An Extended Study Based on Six Charts: Structural Divergence in Valuation, Earnings, and Capital Expenditure
- lx2158
- Aug 23
- 32 min read
Updated: Aug 27
I. Figure 1 (P/E Box Plot Distribution): The S&P 500 at 22x is Above the 95th Percentile of its 20-Year Range
Chart Highlights: The current Next Twelve Months (NTM) Price-to-Earnings (P/E) ratio for the S&P 500 is approximately 22x (red box), placing it above the >95th percentile of its distribution over the past 20 years. The Nasdaq 100's valuation has reached 27x, also an extremely high level. The Russell 2000 small-cap index stands at 24x, with a wider box plot, reflecting greater historical volatility in its valuation. In contrast, the Equal-Weight S&P 500 and the S&P Midcap 400, which better represent the market's core, are valued at a more reasonable 17x and 16x, respectively. The box plot clearly displays the 75th–25th percentile range (the box), the 95th–5th percentile range (the whiskers), and the 20-year median as a long-term benchmark (the gray line).
Interpreting Valuation:
Standard Definition and Implied Return: The basic formula for the P/E ratio is P/E = Price / EPS_NTM. The core of this ratio is the price the market is willing to pay for each unit of earnings over the next year. In a simplified, static expectation where the valuation multiple is assumed to remain constant, the nominal expected return for an investor is approximately the inverse of the P/E ratio, known as the Earnings Yield: E/P = 1 / (P/E). Based on the current S&P 500 P/E of 22x, the earnings yield is approximately 1/22 ≈ 4.55%. It must be emphasized that this 4.55% is merely the return at the earnings level; it is by no means equal to the total return a shareholder can expect. It serves more as a baseline, measuring a company's earnings generation capacity relative to its market value, without considering growth or changes in valuation.
A Deeper Analysis from a "Return Decomposition" Perspective: A more complete and dynamic framework for expected returns breaks down the total shareholder return into three core components: Expected Total Return ≈ Earnings Yield + g + Δ(PE).
Earnings Yield (E/P): This is the 4.55% we calculated, representing the "cornerstone" or "coupon" component of the return.
g (Long-term Sustainable EPS Growth): This is the "growth engine" of returns. This g should not be short-term cyclical earnings volatility but rather a structural growth rate reflecting a company's long-term competitive advantages, industry structure, macroeconomic growth, and return on capital. In the current environment, the market's expectation for g, especially for a few mega-cap companies, is a key pillar supporting high valuations. However, if the central tendency of macroeconomic growth declines or industry competition intensifies, the realization of g will face significant challenges.
Δ(PE) (Change in Valuation Multiple): This is the "sentiment" or "cyclical" component of returns, reflecting changes in market risk appetite, the interest rate environment, and growth expectations. When Δ(PE) > 0, a valuation expansion (rerating) occurs, leading to excess returns. When Δ(PE) < 0, a valuation compression (derating) occurs, eroding total returns. Given that the current P/E of 22x is at a >95th percentile historical high, from the perspective of probability and mean reversion, the likelihood of a negative Δ(PE) in the future is far greater than a positive one.
The Profound Meaning of Historical Percentiles:
The Gravitational Pull of Mean Reversion: When an index's valuation level, such as the S&P 500's 22x P/E, is above the 95% percentile of its own 20-year historical distribution, it means the price the market is currently paying for a dollar of earnings has significantly surpassed the levels of the vast majority of the past two decades. This in itself is a warning signal. We can roughly depict the dynamic changes of the valuation multiple using a stochastic process, such as the Ornstein-Uhlenbeck process: d(PE)_t = θ(μ - PE_t)dt + σ dW_t. In this model, μ is the long-term mean (or median) of the valuation, PE_t is the current valuation, and θ is the speed of reversion. When PE_t is far above μ (as is the case now), the model's drift term θ(μ - PE_t) is significantly negative. This mathematically describes a powerful "gravitational pull"—valuation tends to revert to its long-term mean, thereby creating a risk of potential valuation compression. Historical data has repeatedly shown that investing from extreme valuation highs often leads to disappointing long-term returns.
The Essence of Structural Divergence:
Quantifying the "Mega-Cap Lift Effect": A vast chasm exists between the market-cap-weighted S&P 500 (22x) and the equal-weight S&P 500 (17x). This 5x valuation premium (Premium ≈ 22x - 17x) precisely quantifies the market's fragmentation. It means that if we view the market as a whole, a small number of mega-cap tech giants, with their exorbitant valuations, have significantly pulled up the average level of the entire index. For the same dollar of expected earnings, the price the market assigns to these giants is about (22-17)/17 ≈ 29% higher than the price assigned to other average companies in the market. Behind this phenomenon lies the market's extremely optimistic pricing of these giants' near-monopolistic growth prospects and profitability in fields like artificial intelligence. However, this also implies a decline in market health, with the overall rally driven by a few stocks, while the majority of stocks have not shared in the prosperity to the same degree.
Risk Characteristics of Different Indices: The 27x valuation of the Nasdaq 100 reflects its high concentration of technology stocks. The 24x valuation of the Russell 2000 is more complex; its high valuation is partly due to the inclusion of a large number of small-cap companies that are not yet profitable but have high growth expectations (the denominator, EPS, is small or even negative). Therefore, interpreting its valuation requires more caution and should be combined with other metrics like EV/Sales. Meanwhile, the 16x valuation of the S&P Midcap 400 shows that the valuation of the mid-sized, mature corporate sector is more conservative and closer to historical norms.
A Simplified Derivation and Extension of Valuation-Interest Rate Sensitivity:
Application of the Gordon Growth Model: In the simplified form of the classic Dividend Discount Model (DDM)—the Gordon Growth Model—the stock price is P = D1 / (r - g), where D1 is the expected dividend in the next period, r is the discount rate (or required rate of return), and g is the perpetual growth rate. If we substitute earnings E for dividends D and introduce a payout ratio payout (D = E * payout), we can derive P/E ≈ payout / (r - g). This formula intuitively reveals the relationship between valuation and interest rates.
