At the AI Investment Summit held in Beijing on September 16, hosted by Sina Finance under the theme "Certain Opportunities in the AI Infrastructure Era," Wang Qing, Chairman of Chongyang Investment, shared his perspectives on the AI computing supply chain. From the vantage point of an active capital market participant, he analyzed the potential impact of super node technology on the landscape of AI inference and training, as well as the funding constraints and market risks underlying the current global capital expenditure expansion.
Super Nodes: A New Variable in Inference and Training
Wang Qing candidly admitted that he is not a technical expert, but as a capital market participant, he believes super nodes open up fresh possibilities for the entire AI industry. On the inference side, the parameter scale of leading domestic large language models continues to climb, with Kimi3 already reaching the trillion-parameter level and potentially ascending further within the year. This demands simultaneous upgrades in both computing power and model architecture, and the adoption of Mixture of Experts (MOE) architectures places higher demands on parallelization strategies. Wang Qing predicts that such requirements may only be met through systematic integration approaches like super nodes. If successfully implemented, AI model inference could enter a new cycle.
The training side presents a different picture. Due to physical constraints in advanced chip manufacturing processes, domestic efforts in training are comparatively lagging. However, domestically produced chips are playing an increasingly important role in inference tasks. Wang Qing believes that breakthroughs in training will also rely on engineering optimization to overcome the limitations posed by individual chips. If super nodes achieve widespread adoption, they could propel domestic AI training into a new phase, which would carry significant implications.
Capital Expenditure: Demand Is Not the Constraint, Funding Costs Are Key
From a demand perspective, Wang Qing believes there is currently no constraint. The fundamental funding landscape is as follows: capital expenditures by major US firms stood at roughly $400 billion last year, are expected to approach $800 billion this year, and could reach $1 trillion next year, with the scale still expanding. The sources of funding are shifting. Last year, these expenditures were primarily funded by internal cash flows; this year, those firms' operating cash flows have turned negative, prompting them to seek external financing, including equity and debt issuance.
Domestically, the pace is slightly slower. Last year, capital expenditure was around RMB 500 billion, this year it is RMB 1 trillion, and it is expected to increase further next year. Domestic major firms' operating cash inflows and outflows have now roughly reached equilibrium this year, and some have begun issuing shares or bonds. However, Wang Qing emphasized that for domestic capital expenditure, the most immediate constraint remains the availability of computing power.
US tech giants have ample borrowing capacity. Wang Qing estimates that at a 1x net leverage ratio, there is theoretically still $1 trillion in financing capacity; at 2x net leverage, it approaches $1.9 trillion to $2 trillion. Current on-balance-sheet debt ratios are around 27%, and even after accounting for off-balance-sheet and contingent liabilities, they remain below 40%, at approximately 36% to 37%, leaving room for further expansion. The critical issue is cost.
Financing methods are evolving from internal cash flows, bond issuance, and finance leases to independent data center construction and cloud providers leasing data centers. These data centers, in turn, are financed through private equity and private credit. Wang Qing noted that financing rates for projects backed by or credit-enhanced by major cloud providers are around 5.3% to 6.6%, close to US long-term Treasury yields. Projects without such credit backing command higher rates of 8.5% to 9%.
The cost constraint is closely tied to the macroeconomic and capital market environment. US inflation remains persistently above target, fiscal conditions mean the government is continuously borrowing from markets, and interest rates stay elevated. Funding costs may well be the factor that caps capital expenditure, which is why the market closely watches the Federal Reserve's monetary policy and interest rate movements.
Certain Directions and Potential Risks
Wang Qing believes the most certain directions are, first, computing power demand and a new wave of capital expenditure. Although China trails the US by half a step, capital expenditure by major domestic firms is clearly accelerating, from RMB 500 billion last year to RMB 1 trillion this year. These firms are intensifying their deployment efforts, with particularly visible opportunities in the office applications sector.
Infrastructure leasing and commercial application scenarios are gradually opening up, and the supply-side channel for storage, computing, design, and production has been established, creating synergy between supply and demand. Another certain direction is cloud services themselves. Cloud services require higher-level and more sophisticated offerings, and demand-side large model iterations continue to drive growth. Both directions appear relatively assured.
On the risk front, Wang Qing highlighted potential pressure on upstream component suppliers in the supply chain. While demand is robust, short-term supply cannot keep up, and this year's price increases have put pressure on downstream players. However, these sectors have low barriers to entry and limited moats. Chinese manufacturing capacity catches up quickly, and recent financing activities by listed companies in this space suggest that capacity will ramp up soon, making sustained price increases unlikely. If prices cannot hold, the market's current earnings forecasts and valuations for these companies would need adjustment, posing a potential risk.
The final risk stems from market behavior. Wang Qing pointed out that these opportunities are highly consensus-driven, particularly within China's market structure, which tends to facilitate rapid crowded trades and herding behavior. This could create risks at the capital market level, though not necessarily at the industry level. Given the market correction that began in July, this serves as a reminder to investors: changes in fundamentals and changes in the capital market do not always move in tandem. It is essential to monitor not only the pace of fundamental evolution but also the market's own valuation levels, risk premium, and even investor behavior characteristics.
MACD golden cross signals have formed, and these stocks are performing well!
Comments