Profit distribution along the US AI industrial chain is strikingly vertical and hierarchical. Chip and hardware makers — the upstream "water sellers" — occupy an absolutely dominant position with trillion-dollar-scale revenue, while the combined revenue of model companies at the downstream application layer still falls short of $100 billion. Baorong Wanwu Hengheshui (包容万物恒河水, a commentary account on Weibo) identifies a structural problem here: profits from the AI wave penetrate upward with extreme ease — the upstream enjoys the dividend with certainty — while the cost burn flowing downward is equally enormous, as the downstream is still hunting for a viable business model. If this fault line cannot be absorbed by downstream commercialization, it may evolve into a systemic risk driven by the extreme concentration of financial capital.
The Profit Fault Line Between Upstream and Downstream — Trillions Versus Billions
By early May 2026, the US stock market was in a state of "extreme euphoria." Beyond the traditional "Magnificent Seven" (NVIDIA, Alphabet, Apple, Microsoft and the rest), Broadcom had broken through $2 trillion, and Walmart and Berkshire Hathaway sat firmly in the trillion-dollar club. Eli Lilly briefly topped $1 trillion before pulling back, while JPMorgan and Micron were closing rapidly on the trillion-dollar threshold.
But one question is decisive: what is holding up the distribution of profit along the AI industrial chain?
Downstream model companies: the combined annualized revenue of frontier model companies such as OpenAI, Anthropic and Gemini is estimated to remain below the $100-billion level (OpenAI roughly $25 billion, Anthropic roughly $44 billion).
Upstream hardware makers: semiconductor giants such as NVIDIA, Micron, Samsung and SK Hynix, together with the AI-chip-adjacent segments of SSDs, optical modules and foundry services, have reached trillion-dollar-scale total revenue.
This fault line means: profits of the AI wave penetrate upward with extreme ease (upstream vendors enjoy a certain dividend), but the cost burn flowing downward is equally enormous (the downstream application layer is still exploring viable business models).
A New Peak of Capital Fictitiousness — Financial Capital Ascendant Over Industrial Capital
This AI feast is not merely technology-driven; it is dominated by financial capital:
"Capital — especially monopolistic financial capital — is aggregating on an unprecedented scale and showing a tendency to tower above all industrial capital. A handful of enterprises harvest outsized financial profits through the amplifying effect of market expectations and valuation mechanisms, while the profits of physical industrial capital remain comparatively subordinate. This is a new peak in the twenty-first century of the process of capital fictitiousness — the movement of monetary capital is increasingly detaching itself from the valorization process of industrial capital and operating on its own."
Concretely:
- Semiconductors and chips are the most easily ignitable point of entry for finance — every breakthrough in chip architecture and every upgrade in process technology is extraordinarily well suited to absorbing enormous volumes of monetary capital, because it fits perfectly the financial narrative template of "optimistic expectations for future productive forces."
- Large market capitalization encodes the market's optimistic expectation of long-run future cash flows — once ever more capital floods into a given field, later investors follow without a second thought, pushing prices still further away from fundamentals.
Historical Analogies and Key Differences — What the AI Boom Actually Resembles
Hengheshui's analysis supplies a useful historical framework:
| Analogy | Structural Similarity | Key Difference |
|---|---|---|
| The .com bubble (1990s) | Financial capital capturing value in high-growth fields; valuations detached from fundamentals | AI has a genuine technological driving force (the improvement in LLM capability is a fact, not pure narrative) |
| The metaverse frenzy | Capital hyping "future narratives" | AI's compute demand genuinely exists and is rapidly consuming the world's electricity resources |
| Overcapacity in traditional industries (steel, automobiles) | High concentration of capital → accumulation of overcapacity | AI's fixed-investment threshold is extremely high (trillions of dollars in compute infrastructure), and the cost of recovery after a bubble burst would be even greater |
The US AI boom is distinctive for two reasons: on one hand, its potential uplift to labor productivity may be among the largest since the Industrial Revolution; on the other, realizing that uplift requires colossal up-front infrastructure investment (trillions of dollars in compute-infrastructure spending worldwide). Financial capital supplies the financing channel this construction requires — yet it may also interrupt the rhythm of that investment if the bubble bursts.
