantage Is Compounding
Morgan Stanley's finding that 80 percent of Chinese consumers use AI weekly versus 54 percent in the US is not merely a market share statistic — it is a signal about where training data, user feedback loops, and product iteration cycles are concentrating. China's super-app architecture (Tencent's WeChat, Alibaba's ecosystem) embeds AI into daily transactions, social interaction, and commerce in ways that Western app-store fragmentation does not replicate. This creates a structural data advantage that compounds over time independently of frontier model capability rankings. Senior strategists should track whether this adoption gap translates into Chinese AI systems outperforming US equivalents on tasks requiring dense real-world behavioural data — consumer preference, social dynamics, logistics optimisation — even if US models retain leads on benchmark reasoning tasks.
The 'Safety as Geopolitical Weapon' Narrative Is Hardening in China — and Will Outlast This Summit
The convergence of a DeepSeek engineer's public attack, Chinese industry resistance to pacing proposals, and Foreign Policy's analysis of mutual data vulnerabilities points to an emerging Chinese strategic consensus: that US-led AI safety frameworks are instruments of technological containment rather than genuine risk management. This narrative, if it solidifies into official Chinese policy, does not merely complicate the September 24 summit — it forecloses the possibility of good-faith multilateral AI governance for the medium term. The structural problem is that the narrative has empirical grounding that makes it hard to counter: US safety advocates are frequently the same firms seeking regulatory capture, and US export controls are explicitly designed to maintain competitive advantage. Foreign policy professionals should anticipate that China will use this framing at multilateral venues — the UN, ITU, and regional AI governance forums — to position itself as the defender of open AI development against Western monopolisation.
China's Semiconductor Self-Sufficiency Push Is Moving from Policy to Demonstrated Results
The ferroelectric memory endurance breakthrough — 10 billion write cycles, 100 times prior benchmarks — is a peer-reviewed research result, not a commercial product announcement, and should be assessed accordingly. But the pattern it represents matters: Chinese semiconductor research is producing measurable advances specifically in the areas where US export controls are designed to create bottlenecks, namely high-performance memory for AI training and inference. Whether any individual advance reaches mass production is secondary to the directional signal that Chinese research institutions are systematically targeting control points in the AI hardware stack. Combined with continued capital flows to Chinese AI firms through Hong Kong markets, the medium-term trajectory is toward reduced Chinese dependency on US-adjacent semiconductor supply chains — which will erode the leverage that export controls currently provide.
Explore Other Categories
Read detailed analysis in other strategic domains