The Transmission Mechanism of Interest Rates: The discount rate r can typically be broken down into the risk-free rate r_f (such as the 10-year Treasury yield) plus an equity risk premium ERP. Therefore, P/E ≈ payout / (r_f + ERP - g). When the risk-free rate r_f rises, even if all other variables (payout, ERP, g) remain constant, the denominator of the formula will increase, leading to a downward trend in the P/E valuation. This explains why high-valuation growth stocks are usually hit harder during interest rate hike cycles or in high-interest-rate environments. Their future earnings growth needs to be discounted at a higher rate, thus reducing their present value.
Direct Implications for Operations:
The Urgency of Finding a "Safety Cushion": In a macro combination of "high historical percentile valuation + an interest rate environment that has not clearly turned dovish," relying solely on a 4.55% earnings yield (E/P) is far from sufficient. Investors must hope for two rather fragile assumptions to hold true simultaneously: 1) companies can achieve sufficiently high real earnings growth g to offset the impact of high valuations; and 2) there will be no significant valuation compression in the future, i.e., Δ(PE) will be at least zero or close to it. In a period of high macroeconomic uncertainty and escalating geopolitical risks, both of these assumptions face severe tests. Therefore, the focus of investment strategy must shift from purely chasing growth (Beta) to selecting individual stocks (Alpha), emphasizing valuation discipline, and seeking out companies with high growth certainty, robust financial health, and valuations that have not yet excessively priced in future potential.
II. Figure 2 (Historical Decomposition of the Share of Stocks with EV/Sales > 10x): The Third Emergence of ~20% of Constituent Stocks Being "Hyper-Expensive"
Chart Highlights: This time-series chart, stacked by industry, reveals a startling phenomenon: currently, the proportion of companies in the S&P 500 index with an Enterprise Value-to-Sales (EV/Sales) ratio exceeding 10x has reached approximately 20%. This is the third time in history that this warning level has been reached or surpassed. The previous two occasions were at the peak of the dot-com bubble in 2000 and during the tech stock frenzy of 2021, driven by zero interest rates and the mobile internet wave. In terms of industry composition, Information Technology (Info Tech) and Communication Services (Comm Svcs) are the main contributors to this extreme valuation phenomenon, which is highly consistent with the current market narrative driven by Artificial Intelligence (AI).
Why a 10x Sales Multiple is a Dangerous Threshold:
Reverse-Engineering Implied Assumptions from the Valuation Formula: The full definition of EV/Sales is (Market Capitalization + Net Debt) / Revenue. It measures how much total enterprise value the market is willing to pay for one unit of sales revenue. Unlike P/E, it is also applicable to growth companies that are unprofitable or have unstable earnings. What kind of extreme expectations are hidden behind an EV/Sales > 10x valuation? We can reverse-engineer this through a simplified steady-state free cash flow model. In a perpetual growth state, EV ≈ FCF / (r - g), and free cash flow FCF can be expressed as Sales × m × (1-τ) × (1-ρ), where m is the pre-tax profit margin, τ is the tax rate, ρ is the reinvestment rate, and g = ρ × ROIC. After simplification, we get EV/Sales ≈ m × (1-τ) / (r - g).
A "Sobering" Calculation: Let's plug in a set of long-term assumptions that are relatively optimistic but still reasonable: a top-tier tech company's steady-state operating profit margin m = 20%, an effective tax rate τ = 15%, and a discount rate minus growth rate spread of r - g = 5% (which already implies decent growth or low risk). In this scenario, its reasonable EV/Sales multiple would be approximately 0.20 × (1 - 0.15) / 0.05 = 3.4x. To support an EV/Sales of 10x or even higher, what kind of extreme assumptions are needed? Keeping other variables constant, the profit margin m would need to reach (10 × 0.05) / 0.85 ≈ 58.8%, or the r-g spread would need to be compressed to (0.20 × 0.85) / 10 = 1.7%. The former implies that the company needs to have mythical pricing power and cost control, almost unaffected by competition; the latter means the market expects the company to grow perpetually at an extremely high rate, or that the required risk compensation is extremely low. Both scenarios are extremely rare and difficult to sustain in the real business world.
The Risk of Non-Linear Declines: Therefore, an EV/Sales > 10x valuation is essentially a pricing of "belief." It prices not reality, but the expectation of the best possible future. Once the actual operating conditions (such as slowing growth, margins failing to meet expectations, or increased competition) deviate from this extreme expectation, the decline in the valuation multiple will be sharp and non-linear. The drop in stock price will far exceed the downward revision of earnings, as the negative contribution of Δ(PE) will be amplified.
Structural Similarities and Differences of the Three "Bubbles":
2000 vs. 2021/2025: In the 2000 dot-com bubble, the group of companies with EV/Sales > 10x was mainly composed of "dot-com" companies that had only a business concept, lacked a clear profit model, or even had no revenue. It was a frenzy about "eyeballs" and "click-through rates," which ultimately proved to be a castle built on sand. In the current high-valuation wave from 2021 to the present, the main players have changed. Many of the companies in the high-valuation range, such as the "Magnificent Seven," are platform-based enterprises with strong moats, high profit margins, massive cash flows, and mature business models. Their "expensiveness" seems to have a more solid earnings foundation. However, the danger lies in the "valuation spillover effect"—under the halo of the leading companies, many smaller and medium-sized companies of far inferior quality and with unproven business models are being assigned similar extreme multiples by the market, simply because they have been labeled with "AI," "cloud computing," or "SaaS." This segment of companies is the most fragile link in the current market structure.
Practical Rules for Investment Strategy:
The "Triple-Constraint Test" Framework: For any investment target with an EV/Sales > 10x, it should be treated like a high-risk experiment and subjected to a rigorous "Triple-Constraint Test":
Feasibility Analysis of the Five-Year Target Profit Margin (m*): Is it truly possible for this company to achieve and sustain the ultra-high profit margins required to support its valuation within five years? Do its industry structure, competitive landscape, technological barriers, and pricing power support this judgment? A detailed Porter's Five Forces analysis is required.
Scrutiny of the Free Cash Flow Path (FCF): High growth often requires high-intensity reinvestment. How large are the capital expenditures (Capex) and net working capital (ΔNWC) investments required for this company to achieve its growth targets? Is its business model asset-light, or is it an asset-heavy model that requires continuous "cash burning"? Is the path of free cash flow, FCF = EBIT(1-τ) - ΔNWC - Capex, healthy? The quality of growth of a company that cannot generate positive free cash flow is questionable.
