Geopolitics & Sovereign Positioning

DeepSeek is closing a pre-IPO funding round valuing it at approximately 500 billion yuan (~$74 billion) before investment, with a ~50 billion yuan raise expected to complete before end of August and a 2027 listing on Shanghai's Star Market targeted, according to sources cited by South China Morning Post. Existing investors are participating in the round, signalling continuity of backing rather than a distressed recapitalisation.

The strategic significance here is not the valuation — it is the trajectory. A Star Market listing would embed DeepSeek within China's domestic capital architecture, making it subject to the same regulatory oversight and state influence that governs other listed technology companies. This reduces the company's operational independence while simultaneously placing it beyond the reach of US secondary sanctions or investor-pressure mechanisms. Simultaneously, DeepSeek's open-weight models are achieving record penetration on US developer infrastructure: open-weight models now account for 54% of token volume on Vercel's AI Gateway, with Chinese models — primarily DeepSeek — driving that share, per South China Morning Post. This is an asymmetric dynamic: Chinese models distributed freely are capturing developer mindshare and dependency within the US ecosystem, while DeepSeek itself moves into a protected domestic institutional structure.

Goldman Sachs projects that China's supply of wafers produced at 7nm and below will grow at a 46% compound annual rate between 2025 and 2035, versus 17% for the rest of the world, per South China Morning Post. The analysis identifies lithography — specifically the absence of domestic EUV capability — as the remaining structural constraint, but frames it as a bottleneck on pace, not a ceiling on eventual capability.

This assessment is consistent with the capital mobilisation visible across the Chinese semiconductor ecosystem. CCSH Corporation, parent of YMTC — China's leading NAND flash manufacturer — is preparing what could be a record-setting IPO on Shanghai's Star Market, seeking to list 10-12% of enlarged share capital, per South China Morning Post. Separately, Xiaomi has unveiled a 3nm AI processor — the Xring O3 — and a 6nm automotive chip, pursuing vertical integration of silicon design for AI workloads on devices and vehicles, per South China Morning Post. These are not isolated corporate decisions — they reflect a system-level response to US controls that is producing distributed chip design and manufacturing capacity across the Chinese tech industry. The second-order consequence of export controls has been exactly what critics warned: accelerated domestic investment, state-backed capital mobilisation, and a faster timeline to partial independence than would have existed without the external pressure.

US visa restrictions are structurally deterring international AI researchers from building and remaining at American institutions, with the effect of redirecting talent toward other jurisdictions including China, per Rest of World. This is not a new concern, but it is sharpening: the article profiles cases where researchers with direct frontier AI relevance are either returning to China or routing their careers through third countries rather than navigating US immigration constraints.

The irony of the policy configuration is acute. The US has simultaneously tightened hardware export controls on the premise that China must not access frontier AI capability, while implementing immigration policies that push the humans who build frontier AI capability out of the American system. The $70 million Bay Area mansion purchased by Tony Wu Yuhuai — China-born co-founder of xAI — reported by South China Morning Post — illustrates the other face of this dynamic: elite Chinese AI talent that has remained in the US commands enormous economic power and sits at the centre of America's most strategically sensitive AI labs. The policy tension between retaining such talent and the broader restrictive posture toward China-linked individuals has not been resolved and is generating contradictory signals.

Two concurrent developments illustrate AI's transition from strategic aspiration to operational military reality. In Ukraine, autonomous systems equipped with on-board AI are conducting strikes deep into Russian-controlled territory where jamming renders remote piloting non-viable, per War on the Rocks. This is confirmed operational use of AI-enabled autonomous lethal systems in a major interstate conflict — not a test, not a demonstration. The piece frames this as a democratisation of military capability, but the geopolitical read is more specific: smaller states can now sustain asymmetric offensive operations against great powers using AI autonomy to circumvent electronic warfare advantages.

In parallel, the UAE has deployed homegrown AI cyber-defence systems to protect banks, aviation, and energy infrastructure against attacks that have escalated since the onset of conflict with Iran, per Rest of World. The UAE case is strategically notable because it reflects a deliberate sovereign AI industrial policy — building domestic capability rather than purchasing foreign solutions — in the context of active threat. Shield AI's Brandon Tseng, in a War on the Rocks interview, articulates the doctrine that the AI pilot matters more than the airframe, a principle that has direct implications for how the Pentagon and its partners are thinking about force structure. Taken together, these cases indicate AI autonomy has crossed from developmental to operational status in at least two distinct conflict environments.