Summary of Key Points
Although large AI model companies (such as Zhipu and MiniMax) are suffering heavy losses, the upstream hardware manufacturers (which provide the necessary components like optical modules, PCBs, and some chips) are making huge profits. Global AI investment is flowing in the tens of trillions of dollars, with funds moving from cloud service providers to companies that solve the physical connectivity issues (such as optical modules and PCBs), then to chip and server manufacturers, and only finally to the application companies. However, these upstream companies facing rapid profit growth also face risks related to market cycles, competition, and the quality of their revenue. In the future, the main sources of profits may shift to domestically produced chips, cloud service providers, or AI application developers.
How Does Money Flow in the AI Industry?
Money in the AI industry doesn't disappear; it moves along the supply chain:
- Starting point: Cloud service providers (Alibaba, Google): These giants spend heavily on purchasing GPUs and building data centers, depleting their profits and cash flows. For example, Alibaba's first-quarter financial report showed significant AI-related expenses eroding its profits.
- First stop: Companies that solve physical connectivity issues: GPUs need to work together, and this requires optical modules to transmit data (similar to laying high-speed fiber optic networks for supercomputers); PCBs serve as the foundation for GPUs. These companies receive the money first because cloud providers must buy their products before building data centers.
- Second stop: Chip and server manufacturers: Domestic chips (such as Cambricon) have secured orders by replacing NVIDIA, while server manufacturers (like Foxconn Industrial Internet) help assemble the equipment, earning substantial profits from the scale of production.
- Final stop: Application companies: Companies that develop large AI models and AI assistants can start making money only after the underlying infrastructure is in place, but most are still losing money for now.
Who Gets the First Taste of AI's Benefits?
The following upstream hardware companies are among the biggest beneficiaries:
1. Optical module manufacturers: Zhongji Xuchuang reported a net profit of 11.85 billion yuan in 2025 (a 112% increase) and another 5.72 billion yuan in the first quarter of 2026 (more than its entire profit for 2024); NeoPhoton Technologies' net profit increased by 14 times in two years, with a gross margin of nearly 48%. They rely mainly on orders from overseas cloud providers (Google, Amazon) and NVIDIA and are essential in the global AI investment process.
2. PCB manufacturers: Shenghong Technology, which originally made money from smart devices like smartphones, saw its AI-related revenue soar from 5.8% to 43.2% in 2025, with a net profit increase of 277%. Its customers include NVIDIA and Microsoft.
3. Some domestic chips: Cambricon turned a profit for the first time in 2025 (1.77 billion yuan) after five years of losses, and its net profit increased by 238% in the first quarter of 2026; Higuang Information is also growing steadily. However, not all domestic chips are profitable—Moor Threads and Muchi are still losing money, though less significantly.
4. Server manufacturers: Foxconn Industrial Internet had revenue of over 90 billion yuan and a net profit of 34.1 billion yuan in 2025, but its gross margin was only 6.98%. Inspur Information saw its revenue increase by 43%, but its profit grew by only 11%, indicating limited bargaining power.
Counterintuitive Facts: The Further from Users, the More Stable the Profits
The order of profit generation in the AI industry is different from what we might expect:
- Companies far from users profit first: Optical module and PCB manufacturers receive orders before cloud providers even start serving customers, so their revenue is more stable. For example, the stock prices of optical module companies increased by 100% in 2025 and another 60% in 2026.
- Midstream chip companies are at risk due to substitution: Cambricon's market value has exceeded one trillion yuan because NVIDIA is under restrictions, forcing domestic cloud providers to buy its chips. However, this could change if policies relax.
- Companies closest to users are the most uncertain: Large AI model companies (like Zhipu) are losing the most money. Although cloud providers' AI revenue is growing, the cost of purchasing GPUs is high, making it difficult to accurately measure their AI-related profits. Internet giants (such as Tencent and Kuaishou) integrate AI into advertising and recommendation services, but it's unclear how much of their earnings come from AI, and their stock prices are not easily predictable.
Will This Good Times Last?
The currently profitable upstream companies may face challenges in the future:
1. Peak in capital spending: Cloud providers don't have unlimited funds; Google's cash flow is nearly depleted. CICC predicts that large cloud providers' free cash flows will drop from 200 billion yuan to 24 billion yuan in 2027. If investment decreases, the high margins of optical module and PCB manufacturers may be at risk.
2. Risks associated with heavy reliance on key customers: Companies like Zhongji Xuchuang and NeoPhoton Technologies rely heavily on overseas markets; a decline in orders from a major customer (e.g., Google) could significantly impact their performance.
3. Intensifying competition: The entry of electronics giants like Lixun Precision and Dongshan Precision into the optical module industry could lead to price wars, reducing profit margins.
4. Questionable revenue figures: NVIDIA's financing of its own chips through OpenAI creates inflated revenue figures. If the industry growth slows, there will be doubts about the actual profitability of these companies.
Who Will Profit Next?
The flow of money in the AI industry will continue, but the beneficiaries may change:
1. Domestic chip alternatives: Companies like Cambricon and Higuang will continue to profit as long as NVIDIA's restrictions remain, creating market opportunities.
2. Cloud service providers offering comprehensive solutions: Alibaba Cloud's MaaS platform (model-as-a-service) aims to generate annual revenue of 30 billion yuan by the end of the year. With a potential gross margin similar to that of overseas companies like Microsoft and Google, there is significant room for growth.
3. AI for business applications: Companies developing AI solutions for enterprises (e.g., Palantir) have shown promise; they could become successful in the future.
Just as in the internet era, companies that built the infrastructure (such as Cisco) eventually gave way to leaders like Amazon and Google, the current fast-profitting upstream players in the AI industry may not be the ultimate winners. The real long-term opportunities will lie with those who can convert AI into stable, recurring revenue streams.
(The entire analysis is written in plain language to explain how money is generated in the AI supply chain, who is making profits, and who might do so in the future.)