Summary of Key Points
The central argument of this article is that the current boom in the AI storage industry is not based on the supply and demand for storage chips themselves, but on the funding that supports the AI industry—huge investments made by AI companies and tech giants through financing and debt issuance. Once the capital market's tolerance for AI losses decreases and the funding chain breaks, the storage industry will be the first to be impacted. The article provides four indicators to determine whether the market has peaked by comparing input-output ratios, analyzing the sources of funds for AI companies and giants, and examining historical cycle risks.
1. Is the Investment-to-Return Ratio of 27:1 Really Exaggerated?
In 2026, the five major tech giants (Microsoft, Google, etc.) invested a total of $690 billion in AI infrastructure, while global AI service revenues were only $25 billion, resulting in an input-output ratio of 27:1—meaning for every $27 invested, only $1 was earned. How is this gap being filled? AI companies rely on financing to survive (for example, OpenAI raised $122 billion), and tech giants issue debt (they issued $159 billion in the first half of 2026, more than in the entire previous year). This cycle can continue only because investors believe that AI will eventually become profitable, but no one knows when that will happen.
2. AI Companies: Either Facing Huge Losses or Just Beginning to Profit
- OpenAI: Had revenue of $13 billion in 2025 and a net loss of $38.5 billion (losing $1.6 for every $1 earned), with a commitment to purchase computing power worth $66.5 billion, so it can only rely on going public to raise funds.
- Anthropic: Is in a slightly better position with annual revenue of $47 billion and is expected to break even for the first time in the second quarter of 2026. However, its profitability is a key signal—if it can continue to make money, the capital market may shift from tolerating losses to demanding profits, which would directly shake the financing logic of the AI industry and affect storage demand.
3. Tech Giants: Spending All Their Profits on Data Centers, Relying on Debt
In 2020, the infrastructure investments of the five giants accounted for only 39% of their cash flow, and they were still able to pay dividends and make acquisitions; by 2026, these investments accounted for 94% of their cash flow—almost all profits were invested in data centers and GPUs, with the remaining money just enough for daily operations. The gap can only be filled through debt issuance: they issued $159 billion in the first half of 2026, nearly four times as much as in 2020. Even Oracle laid off 21,000 employees (13% of its workforce), and the bond market is starting to worry about their ability to repay debts (CDS prices are rising).
4. The Weakness of the Storage Industry: Demand Is “ Borrowed,” with Higher Risks Than in the Past
The storage industry experiences the greatest volatility among semiconductors due to its heavy asset-based factories and the long lead time for capacity expansion (2-3 years), which often results in changing demand when production is released. Looking at history:
- In 2008, the market collapsed due to oversupply (ten manufacturers expanded production recklessly);
- In 2018, it collapsed due to a shift in demand expectations (profit peaks);
- The risk in 2026 lies on the demand side: current storage demand does not come from actual user payments (e.g., for PCs and phones) but from funds borrowed by AI companies and giants. Once the funding dries up, demand will immediately shrink.
5. Four Indicators to Help You Determine if the Market Has Peaked
There’s no need to predict the exact time, but you can watch these four leading indicators:
1. Inventory Levels: Current DRAM inventory is only 2-4 weeks’ worth; if it rises to more than 6 weeks, it indicates that customers are starting to stockpile, suggesting a possible turnaround.
2. Contract Price Increases: Prices are increasing by 30% month-on-month; if they drop to single digits (even if they are still rising), it means buyers are beginning to resist.
3. Giant Companies’ Capital Expenditure Plans: If any giant announces “optimizing investments” (i.e., not increasing spending), it’s a signal.
4. AI Company Financing Pace: Are OpenAI and Anthropic’s IPOs going smoothly? If they underperform or are delayed, it indicates that the market’s tolerance for AI losses has reached its limit.
Conclusion: Not a Bearish Outlook, but a Warning of Risks
The performance of storage companies (e.g., Micron’s quarterly profit of $28.2 billion) is real, but the foundation of the market is “capital’s willingness to cover AI losses.” What will truly make the cycle sustainable is if AI companies can make money through services—like Anthropic’s profitability. Before that happens, every time the market accelerates, we should ask ourselves: How much longer can the funds supporting demand last?
(The full text avoids jargon and explains the underlying logic and risks of the AI storage market in plain language, making it easy for laypeople to understand.)