Vulnerability Assessment of Discount Rate Elasticity (∂P/∂r): As mentioned earlier, high-valuation growth stocks are essentially long-duration assets. They are extremely sensitive to changes in interest rates (the discount rate). It is necessary to assess how much their valuation could be impacted under different interest rate scenarios.
Building a Screener for Cross-Sectional Risk Mitigation: Investors can build a three-dimensional screening model of "extreme multiple—profit margin—free cash flow." For example, one could resolutely exclude companies with EV/Sales > 10x but at the same time have FCF/Sales < 5% or a declining Gross Margin. Through such screening, companies that rely solely on "stories" and "concepts" to support high valuations can be filtered out, and the remainder can serve as a watch list of "potentially true high-quality growth stocks" that require continuous and close monitoring.
III. Figure 3 (1900–2025: S&P 500 Dividend Yield vs. 10Y Treasury Yield): Relative Attractiveness and a Re-examination of the "Fed Model"
Chart Highlights: This magnificent chart, spanning more than a century, illustrates the long-term relationship between the S&P 500's Dividend Yield and the 10-Year U.S. Treasury Yield. A significant historical turning point occurred in the late 1950s to early 1960s, after which the 10-year Treasury yield has been higher than the S&P 500's dividend yield for the vast majority of the time. In recent years, as the global inflation hub has risen and central banks have tightened monetary policy, U.S. Treasury yields have climbed sharply, causing the spread between the two (Bond Yield - Dividend Yield) to become significantly positive again, reaching a multi-decade high. The long-term downward trend in the dividend yield is also closely related to the evolution of corporate financial strategies in the post-World War II era, particularly the rise of stock buybacks.
A Critical Analysis of Three "Relative Valuation" Metrics:
The Misuse of the Fed Model: The Fed Model is a widely circulated but highly controversial valuation tool. Its core idea is to compare the stock market's earnings yield (E/P) with the 10-year Treasury yield (10Y). When E/P > 10Y, the stock market is considered undervalued relative to the bond market. However, this model has several fatal theoretical flaws:
Conflation of Real vs. Nominal: Earnings are a "real" concept, as they tend to grow with inflation, whereas bond yields are a "nominal" concept, with fixed coupon payments. Directly comparing the two in different inflationary environments is like using a ruler to measure weight—the units are wrong.
Absence of a Growth Component: The model completely ignores the core attraction of stocks—the long-term growth of earnings (g). Stocks are perpetual growth assets, while bonds are assets that repay principal and interest at maturity.
Neglect of the Risk Premium: Stocks are inherently riskier than government bonds, so investors should demand an additional risk compensation, the equity risk premium (ERP). The Fed Model implicitly assumes that the ERP is a constant, which is severely at odds with reality. Therefore, simply comparing E/P and 10Y can easily lead to the erroneous conclusion that "stocks are cheap" in certain market environments (especially during periods of high inflation).
The Correct Calculation of the Equity Risk Premium (ERP): A more rigorous framework is to directly estimate the ERP. The core idea is that the expected total return on equity should equal the risk-free rate plus a risk premium: Expected Equity Return = r_f + ERP. The expected total return on equity can, in turn, be broken down into E/P + g - Δshares, where Δshares captures the net effect of stock buybacks (reducing shares, boosting EPS) or issuances (diluting equity). Therefore, ERP ≈ (E/P + g - Δshares) - r_f.
Estimating the Current Market's ERP: Let's perform a rough calculation using current data. The S&P 500's E/P ≈ 4.5%. For long-term sustainable earnings growth g, a reasonable estimate might be between 3-4% (close to nominal GDP growth). The boost from net buybacks (-Δshares) has been around 1-2% in recent years. The risk-free rate r_f (10-year Treasury yield) is around 4.2-4.5%. Thus, ERP ≈ (4.5% + 3.5% + 1.5%) - 4.5% ≈ 5.0%. This level is not particularly low compared to the long-term historical average, but given the high degree of current macroeconomic uncertainty and the significant question of whether earnings growth g can be realized, it is debatable whether this risk premium level is sufficient to compensate investors for the risks they are taking. Once expectations for g are revised downwards, or if the market experiences valuation compression (Δ(PE) < 0), the attractiveness of equities relative to safe government bonds will rapidly decline.
The Dividend Yield Gap: The DY - 10Y metric, though simple, has a very intuitive meaning for certain types of investors (such as pension funds and retirees who rely on dividend income). This gap is currently significantly negative, which means that from an "immediate cash flow" perspective, the stable coupon income that can be obtained from investing in risk-free government bonds is already far higher than the average dividend income that can be obtained from investing in the entire stock market. This completely invalidates the "TINA" (There Is No Alternative) logic. For risk-averse or income-oriented investors, the appeal of fixed-income assets like government bonds and high-grade credit bonds is returning.
An Intuitive Understanding and Application of the "Equity Duration" Concept:
The Analogy of Duration: In bond analysis, "duration" measures the sensitivity of a bond's price to changes in interest rates. We can apply this concept by analogy to stocks. Using the same formula P = CF1 / (r - g), we can derive that the sensitivity of a stock's price to the discount rate r is approximately ∂ln(P) / ∂r ≈ -1 / (r - g). This -1 / (r - g) can be thought of as "equity duration."
The Long-Duration Characteristic of the Current Market: Assuming the current market's r - g (the difference between the market's required risk premium and growth expectations) is 3%, then the equity duration is approximately 1 / 0.03 ≈ 33 years. This is a very long duration. It means that, all else being equal, if the market's required discount rate r increases by 1 basis point (0.01%), the stock price would theoretically fall by about 0.33%. This long-duration characteristic explains why the stock prices of growth stocks, especially those that rely on distant future cash flows, fluctuate so dramatically when interest rate expectations change. The current high valuation of the market itself implies that its implicit duration is very long, and therefore its exposure to the risks of rising interest rates or a downward revision of growth g is also exceptionally large.
