CITIC Securities: Code Inflation, Physical Scarcity

This February marked a significant watershed moment. The leap in AI coding capabilities has officially pushed the global scale of effective code into an exponential growth phase. Under current physical AI technology conditions, the expansion rate of society’s tangible production value and total income lags far behind the growth rate of AI-generated code. The world is highly likely to first experience a phase of code proliferation, excess execution capacity, intensified competition, and diminished returns on capital investment.

Based on two dimensions—physical dependence across industries and regulatory/emotional barriers—we can categorize industries into four groups: damaged (low physical dependence, low regulatory/emotional barriers), reshaped (low dependence, high barriers), fortresses (high dependence, high barriers), and beneficiaries (high physical dependence, low regulatory/emotional barriers). In the near future, the gap in benefits between tangible resource beneficiaries and those harmed by code expansion may continue to widen, and this divergence trend will persist. This is a new factor that must be considered when analyzing market conditions and sector allocations.

From a short-term market perspective, the structure of A-shares is primarily centered on manufacturing and finance. Under this AI wave, their impact has been relatively smaller compared to US and Hong Kong stocks. The pattern of capital inflow and investor optimism remains unchanged. The spring market after the holiday is expected to continue, and price increases remain one of the key investment clues for the first quarter.

The scale of effective global code has officially entered an exponential expansion phase

If last year’s Spring Festival’s DeepSeek hype fueled market enthusiasm for AI applications, then this year’s explosion of Coding Agents has caused widespread anxiety about the expansion of global code and the disruption of traditional software applications. On February 5, 2026, OpenAI and Anthropic released new models on the same day: GPT-5.3 Codex and Claude Opus 4.6. The experience they offer is revolutionary—AI has evolved from a “helper tool” to an independent executor at the coding level. According to official sources, GPT-5.3 Codex and Claude Opus 4.6 are both positioned as agentic models capable of planning, debugging, and multi-step modifications within large codebases, no longer limited to passive “prompt and supplement code” assistance.

This means that any workflow describable in language and expressed in code will be rapidly replaced by AI. It also signifies that the scale of effective code worldwide is entering an exponential growth stage.

The total value of physical societal production and total income far lag behind the growth of code volume

According to IEA and Ember Energy data, global electricity generation is expected to grow from about 30,000 TWh in 2024 to approximately 32,000 TWh in 2026, with a compound annual growth rate of only 3.3%. Meanwhile, data center energy consumption will increase from about 600 TWh in 2024 to roughly 1,050 TWh in 2026 (optimistic scenario), with a high compound growth rate of 32.3%. In the short term, energy growth clearly cannot keep pace with the total increase in code and token consumption. Addressing energy consumption and latency in engineering is more important than simply scaling up computing power.

The ratio of total GitHub code repositories (in millions) to global GDP (trillions of dollars) was only 3.93 in 2023, rising to 5.38 by 2025. Based on the growth rate of GitHub repositories disclosed in the 2025 GitHub Octoverse and IMF GDP forecasts, we estimate that by 2026, this ratio could further increase to 6.29.

Meanwhile, competition among large models worldwide has intensified. The capabilities of leading models have not diverged with increased compute investment—in fact, the gap is narrowing. Future income competition is expected to become more intense, with marginal costs rising due to hardware price increases. Perhaps someday, the maturity of embodied intelligence will enable society to make rapid advances in resource acquisition and physical production, potentially establishing an effective distribution mechanism to curb wealth disparity—though this remains unlikely in the short term.

Overall, the short-term payment capacity clearly cannot keep up with the expansion of code-based products and the associated compute costs. We are likely to experience a process of widespread code proliferation, excess execution capacity, intensified competition, and diminished returns on capital.

Which industries are harmed by code expansion? Which benefit from physical resource scarcity?

Using the two dimensions of physical dependence and regulatory/emotional barriers, industries can be divided into four quadrants: 1) Damaged (low physical dependence, low barriers)—AI-generated code and content are highly usable here, and business models are relatively easy to be replaced. Lacking exclusive data or know-how as a moat, typical industries include basic code outsourcing, general SaaS, marketing, and PR; 2) Reshaped (low dependence, high barriers)—businesses are primarily digital but involve legal accountability, financial regulation, or rely on psychological trust and emotional connections, forming moats that are hard to replace with pure code. AI acts as a “super leverage,” mainly assisting decision-making and improving efficiency, driving “downsizing and efficiency gains,” rather than disruptive replacement. Typical industries include legal litigation, high-end strategic consulting, asset management; 3) Fortresses (high dependence, high barriers)—involving monopolistic assets or scarce resources, such as core minerals, military manufacturing, transportation infrastructure, and high-end luxury goods, brands, or emotionally valued products that must connect with physical goods; 4) Beneficiaries (high dependence, low barriers)—related to physical carriers like copper, aluminum, energy metals, semiconductor manufacturing, PCBs, optical modules, servers, energy infrastructure like power equipment and transformers.

