China’s AI race heats up: Goldman warns on cost wars, Hong Kong IPOs loom, and Wall Street funds Anthropic’s data centers

Goldman Sachs’ Ronald Keung, Head of Asia Internet Research, argues that competition among Chinese AI developers will intensify around the “best performance-to-cost balance,” implying a faster commoditization cycle for model quality. In parallel, Rhodium Group analysis highlighted that China leads the US and other G7 countries in several key AI “tech stack” areas, even while it lags in chips—suggesting the strategic battleground is shifting from compute scarcity to software, tooling, and deployment layers. Separately, Evoken, the Chinese startup behind the AI design agent Lovart, is reportedly exploring an IPO in Hong Kong, positioning itself to monetize investor demand for AI tools after the breakout success of Manus. Finally, the Financial Times reports that banks are set to offload about $15bn of debt tied to an Anthropic data center backed by Google, a move framed as freeing up lending capacity as mega AI deals strain Wall Street financing limits. Geopolitically, the cluster points to a dual contest: China’s push to win on the broader AI stack while the US-led financial system and hyperscalers manage the capital intensity of scaling frontier capabilities. China benefits from a narrative of software and systems leadership that can partially offset chip constraints, while the US and G7 face pressure to defend not only model innovation but also the industrial organization of AI deployment. The Hong Kong IPO angle matters because it links China’s AI commercialization to a global capital venue that can accelerate fundraising, talent retention, and platform consolidation. Meanwhile, Wall Street’s need to recycle balance-sheet capacity for AI infrastructure underscores that the “bottleneck” is increasingly financial intermediation, not just engineering talent. In short, the winners are likely to be those that combine cost-efficient model development, rapid productization, and scalable funding channels. Market and economic implications are immediate for AI-linked capital markets and for the supply chain that supports data center buildouts. The reported $15bn debt offload connected to an Anthropic data center backed by Google signals potential near-term liquidity relief for banks, but also highlights that AI infrastructure financing is approaching a scale where balance-sheet constraints become a pricing factor. This can influence spreads on AI-adjacent credit, mortgage-like securitization structures, and corporate bond issuance calendars, with knock-on effects for insurers and asset managers that buy yield. On the equity side, a potential Hong Kong IPO for Lovart would add to the flow of AI tool listings, potentially increasing volatility and valuation dispersion across Chinese AI software names. Commodities are not directly cited, but the data-center financing theme typically transmits into demand expectations for power equipment, cooling systems, and construction inputs, which can later feed into industrial procurement cycles. What to watch next is whether Chinese developers can sustain the “performance-to-cost” advantage without triggering a price war that compresses margins across the model ecosystem. For investors, the key trigger is confirmation of Lovart’s Hong Kong IPO timetable, including underwriting partners, deal size, and whether it targets a valuation premium or a growth-at-a-reasonable-price strategy. On the financing side, monitor follow-through on the $15bn debt offload mechanics—who buys the paper, how it is structured, and whether it reduces funding costs for subsequent AI infrastructure tranches. For policy and competitive intelligence, Rhodium’s “tech stack” findings should be tracked for measurable milestones in tooling, agent frameworks, and deployment platforms that can translate into revenue faster than chip progress. Escalation risk is not kinetic, but the competitive stakes are high: if cost competition accelerates faster than monetization, consolidation and regulatory scrutiny over AI disclosures and client communications could intensify across jurisdictions.

Geopolitical Implications

• 01

  Software and deployment layers may become the primary battleground in the US–China AI contest as chip constraints persist.

• 02

  Hong Kong’s role as a capital bridge can amplify China’s AI commercialization and reduce reliance on purely domestic funding.

• 03

  Wall Street’s balance-sheet limits for mega AI deals may reshape who can scale infrastructure fastest, affecting long-run competitive positioning.

Key Signals

• —Confirmation of Lovart’s IPO filing/underwriting, including deal size and valuation range.
• —Details on the $15bn debt offload: buyer base, structure, and whether it lowers subsequent AI infrastructure borrowing costs.
• —Evidence that Chinese developers can sustain performance-to-cost leadership without monetization lag.
• —Measurable progress in agent frameworks and deployment tooling that Rhodium flags as tech stack strengths.