Strategic Implications at the Asset Allocation Level:
Hedging "Valuation Duration Risk": When the 10-year Treasury yield, as a key risk-free rate benchmark, remains persistently high, and the stock market's dividend yield is at a historical low, it means that the "valuation duration risk" of equity assets is very prominent. In asset allocation, it is necessary to introduce tools that can hedge this risk.
Increasing Short-Duration Assets: Within an equity portfolio, one can moderately increase the weight of value-oriented or high-dividend companies that have "more certain cash flows, stable liability structures, and more reasonable valuations." The cash flows of these companies are closer to the present, and their duration is relatively shorter.
Allocating to High-Quality Credit: In cross-asset allocation, shorter-duration, high-quality investment-grade (IG) credit has become a very attractive option. They can provide coupon income similar to current Treasury yields while having manageable credit risk, and can serve as a stabilizer to hedge the volatility of "long-duration" equity assets.
IV. Figure 4 (MAG7 vs. S&P 500 vs. S&P 493: Net Income Growth Profile): A Quantitative Portrayal of Growth "Unipolarization"
Chart Highlights: This bar chart quantifies, in a truly stunning way, the extreme degree of "growth unipolarization" in the current U.S. stock market. Since 2023, the "MAG7" (Magnificent Seven) cohort of tech giants has maintained an astonishing net income growth rate of ~20% to ~60% over multiple quarters. In stark contrast, the "S&P 493," which excludes these seven companies, is projected to have a net income growth rate of only ~2-3% for the second, third, and fourth quarters of 2024. This growth rate is not only far below that of the MAG7 but is also lower than the expected inflation rate for the same period, meaning that the real profits (adjusted for inflation) of the vast majority of companies in the market are actually shrinking. The growth of the entire S&P 500 index is almost entirely contributed by these seven giants.
A Three-Tiered Deconstruction of Profit Drivers:
An Accounting Identity Perspective: The growth of earnings per share (EPS) can be broken down into three levels of drivers: ΔEPS ≈ ΔSales × Operating Leverage × Financial Leverage - Share Dilution.
ΔSales (Sales Growth): This is the source of growth.
Operating Leverage: This measures the elasticity of profit growth relative to sales growth and is mainly affected by the cost structure (fixed costs vs. variable costs) and gross margin.
Financial Leverage: This measures the elasticity of net income growth relative to operating income growth and is mainly affected by interest expenses.
Share Dilution: This reflects the accretive or dilutive effect of stock buybacks or issuances on earnings per share.
The Power Sources of "MAG7's" High Growth:
(i) Strong Revenue Growth: The sectors these giants operate in (cloud computing, digital advertising, e-commerce, high-end consumer electronics) continue to benefit from structural tailwinds, allowing their revenue to maintain high single-digit or even double-digit growth.
(ii) Operating Leverage Boost from AI: The explosive growth of artificial intelligence workloads has greatly propelled their high-margin cloud services and software businesses. These businesses have significant economies of scale; once the initial infrastructure is in place, the marginal cost of serving more customers is extremely low, leading to powerful operating leverage that allows profit growth to far outpace revenue growth.
(iii) The Accretive Effect of Massive Buybacks: Abundant cash flow enables these companies to continuously conduct large-scale stock buybacks, constantly reducing the number of outstanding shares. This further boosts earnings per share (EPS) by shrinking the denominator (number of shares) while the numerator (net income) is growing.
The Predicament Facing the "S&P 493":
(i) Anemic Revenue Growth: Most companies in traditional industries are facing pressure from a slowing macroeconomy and sluggish demand, leading to weak revenue growth.
(ii) The Double Squeeze of Costs and Interest: They are more susceptible to the impact of rising raw material prices and rigid labor costs, leading to negative operating leverage. At the same time, in a higher interest rate environment, the interest expenses of many highly leveraged companies have risen sharply, further eroding net profits. This creates a "pincer attack" on profit margins.
The Severe Mismatch Between Valuation and Growth:
The "493's" Implied Return is Too Low: If we accept the overall valuation of the "S&P 493" (we can refer to the 17x P/E of the equal-weight S&P 500) and combine it with its projected earnings growth of only 2-3%, then, without considering changes in valuation, its nominal expected return would be approximately Earnings Yield + Growth Rate ≈ 1/17 + 2.5% ≈ 5.9% + 2.5% = 8.4%.
The Risk Premium is Too Thin: This 8.4% expected return, when compared to the ~4.5% risk-free return offered by 10-year Treasuries and the ~5.5% return offered by investment-grade corporate bonds, provides a very thin equity risk premium. Once the macroeconomy experiences an unexpected downturn, causing this 2-3% growth to not materialize, or if market sentiment reverses, leading to valuation compression (Δ(PE) < 0), then the nominal rate of return on investing in the broad "S&P 493" companies could quickly fall to 5-6% or even lower, making it almost unattractive.
Potential Triggers for a Style Rotation:
From "A Single Flower Blooming" to "A Hundred Flowers Blossoming"? The current "growth unipolarization" market pattern has greatly diminished the effectiveness of traditional investment strategies that rely on macroeconomic beta, while the importance of bottom-up stock picking (Alpha) has been elevated to an unprecedented height. For the market style to shift from the current state of extreme divergence to a healthier rally driven by a broader range of companies, one of two typical paths usually needs to be triggered:
(i) Marginal Improvement in the Macroeconomy: If positive macro signals emerge, such as inflation receding, central banks beginning to cut interest rates, and the economy achieving a "soft landing," then the earnings expectations g of the "non-giant" companies that are more sensitive to the economic cycle and are currently struggling are likely to be revised upwards. This would improve their relative attractiveness and could trigger a flow of funds from the overcrowded giants to the broader value and cyclical stocks.
(ii) A Faltering of the Giants' Own Logic: Another possible trigger comes from within the giants themselves. If the return on investment (ROIC) from their current massive capital expenditure cycle (which we will discuss in detail in the next section) falls short of the market's optimistic expectations, or if the path to commercializing AI technology proves to be slower and more tortuous than anticipated, then the "faith" supporting their high valuations could be shaken, triggering a valuation compression of Δ(PE) < 0. This negative sentiment could spread, prompting investors to reassess risk and seek out companies with lower valuations and more solid fundamentals.
