DeepSeek Anniversary: From AI Shock to China’s Tech Self-Reliance Era

What happened

January 20, 2025 marked the release of DeepSeek-R1. One year later, the event's significance extends well beyond a single large language model. It crystallized three structural shifts already underway in China's technology ecosystem: innovation under constraint, domestic revenue anchoring, and a decoupling of return drivers from U.S. technology.

DeepSeek-R1 achieved near-frontier reasoning performance at an estimated 3% of comparable training costs. The immediate market reaction, Nvidia's 18% intraday drop, DeepSeek overtaking ChatGPT on the App Store, was dramatic but short-lived. The structural implications are more durable.

Impact on our Investment Case

A Watershed Moment

Structural implications of the DeepSeek moment are now visible in how Chinese AI is built, commercialized, and brought to market.

Innovation under constraint

U.S. export controls on advanced semiconductors were designed to slow Chinese AI development. Instead, they accelerated architectural efficiency. DeepSeek's mixture-of-experts approach demonstrated that compute optimization may matter more than raw compute access for commercially viable AI. This has implications beyond China: it suggests the capex intensity of frontier AI may be lower than current hyperscaler spending implies.

Open-source as competitive strategy.

DeepSeek and Alibaba's Qwen adopted an open-weight strategy that has proven remarkably effective. Qwen has overtaken Meta's Llama in global downloads. Open-sourcing shifts monetization from model licensing toward infrastructure and enterprise services, areas where Chinese firms have domestic scale advantages and lower customer acquisition costs.

Speed to production.

Within months of release, DeepSeek models were deployed in enterprise environments across manufacturing, logistics, and financial services. This was not research demonstration but production-grade integration, suggesting a deployment velocity that compresses the gap between model capability and commercial application. 

Most importantly, the DeepSeek moment clarified something fundamental: China technology is not a derivative play on global tech. It is a distinct asset class with its own return drivers.

A Distinct Asset Class

If China's technology is not a derivative play on global tech, the investment case must stand on its own fundamentals. Three dimensions define it: revenue structure, valuation, and correlation.

The Domestic Revenue Structure

A defining characteristic of China's technology sector, often misunderstood by global investors, is its domestic revenue concentration. Using our proprietary market-cap-weighted China Technology index, we estimate that approximately 75% of revenues are generated within China.

This could be, wrongly, interpreted as a limitation: lack of global reach, vulnerability to domestic slowdown, restricted addressable market. The alternative interpretation is that domestic concentration is the core feature of the tech self-reliance model, and we do believe in the latter thesis, being a long-standing investor in Chinese companies in our offering of thematic equity portfolio.

Domestic concentration delivers three structural advantages that the market has yet to fully price. The transmission mechanism from policy to earnings is unusually direct. When the 15th Five-Year Plan allocates capital toward digital infrastructure and SOE modernization, domestically-anchored firms capture that spending as revenue. This creates a shorter, more predictable path from policy announcement to earnings than global consumer-facing models offer. At the same time, what appears as geographic limitation actually functions as regulatory insulation. Firms generating most of their revenue in China face materially lower exposure to U.S. sanctions, export controls, and delisting risk, and allocation through A-shares and H-shares sidesteps ADR vulnerability entirely. Finally, the demand profile itself has shifted. Policy-driven procurement from SOEs and provincial governments replaces the advertising-dependent revenue that characterized the "China Trade" of 2020. The 90% AI adoption target for SOEs by 2030 is not an aspiration but a procurement schedule, translating directly into enterprise software and services demand. 

Valuation Context

As of December 2025, China technology currently trades at approximately 26x forward earnings against consensus earnings growth of 29% (PEG ratio ~0.9x). U.S. technology trades at approximately 38x forward earnings against 16% expected growth (PEG ratio ~2.4x). On a growth-adjusted basis, China tech is priced at roughly one-third the valuation of its U.S. counterpart.

This gap reflects a market still anchored to the pre-2024 narrative, cyclical China exposure, regulatory overhang, and global demand dependency. It has yet to fully price the structural shift toward policy-anchored, domestically-driven growth. With 75% of revenues generated internally and demand increasingly tied to visible government capex cycles, earnings volatility should compress relative to the platform-dependent U.S. model. The current discount prices-in uncertainty that the new revenue structure is designed to reduce.

The investment case does not require multiple expansion. It requires only that earnings compound at the rate implied by consensus. The valuation discount provides margin of safety; any re-rating is upside optionality.

Correlation and Portfolio Implications

From an asset allocator perspective, the recent correlation of Chinese versus U.S. technology reveals an extremely low level, approximately 0.3. Mapping the structural differences over those two themes helps grasp the different dynamics behind.

The implication is that China technology is not a leveraged bet on global tech sentiment. It is a structurally distinct exposure with different return drivers. For portfolios concentrated in U.S. technology, whether through explicit allocation or passive index exposure, China technology offers genuine diversification rather than amplification.

Our Takeaway

The case for China technology rests on structural shifts, not cyclical hope. But structural does not mean risk-free. Here is what could go wrong and how we mitigate those concerns in the construction of the portfolio for our systematically enhanced strategy that banks on this sector.

Further U.S. restrictions on technology transfer, investment flows, or aggressive secondary sanctions remain the primary tail risk. The domestic revenue anchor provides partial insulation, not immunity, so we filtered the investable universe based on various sanction exclusion lists, such as the Chinese Military Industrial Complex Companies list from the U.S. Department of the Treasury.

Foreign investor access to A-shares remains constrained. The Chinese market microstructure differs materially from developed markets. Position sizing must account for execution costs and exit liquidity, hence we applied relative and absolute liquidity filtering constraints for the investable universe and put a weight cap on smaller companies.

One year after DeepSeek, our thesis is not speculative. The evidence is observable:

• Open-source AI models with global production adoption.
• 75% domestic revenue concentration provides policy transmission.
• 0.3 correlation with U.S. technology over the past year.
• Valuation at ~0.9x PEG versus ~2.4x for U.S. technology.

The real question for allocators is whether portfolios are positioned for a technology growth cycle that is domestic in revenue, industrial in application, and structurally uncorrelated with the platform-centric U.S. model. China technology is not a substitute for U.S. technology exposure. It is a complement, a distinct return stream, driven by different factors, with different risks, at different valuations. The DeepSeek moment clarified that this distinction is structural, not cyclical. Our dedicated strategy stands to benefit.