Two Key Variables — Profit That Cannot Be Transmitted, and a Missing Manufacturing Base
Hengheshui singles out two key variables that run against the United States:
1. Upstream profits cannot propagate downstream. The high margins of chip companies may prove unable to diffuse into the downstream application layer through ordinary industry competition. This matches the classic structure in industrial economics of "upstream monopoly → downstream compression": when upstream hardware vendors possess extremely high technical barriers (NVIDIA's CUDA ecosystem, TSMC's process technology) and extremely short Moore's-law iteration cycles, the downstream application layer can only keep paying the bill, unable to win profit room by competing prices down.
2. The manufacturing base is missing. AI's broad effect on social labor productivity has to be realized through implementation in the real economy, and that requires the cooperation of a strong manufacturing sector. The structural weakness left by decades of US deindustrialization is further amplified at this moment — even if AI achieves breakthroughs in software, converting those breakthroughs into real-economy productivity gains (factory automation, supply-chain optimization, energy-efficiency improvement) requires a still-active manufacturing base, and the gap between the United States and China on this front is widening.
Hengheshui points to a paradox: among the world's top 20 listed companies by market capitalization there is not a single Chinese enterprise. In itself, this reflects how "negligible" the US capital market holds the valuations of large Chinese companies to be. But viewed through the combined competitiveness of "manufacturing + AI," a global top-20 list without Chinese companies may precisely reflect the capital market's severe underestimation of China's "manufacturing + AI" integration potential — rather than any actual lack of competitiveness of Chinese firms in these fields.
Short-, Medium- and Long-Term Prospects
Short term (6–12 months): the bubble need not burst immediately. As long as compute demand keeps outstripping supply, the seller's market in AI chips will persist; the supply shortage of NVIDIA's B200/N100 series will not change in the near term.
Medium term (1–3 years): valuations need the support of delivered earnings. If the downstream application layer's commercialization lags Wall Street's aggressive expectations, overstretched valuations will come under pressure. Key metrics to watch: revenue growth versus operating-expense growth at OpenAI/Anthropic, and the return on AI capital of CSPs (cloud service providers).
Long term (3–5 years): real innovation is what counts. Whether AI can shift from a "cash-burning race" into a "self-sustaining" business model depends on finding implementation scenarios of sufficient scale around the world — and the density of "manufacturing + AI" scenarios may be higher in China than in the United States.
Trackable Verification Metrics
The analytical framework on this page yields the following trackable indicators:
- NVIDIA's revenue growth and gross-margin trajectory: once revenue growth begins to slow — from intensifying competition or demand saturation — that will be the first trigger point for bubble pressure.
- The timetable for downstream model companies' gross margins to turn positive: if OpenAI/Anthropic fail to reach positive operating cash flow before 2027, a valuation correction becomes unavoidable.
- The trend in AI-related electricity consumption and compute costs: a rising share of power costs will compress profit room across the entire industrial chain.
- The moment Chinese AI companies enter the global top 20 by market capitalization: this is the verification — or falsification — of the judgment that "the US capital market underestimates China's AI competitiveness."
This essay is based on an in-depth analysis published by Baorong Wanwu Hengheshui at 17:40 on May 9, 2026, exposing the structural fault line in upstream–downstream profit distribution along the US AI industrial chain and its systemic risk. Hengheshui advances a core judgment: the concentration of financial capital in the AI wave has already reached a critical point — if the downstream application layer cannot close its business-model loop in the medium term, the valuation risk released in concentrated form may go far beyond a simple stock-market correction and strike at the financing foundations of the US real economy.