The Demands on Investors: Regardless of which path ultimately materializes, it places extremely high demands on investors. Before path (i) occurs, it is necessary to conduct rigorous screening of the companies in the "S&P 493" for earnings quality and financial resilience to identify the survivors most likely to benefit first in an economic recovery. Under the risk of path (ii), it is necessary to closely track the capital expenditure efficiency of the giants and the real progress of their AI businesses to guard against the risk of a valuation bubble bursting.
V. Figure 5 (Capex/OCF: Big 7 Converging with "The Rest"): A Transition from an Asset-Light Model to a Compute-Heavy, Asset-Intensive One
Chart Highlights: This chart reveals a fundamental shift in the business model of tech giants. Historically, the Big 7 Tech companies were considered paragons of the "asset-light" (capex-light) model, with their ratio of capital expenditures to operating cash flow (Capex / Operating Cash Flow) being far lower than that of other companies in the market (The Rest). However, since 2022, the situation has changed dramatically. The Capex/OCF ratio for the Big 7 has been climbing rapidly and is projected to reach about 45% by mid-2025, nearly converging with the 45-47% level of "The Rest." This signals that, driven by the wave of artificial intelligence, these once "asset-light" tech companies are rapidly transforming into "asset-heavy" providers of computing infrastructure.
Changes in the Cash Flow Equation and Free Cash Flow Elasticity:
Direct Pressure on FCF: A company's Free Cash Flow (FCF) is the core source of its value, and its basic formula is FCF = OCF - Capex - ΔNWC. It is clear from this formula that when the Capex/OCF ratio rises sharply, if OCF (operating cash flow) does not grow at a proportional and similar pace, it will inevitably lead to a squeeze on FCF (free cash flow). ΔFCF ≈ -ΔCapex.
Impact on Valuation: The Free Cash Flow Yield (FCF Yield = FCF / EV) is a key metric for measuring the attractiveness of a company's valuation. When FCF declines temporarily due to a surge in capital expenditures, the FCF Yield will also fall. If the market still maintains high valuations for these companies, it means the market believes that the current massive investments will bring higher and more sustainable cash flow growth in the future, thus compensating for the short-term sacrifice. This is a typical strategy of "trading today's cash flow for tomorrow's growth." However, the success or failure of this strategy depends entirely on the efficiency of the investment.
The Core Test of Investment Returns: ROIC vs. WACC
The Iron Law of Value Creation: No matter how appealing the story of an investment is, its ultimate ability to create value for shareholders depends on a simple standard: whether the Return on Invested Capital (ROIC) is higher than its Weighted Average Cost of Capital (WACC).
ROIC = NOPAT / Invested Capital (where NOPAT = Net Operating Profit After Tax)
If ROIC > WACC, then growth creates value.
If ROIC < WACC, then growth is actually destroying value. The more capital the company invests, the more value it destroys.
The Ultimate Question for AI Infrastructure Investment: The current investment of hundreds of billions of dollars by tech giants to build AI infrastructure (data centers, GPU clusters, etc.) is essentially a huge gamble. The reason the market gives them high valuations is the belief that these investments can build an impregnable economic moat, thus ensuring that ROIC - WACC > 0 for a long time to come. However, this is not a foregone conclusion. The success or failure of this investment race is the key determinant of whether the current market pricing is reasonable.
The Economics of Computing Infrastructure:
Deconstructing the Drivers of ROIC: The ROIC of AI infrastructure investment is affected by a series of complex and dynamic factors. We can roughly break it down as: ROIC ≈ (Price_per_compute_unit - Variable_cost_per_unit) × Utilization_rate × Operating_hours / (Total_Capex × Depreciation_factor).
Price: In the early stages of AI development, computing power is scarce and prices are high. But as major giants and numerous startups enter the field, the supply side is expanding at an unprecedented rate. This will almost inevitably lead to fierce price wars in the future.
Utilization: What percentage of the time are the massively expensive computing centers actually being used effectively and generating revenue? This depends on the speed at which killer AI applications emerge and become widespread. If innovation on the application side cannot keep up with the pace of infrastructure construction, a large number of GPUs may be idle, which would severely drag down ROIC.
Cost: In addition to the procurement cost of GPUs, the operating costs of data centers (especially electricity and cooling costs) are also extremely high and vary with fluctuations in energy prices.
The Potential Cyclical Trap: Any capital-intensive industry is prone to falling into a "boom-bust" cyclical trap. In the current AI gold rush, major manufacturers are scrambling to expand production capacity, which may lead to a temporary oversupply of computing power in the coming years. Oversupply will trigger price wars, which will lead to a decline in the overall industry ROIC. A decline in ROIC will, in turn, curb the willingness to make new capital expenditures, until the next round of technological breakthroughs or demand explosion.
Bridging from Accounting Profit to Economic Profit:
The Impact of R&D Capitalization: When assessing the true profitability and return on investment of these tech companies, it is also necessary to be aware of the distortions that may be caused by accounting methods. For example, for R&D investment in large models, some companies may expense it directly against current profits, while others may capitalize a portion of it as an intangible asset to be amortized over the next few years.
Making an "Economic Adjustment": To assess more accurately, analysts need to make an "economic adjustment" for R&D expenditures (R&D Capitalization). This involves adding back the currently expensed R&D to profits, while subtracting the economic amortization of the past accumulated R&D capital stock. Only the adjusted Adj_EBIT and Adj_ROIC can more truly reflect the company's earnings quality and capital efficiency, avoiding overestimation or underestimation of its intrinsic value due to differences in accounting choices.
VI. Figure 6 (GAMA's Capex This Year is Approx. $317 Billion ≈ 1% of U.S. GDP): The Macro Scale of the AI Arms Race
Chart Highlights: This chart, in absolute terms, further highlights the unprecedented scale of current AI investment. The combined capital expenditures of just four companies—Alphabet, Amazon, Microsoft, and Meta (GAMA)—are projected to reach a staggering ~$317 billion this fiscal year. This figure is not only astronomical, but more importantly, it has macroeconomic significance—it is approximately equal to ~1% of the U.S. nominal GDP.
