AI capital expenditure has exceeded the "historical record", and tens of billions of private equity firms are reflecting on the valuation dilemma under the prosperity narrative
The tide of artificial intelligence (AI) is sweeping every corner of the global economy and society, yet amid this unprecedented technological wave, there have been many sober voices from the industry and the investment community.
Following META's internal assessment that "AI agent technology is not accelerating as expected", Hexie Huiyi, a tens-of-billions-level private equity institution, recently released its latest market perspective to calmly reflect on the AI boom. They believe that the current intensity of global capital gathering into AI has approached or even surpassed that of the 2000 dot-com bubble era, and the diverging market trends may be gestating a sharp correction in the future.
The investment managers of Hexie Huiyi also pointed out that the massive AI infrastructure investments by North American Cloud Service Providers (CSPs) are rapidly depleting their cash flow, and they will need to raise funds through the capital market to cover the huge capital expenditures.
They further argued that there could be a huge gap between these North American enterprises' preemptive investments driven by corporate Fear Of Missing Out (FOMO) and the actual growth expectations of end-market revenues.
Despite remaining optimistic about AI's long-term development potential, the institution's investment managers believe that no matter how AI evolves in the future, the valuations of many stocks have seriously deviated from rationality, even reaching extremely optimistic levels. Historically, this kind of state only appears at the end of a market rally.
Hexie Huiyi has many well-known senior equity fund managers in the industry, including Lin Peng, a representative of growth investing, Han Dong, a representative of value investing, as well as Wang Yanfei and Liang Shuang. Their research on industry and company fundamentals is rigorous and in-depth. Therefore, as an independent perspective, it is worthy of prudent consideration by investors.
01
AI Investment Intensity Has Surpassed the 2000 Dot-Com Bubble
Han Dong, an investment manager of Hexie Huiyi's Vision Series, mentioned a viewpoint in his latest monthly report: the global impulse to concentrate investments into AI has pushed the current investment intensity to an extremely high level, even exceeding that of the 2000 dot-com bubble period.
Han Dong noted that from 2024 to 2025, the AI infrastructure investments of major North American CSP (Cloud Service Provider) manufacturers will exhaust their free cash flow. From 2025 to 2026, these expenditures will further consume their operating cash flow. It is estimated that by 2026 to 2027, these manufacturers will face AI capital expenditures exceeding 700 billion US dollars.
Measured by the ratio of capital expenditures to cash flow or revenue, the current investment intensity has approached or even surpassed that of the 2000 dot-com bubble era.
02
High-Intensity Preemptive Investments Bring Cash Flow Pressure
Such high-intensity preemptive investments have brought severe cash flow pressure to tech giants.
Han Dong predicts that the investment volume of 700 billion US dollars this year and over 1 trillion US dollars next year is equivalent to half of the total revenue of several largest North American CSP manufacturers or all of their operating cash flow. This has gradually exceeded the volume they can afford on their own.
As a result, this cash flow pressure will force major overseas CSPs to raise funds through the capital market to balance their cash flow.
This clearly embarks on a path of cash flow balancing with greater risks and pressures.
03
Large Model Vendors Need 20-Fold Revenue Growth to Break Even
Since they have started raising funds from the secondary market, they naturally need to present attractive asset statements or future growth plans, which seems unlikely for current large model vendors to achieve.
Han Dong believes that based on reasonable operational model assumptions for large model vendors (60%-80% gross margin, 20% depreciation-to-revenue ratio), it means that the end market needs to generate around 1 trillion US dollars in annual revenue to recover one year's investment cost within 5 years.
If the current investment intensity is maintained for a long time, the end-market revenue will need to rise to 5 trillion US dollars in the future. Considering that the end-market revenue of large model vendors will reach 30 billion US dollars in 2025, the annualized revenue volume is expected to only reach 200-300 billion US dollars by the end of 2026.
Therefore, to maintain the current intensity of AI infrastructure investment, the revenue of large model vendors in the future needs to increase by 20 times on the basis of the current revenue volume.
04
FOMO Sentiment Drives the Corporate Preemptive Investment Boom
The investment managers of Hexie Huiyi also analyzed the micro-ecosystem and concluded that in terms of revenue structure, the short-term commercial implementation of AI faces bottlenecks.
They believe that the current payers of AI show a ratio of 8:2 between enterprises and individual consumers. The main revenue driver for large model vendors is helping enterprises reduce costs and improve efficiency, rather than creating new demands.
If overseas large model vendors want to reach a revenue volume of 5 trillion US dollars in the future, it means that North American enterprises need to spend 10% of their revenue on purchasing tokens, while current IT companies' token expenditures only account for 1% to 3% of their revenue, making such a leap difficult to achieve in the short term.
Hexie Huiyi's analysis points out that there are hidden dangers in the current prosperity transmission of the entire industrial chain. Corporate FOMO sentiment towards AI has led to cost-agnostic preemptive investments, which prompts large model vendors to extrapolate revenues linearly, thereby driving real orders across the entire AI infrastructure industrial chain.
However, if these real orders stem from corporate panic rather than sustainable demand, then the current underlying assumptions about the growth rate and performance models of AI companies are worthy of careful consideration.
05
Optimism About AI Does Not Mean Recognition of Related Stock Valuations
Lin Peng, founder of Hexie Huiyi and investment manager of the Vision Series, stated that Hexie Huiyi is not bearish on AI. On the contrary, he personally is very optimistic about AI's long-term development potential.
But being optimistic about the AI industry does not mean recognizing the valuations of many related stocks at present. He believes that no matter how AI evolves in the future, the valuations of many stocks have seriously deviated from rationality, even reaching extremely optimistic levels. Not to mention that many listed companies are now riding the AI trend to hype their own stock prices. This kind of frenzied state, historically, only appears at the end of a market rally.
Wang Yanfei, investment manager of Hexie Huiyi's Yuanzhen Series, pointed out that the current AI market is essentially a cycle. Reviewing the internet investment phase from 1995 to 2000, we can find that despite strong short-term fundamentals, at some point in the future, many AI targets will inevitably enter a prolonged downward cycle. The current questions we need to consider are: How long will the continued divergence last? Can we optimize our positions during this period? And how to carry out such optimization?
Han Dong stated that although current CSP manufacturers have just started using financing leverage, it seems that even if the bubble bursts, it will not cause a strong systemic risk (historically, the peak leverage ratios during the Grand Canal, railway infrastructure, and dot-com bubble periods all reached around 100%). However, from a responsible perspective, we cannot bet on the probability of an extreme situation evolving into something even more extreme to seek gains amid dangers.
This article is from the WeChat public account "Capital Deep Dive", written by Jiaer, edited by Yuan Chang, and published with authorization from 36Kr.