Physical resource beneficiaries and the ongoing divergence under code expansion

Based on the logic of “physical resource benefits” versus “code proliferation harms,” we have separately analyzed the benefit and harm portfolios in US and Chinese markets. Since 2026, the cumulative return gap between these two groups in the US market has widened by 64 percentage points. In China, benefiting from abundant liquidity and capital inflows, the divergence is less pronounced. Compared to the end of 2025, the excess return of the “physical resource benefit” portfolio over the “code inflation” portfolio in A-shares has only increased by 3 percentage points. This is partly because some software and media stocks in A-shares and Hong Kong stocks surged due to AI hype in January, contrasting sharply with the 20% decline in US software and services sectors since early this year.

However, as global markets become more interconnected and liquidity premiums dissipate, we believe Chinese assets will eventually reflect the divergence between “physical resource scarcity” and “code expansion.”

A-shares industry structure remains centered on manufacturing and finance, with less impact from the current AI wave compared to US and Hong Kong stocks

Because China’s B2B enterprise services and software markets are relatively small, the impact of AI-generated code inflation on physical industries will be much less in early stages than in North America. For example, as of February 13, 2026, the market cap share of software and service companies in US stocks is 22.8%, compared to 31.5% in Hong Kong and only 5.6% in A-shares. Software and enterprise services have historically been the “third pillar” of North American tech outside mobile internet and biotech, with stable, mature business models, high barriers, high capital returns, and steady cash flow. But in the face of disruptive AI innovation, they are most affected.

Hong Kong’s internet giants mainly focus on consumer-to-company (ToC) businesses and are also trying to reconstruct their operations with AI, but they are inevitably affected by global market linkages. In contrast, A-shares are still dominated by monopolistic finance, manufacturing, and energy sectors. During the code expansion, traditional resource-based and manufacturing industries that build AI infrastructure are likely to benefit—they represent truly scarce physical assets and are important safe havens for global capital in the coming years.

Capital inflows and investor optimism remain unchanged; spring market after the holiday is expected to continue

The trend of capital inflow remains steady. According to PBOC data, in January, household deposits increased by 33.9 billion yuan year-on-year, while non-bank financial institutions’ deposits increased by 2.56 trillion yuan year-on-year. As “high-yield” deposits mature, the decline in January deposit data indicates a continued shift toward financial products like wealth management and savings policies, which will eventually channel some funds into equity markets. Overall, the pattern of capital inflow and investor optimism remains intact.

The slight market correction before the Spring Festival may be related to the large gains in January, increased overseas market volatility starting in February, and the longer holiday period. Risk-averse demand before the holiday has been rising. As of February 13, our constructed A-share investor sentiment index (single-day, MA5, MA10) stood at 43.6, 55.7, and 65.5 respectively, showing a clear retreat from January’s levels and hitting the lowest since 2025. By the end of January, the latest positions of private funds surveyed through CITIC Securities channels averaged 79.3%, down significantly from 84.3% in the week of January 23. These signs suggest that risk-averse capital had already reduced positions before the holiday, creating room for post-holiday replenishment.

Price increases remain one of the key investment signals for Q1

1) Our annual strategic framework is based on China’s resource and traditional manufacturing pricing power revaluation. The core logic is that China’s market share advantage is clear, overseas capacity reset costs are high or difficult, and supply elasticity is somewhat influenced by domestic policies. Accordingly, we allocate to sectors such as chemical, non-ferrous metals, electrical equipment, and new energy as the foundation. We also increase positions in undervalued insurance and brokerage (exposing some low-valuation factors), and expand exposure to consumer chains (duty-free, airlines, hotels, scenic spots, fresh tea drinks) and real estate chains (quality developers, building materials, REITs). The idea is to benefit from the expected broadening of market differentiation and the mild recovery of domestic demand and prices.

Even with the volatility in precious metals and commodities in early February, the rebound in the US dollar index, and adjustments in cryptocurrencies and overseas small-cap tech stocks, this core configuration remains valid. Price increases are the most straightforward and immediate trading signals under this framework.

2) We also need to consider the new framework of “code inflation” versus “physical resource scarcity.” The rapid global expansion of code brought by AI may significantly impact low-dependence, highly competitive businesses, while high-dependence, highly regulated, and emotionally barriered businesses could become more scarce. Under this framework, investors may favor fortress assets—those currently insulated from AI disruption—and avoid businesses vulnerable to disruptive innovation, regardless of whether AI agents can immediately replace them.

In our current allocation, industries such as resources, traditional manufacturing, energy, consumer services, and real estate are, at least temporarily, fortress assets shielded from generative AI impacts.

Risk factors

Intensified US-China technological, trade, and financial frictions; domestic policy implementation or economic recovery falling short of expectations; macro liquidity tightening beyond expectations; escalation of conflicts in Ukraine, the Middle East, and other regions; and China’s real estate inventory digestion delays.

Source: CITIC Securities Research

Risk Disclaimer and Legal Notice

Market risks exist; investment decisions should be made cautiously. This article does not constitute personal investment advice and does not consider individual user’s specific investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions herein are suitable for their particular circumstances. Investment is at their own risk.