Dimensional Verification and Macro Impact:
A Startling Proportion: We can do a simple dimensional check. Assuming the U.S. nominal GDP in 2025 is about $31.7 trillion, then Share_of_GDP ≈ $317 billion / $31.7 trillion ≈ 1.0%. It is extremely rare in peacetime history for the investment in a single industry chain (AI infrastructure) formed by just four companies to account for 1% of the GDP of the world's largest economy. The last time a similar situation occurred might have been during the railway era or the rise of the automobile industry.
Macro-level Misallocation Risks and Spillover Effects: Such a huge and highly concentrated investment is bound to have profound macro impacts:
Resource Crowding-Out Effect: This "arms race" is frantically absorbing various resources, including capital, high-end talent (especially chip and AI engineers), electricity, water resources (for cooling), land (for building data centers), and related infrastructure (such as substations). This may create a "crowding-out effect" on other economic sectors, raising the cost of production factors for the entire society.
Energy Bottlenecks: The energy consumption of AI data centers is notoriously huge. It is estimated that by 2026, the electricity consumption of global data centers could be equivalent to that of Japan. This puts enormous pressure on the already strained global power system and could become a key bottleneck constraining the development of AI.
Geopolitical Dimension: The competition for advanced process chips (like GPUs) has evolved into the core of the great power tech rivalry. Any disruption to the supply chain (for example, due to geopolitical conflicts) could deal a fatal blow to this investment race.
"A One-off Investment Peak" or "A New Normal of High Capital Intensity"?
Maintenance vs. Growth Capex: A company's capital expenditure (Capex) can be divided into two categories: "Maintenance Capex," which is needed to maintain existing operations, and "Growth Capex," which is used to expand and develop new businesses. The key question is, of the current ~$317 billion investment, what proportion is a one-time "Growth Capex" aimed at seizing the first-mover advantage in the AI era, which will decline in the future? And what proportion will be converted into a higher level of "Maintenance Capex" that must be continuously invested every year in the future?
The Structural Rise of the κ Hub: We can set a "Maintenance Capex as a percentage of OCF" ratio, κ. Historically, the κ value for tech giants has been low. But if AI infrastructure needs to be upgraded every few years (for example, a new generation of GPUs every 2-3 years), then high capital expenditures may become the norm in order to maintain a leading position in computing power. Judging from the trend of the Capex/OCF ratio of the Big 7 converging with that of "The Rest" in Figure 5, a reasonable guess is that the long-term κ hub for these tech giants may structurally rise from the historical range of 10-20% to the 40-45% region.
Sustained Pressure on Future Free Cash Flow: If the κ hub does indeed rise systemically, it will have a profound impact on the company's long-term valuation model. It will continuously suppress the company's ability to generate free cash flow (FCF), even if its operating cash flow (OCF) is still growing. This will make the company's valuation unprecedentedly dependent on ROIC. Only by continuously proving that its high investments can generate excess returns can the market tolerate its lower FCF Yield.
Scenario Analysis:
FCF Growth Path Projection: Assume that a BigTech company's future OCF compound annual growth rate (CAGR) falls back to a more normal level of 8-10%.
Scenario 1 (Optimistic): If the Capex/OCF ratio stabilizes at 40% and does not rise further (Δ(Capex/OCF) = 0), then its FCF CAGR will be roughly in line with its OCF CAGR, also at 8-10%.
Scenario 2 (Pessimistic): If the Capex/OCF ratio continues to rise in the coming years (Δ(Capex/OCF) > 0), then its FCF CAGR will be significantly lower than its OCF CAGR. For example, if OCF grows by 10%, but Capex grows by 15%, the growth of FCF will be very slow or even negative.
Second-Order Effects and Risks in the Supply Chain:
The Migration of Cash Flow: This epic level of capital expenditure will undoubtedly cause cash flow to migrate from the downstream tech giants to the upstream and midstream suppliers. This will be a very certain investment theme in the coming years.
Upstream Core Beneficiaries: GPU manufacturers (like Nvidia), optical module/optical chip companies, high-speed copper/fiber optic cable suppliers, power semiconductor manufacturers, and transformer equipment companies.
Midstream Beneficiaries: Data center real estate investment trusts (IDC REITs), and photovoltaic/wind power operators that provide green electricity for data centers.
The "Bullwhip Effect" and Cyclical Oscillations: However, companies in the supply chain also face huge risks. This highly concentrated demand could lead to the "Bullwhip Effect." Upstream manufacturers, upon seeing huge orders, frantically expand production capacity ΔCapacity. But once downstream demand (for example, the implementation of AI applications) falls short of expectations, or if the giants cut back on capital expenditures, these upstream manufacturers will immediately face overcapacity, a sharp drop in prices ↓Price, and inventory accumulation, leading to a sharp decline in their ROIC ↓ROIC, and ultimately triggering a painful destocking cycle ↓Capex.
VII. A Strategic Framework Combining Cross-Sectional Information: The Trilemma of High Valuations, Divergent Growth, and the Interest Rate Hurdle
After integrating the structural features of valuation, growth, interest rates, and capital expenditure revealed by the first six charts, we are faced with a trilemma composed of a "valuation hurdle, growth divergence, and interest rate constraint." Any effective investment strategy framework must address all three of these challenges simultaneously.
A. Quantifying the Valuation Hurdle and "Margin of Safety":
Valuation Discipline for Individual Stocks: When the overall market valuation is high, valuation discipline for individual stocks becomes crucial. Traditional relative valuation methods (like P/E) can easily fail in a bubble, and it is necessary to return to more rigorous absolute valuation methods, such as the Discounted Cash Flow (DCF) model.
Two-Stage DCF Model: We can construct a two-stage model of a "high-growth period + perpetual growth period." First, set a clear high-growth period (e.g., T = 5 or 10 years) and make prudent forecasts for the growth rate g1, profit margin, capital expenditure, and ROIC during this period, ensuring that ROIC > WACC. Then, assume that after T years, the company enters a stable perpetual growth phase, with a growth rate g2 that cannot exceed the long-term nominal growth rate of the macroeconomy. By discounting the expected free cash flows (FCF) of these two stages to the present, an intrinsic value P* can be obtained: P* = PV(FCF_{1..T}) + PV(Terminal Value_{T+1..∞}).
Strict Requirement for Margin of Safety: Only when the current market price Price is significantly lower than our estimated intrinsic value P* is there a sufficient "Margin of Safety" (MOS). MOS = 1 - Price / P*. In the current environment, investors should demand a higher margin of safety than usual to guard against the risk of potential valuation compression.
Systematic Screening for Portfolios: For portfolio construction, a multi-dimensional quantitative screening model can be designed to systematically avoid risks and find opportunities.
Three-Dimensional Scorecard: A three-dimensional scoring system of "valuation percentile—earnings recovery probability π—multiple compression elasticity ε" can be constructed.
Score = w1 × (1 - PE_percentile) + w2 × π - w3 × ε
PE_percentile: The percentile of the stock's current P/E in its own historical range; the lower the percentile, the higher the score.
π: A score for the probability of the company's future earnings exceeding expectations, based on financial indicators (such as profit margin trends, operating leverage, and balance sheet health) and industry sentiment.
ε: A measure of the sensitivity of the stock price to changes in the valuation multiple, ε ≈ ∂ln(P) / ∂(ΔPE), which can be estimated through historical data regression. High-valuation growth stocks typically have a higher ε value and should be penalized.
w1, w2, w3 are the weights for each item, which can be dynamically adjusted according to market conditions.
B. Macro Constraints of Interest Rates and Term Premiums:
Active Management of "Equity Duration": Since we have already recognized that the current market implies a rather long "equity duration," we should actively manage and hedge this risk in portfolio management.
Increasing "Short-Duration Cash Flow" Assets: When future interest rates are expected to remain high (higher for longer), or the risk of interest rate volatility (right-tail risk) increases, the weight of assets with cash flows closer to the present and shorter durations should be increased in the portfolio. This includes high-dividend, low-valuation value stocks, as well as industries with shorter capital recovery cycles (such as certain consumer goods or industrial sectors).
Using Fixed Income for Hedging: At the cross-asset allocation level, bonds can be used to hedge interest rate risk. In particular, investment-grade (IG) or government agency (Agency) bonds with maturities in the middle of the yield curve (e.g., 3-7 years) can provide stable coupon income while effectively hedging the impact of a rise in the risk-free rate Δr on long-duration equity assets.
C. Testing for Consistency Between Growth and Capital Expenditure:
Testing the Sustainability of Growth: A company's sustainable internal growth rate is theoretically constrained by its Return on Invested Capital (ROIC) and its Reinvestment Rate: Sustainable g ≈ ROIC × Reinvestment Rate.
Beware of "Inefficient Growth": The current "AI Capex arms race" among tech giants has greatly pushed up their reinvestment rates. However, if these massive investments ultimately fail to lead to a corresponding increase in ROIC, or even cause ROIC to decline (for example, due to excessive competition or a wrong technological path), then the sustainability of their growth will be severely overestimated. This kind of growth, which "increases revenue but not profit" or even "increases revenue at the expense of profit," is value-destructive.
Key Screening Indicators: Investors need to establish a tracking system to closely monitor the investment efficiency of these high-capex companies. Key indicators should include:
Marginal ROIC: ΔROIC / ΔCapex, which measures the incremental return generated by each additional unit of capital invested.
Operational Metrics: GPU utilization rate U, the slope of the unit compute price curve, and customer net dollar retention (especially for AI-related businesses).
Cost Control: The proportion of long-term electricity/cooling costs that are locked in, which reflects the company's ability to manage the risk of future operating cost volatility.
D. Finding Alpha in the Forgotten "493":
Opportunities for Contrarian Investing: In a market dominated by a few giants, the vast majority of companies (the S&P 493) receive less attention and have underperformed, which provides potential contrarian investment opportunities for bottom-up stock pickers.
Using "Quality" as the Core Screening Axis: In this group with a nominal growth rate of only 2-3%, the core screening criteria should not be "growth," but rather "quality" and "resilience."
Potential for Margin Recovery: Look for companies that have shown signs of stabilizing and recovering profit margins during an industry downturn through cost control, efficiency improvements, and product mix optimization.
Quality of Operating Cash Flow: Focus on companies whose operating cash flow (OCF) consistently exceeds their accounting profit (EBITDA) (OCF/EBITDA > 1), which usually indicates good working capital management and a lower risk of earnings manipulation.
Balance Sheet Strength: In a high-interest-rate environment, a strong balance sheet is the foundation for survival and development. Key indicators include: Interest Coverage = EBIT / Interest Expense, and a prudent level of leverage. Look for companies that have the ability to make counter-cyclical investments or acquisitions during an industry downturn.
VIII. A Numerical Calculations Linking the Six Charts
To translate the above qualitative analysis into more actionable quantitative insights, we can perform some simplified numerical calculations. This helps to communicate risks and returns more clearly in investment decision-making meetings.
Calculation 1: Calculating the Implied Ten-Year Total Return of the S&P 500
Input Assumptions:
Initial P/E ratio P/E0 = 22x (current level)
Future ten-year EPS compound annual growth rate EPS CAGR = 6% (a relatively optimistic assumption, higher than long-term nominal GDP growth)
Change in valuation after ten years Δ(PE) = -2x, i.e., P/E moderately declines from 22x to 20x (a neutral to conservative assumption)
Average dividend payout ratio Div Payout = 35%
Approximate Calculation:
Capital Gains Portion (Price CAGR): Price CAGR ≈ g + (Δ(PE) / T) / PE0
≈ 6% + ((-2 / 10) / 22) ≈ 6% - 0.009 ≈ 5.1%
Dividend Return Portion (Dividend Yield, DY): DY ≈ (E/P) × payout ≈ (1/22) × 35% ≈ 4.55% × 35% ≈ 1.6%
Expected Total Return: Total Return ≈ Price CAGR + DY ≈ 5.1% + 1.6% ≈ 6.7%
Scenario Analysis and Stress Testing:
If Growth is Revised Downward: If g is revised down to a more realistic 4%, then Price CAGR ≈ 4% - 0.9% = 3.1%, and Total Return ≈ 3.1% + 1.6% = 4.7%.
If Valuation Compression is More Severe: If Δ(PE) = -4x (declining to 18x), then Price CAGR ≈ 6% + ((-4 / 10) / 22) ≈ 6% - 1.8% = 4.2%, and Total Return ≈ 4.2% + 1.6% = 5.8%.
Conclusion: This simple exercise shows that from the current valuation starting point, even with fairly optimistic assumptions about future earnings growth, the long-term return of the S&P 500 may be quite mediocre, only slightly higher than the current risk-free rate. Once growth falls short of expectations or the pressure for valuation reversion increases, its rate of return will quickly fall into the 4-5% range, offering almost no risk compensation.
Calculation 2: A "Reality Check" Calculation for Stocks with EV/Sales > 10x
Objective: To test how extreme the operating assumptions need to be to support a valuation of EV/Sales = 10x.
Formula: EV/Sales ≈ m × (1-τ) / (r - g)
Verification Process:
Initial Optimistic Assumptions: Steady-state operating profit margin m = 25%, tax rate τ = 15%, and discount rate minus growth rate spread r - g = 4%.
Calculation Result: EV/Sales ≈ 0.25 × (1 - 0.15) / 0.04 = 0.2125 / 0.04 = 5.31x.
Conclusion: Even with a set of already quite optimistic assumptions, the reasonable valuation is only 5.3x, far from 10x.
Reverse-Engineering the Required Conditions:
To make EV/Sales = 10x hold, if we keep r-g=4%, then the profit margin m needs to reach (10 × 0.04) / 0.85 ≈ 47.1%.
Alternatively, if we keep m=25%, then r-g needs to be compressed to (0.25 × 0.85) / 10 = 2.125%.
Significance: This exercise clearly illustrates how extreme the expectations embedded in a 10x EV/Sales valuation are. It requires the company to have either near-monopolistic, incredible profitability, or the ability to grow sustainably at a very high rate, while the market's risk requirement for it is extremely low. This provides a powerful quantitative tool for identifying potential valuation bubbles.
Calculation 3: ROIC Threshold Analysis for BigTech Capital Expenditures
Objective: To estimate what level of operational efficiency a new AI capital expenditure needs to achieve to break even (i.e., ROIC = WACC).
Simplified Assumptions:
A company makes a new Capex = $50 billion.
Weighted Average Cost of Capital WACC = 10%.
Annual depreciation and amortization coefficient for the capital φ = 20% (i.e., fully depreciated in 5 years).
Tax rate τ = 15%.
Unit compute gross profit (Price - Variable Cost) = $0.12 per compute-hour.
Annual effective utilization hours H = 6,000 hours.
The total compute power brought by this investment is Total_Compute.
The compute utilization rate U is the key variable to be solved for.
Calculation Process:
Target NOPAT: Target NOPAT = Capex × WACC = $50B × 10% = $5B.
Actual NOPAT: Actual NOPAT = (Unit_Gross_Profit × U × H × Total_Compute) × (1 - τ).
Breakeven Point: By setting Actual NOPAT = Target NOPAT, we can solve for the minimum utilization rate U* required to achieve ROIC = WACC.
Analysis and Inference: Although this model is simplified, it reveals the core issue: when the industry enters a phase of "trading price for volume," i.e., when unit compute gross profit declines, or when killer applications are slow to emerge, leading to a prolonged low compute utilization rate U, the ROIC of these massive investments could quickly fall below WACC, thus beginning to destroy shareholder value. Investors should pay close attention to the marginal changes in these operational indicators.
IX. An Actionable Checklist for "Selectivity and Valuation"
Based on all the analysis above, we can formulate an actionable checklist for current investment decisions that does not contain a final conclusion but emphasizes process.
Dual-Threshold Screening for Multiples and Earnings:
Valuation Threshold: Prioritize companies whose forward P/E ratio is below the 70% percentile of its own historical range.
Earnings Threshold: At the same time, require that the company's expected EPS CAGR over the next 3-5 years can exceed a minimum threshold, for example, 6%.
Case-by-Case Review for Extreme Multiples: For any company with an EV/Sales > 10x, an independent, in-depth due diligence must be initiated to ensure that it can pass the rigorous "profit margin—free cash flow—ROIC" triple-constraint test mentioned earlier.
Testing for Cash Flow Quality and Balance Sheet Strength:
Cash Flow Conversion Rate: Require that Free Cash Flow / Net Income (FCF Conversion) is consistently greater than 90% to ensure that accounting profits are effectively converted into real cash.
Leverage Level: The Net Debt / EBITDA ratio should be below 1.5x (for non-financial and asset-heavy industries) to ensure financial flexibility during an economic downturn.
Interest Coverage Capacity: EBIT / Interest Expense (Interest Coverage) should be greater than 6x to ensure the company has sufficient capacity to service its debt and is not overly exposed to a high-interest-rate environment.
A "Verification" Checklist for the AI Capex Cycle:
Efficiency Indicator: Track the marginal trend of ΔROIC / ΔCapex.
Operational Indicators: Pay close attention to industry-level GPU utilization U data, as well as the slope of the unit compute price curve of major cloud vendors.
Cost Management: Investigate whether the company has locked in future data center electricity and cooling costs through long-term power purchase agreements (PPAs) or other means, to assess its ability to withstand energy price volatility.
Decision Rule: When it is observed that a company's Capex/OCF ratio is continuously rising, but its ROIC fails to improve in sync or even begins to decline, this should be treated as a strong warning signal, and consideration should be given to reducing exposure to the "valuation duration" risk of that company.
Relative Value Cross-Asset Allocation and Hedging Strategies:
The "Ballast" Role of Fixed Income: In a macro phase where the spread between the 10-year Treasury yield and the S&P 500 dividend yield is significantly negative, fixed income assets should be given a more important strategic role in asset allocation.
Maturity Selection: Consider using shorter-duration, investment-grade (IG) or government agency (Agency) bonds in the middle of the yield curve (3-7 years) to fulfill some of the "nominal yield" function in the portfolio.
Structural Adjustments within Equities: Within the equity allocation, moderately overweight equal-weight/mid-cap index funds, or actively select "growth-at-a-reasonable-price" companies with verifiable cash flows, to balance the over-reliance on a few high-valuation giants.
Comments