From AI CapEx to SPY All-Time Highs: Is the US Stock Market Realizing AI Returns, or Preemptively Pricing in 2027 Profits?

Too Long; Didn’t Read

1.This rally is supported by real profits. As of July 31, the S&P 500's Q2 blended earnings grew 47.4% year-over-year. Even excluding massive investment valuation gains from Google and Amazon, the growth rate remains at 28.8%. Revenue increased by 14.1%, all 11 sectors experienced growth, and the earnings base is much sturdier than a rally “relying only on a handful of AI stocks”.

2.However, the apparent 47.4% growth rate cannot be simply extrapolated. Investment valuation gains of about $98 billion for Google and $53.4 billion for Amazon significantly boosted the index's earnings and profit margins. While they can increase this period’s net income, they do not demonstrate that enterprise customers are already receiving sustainable cash returns from AI applications.

3.Improvement in breadth and rising index concentration can coexist. The equal-weight index and the Russell 2000 outperformed at times, indicating broadening participation; yet the top 10 securities in SPY still account for 38.01%, and the information technology sector for 36.65%. More stocks rising doesn't negate the decisive impact of a few mega-cap companies on index earnings and valuations.

4.7,736.52 points is already pricing in 2027 earnings. Based on public forecasts by Goldman Sachs and JPMorgan of S&P 500 EPS at $385 for 2027, the current level is about 20.1 times earnings; if you look only at the $330-340 forecasts for 2026, the multiple is 22.8-23.4. Whether the new high can be sustained will depend on continued upward revisions for 2027 earnings, not just a Q2 beat.

5.Realized earnings form the floor, 2027 expectations form the ceiling. Currently there isn’t enough evidence to confirm a cyclical top or declare AI investment returns have closed the loop. The true watershed moment depends on: whether adjusted earnings can hold, whether AI revenues and free cash flow can catch up to capital spending, whether market breadth can persist, and whether 10-year Treasury yields and credit spreads remain stable.

The S&P 500 index tracked by SPY closed at a record high of 7,736.52 on August 4th.The acceleration of profits is real, but one-off valuation gains, concentration, and lagging capex depreciation mean “new highs” don’t yet equal realized AI returns.

1. After New Highs: The Question Has Shifted from “Are There Any Earnings?” to “Can Earnings Last Into 2027?”

On August 4, the S&P 500 index rose 1.8% to 7,736.52; the Nasdaq Composite was up 2.6%, the Dow Jones Industrial Average up 1.7%, and the Russell 2000 up 1.8%.Year-to-date as of that day, the S&P 500, Nasdaq Composite, and Russell 2000 are up about 13%, 14.4%, and 22.4%, respectively. Oil prices fell, the yield on 10-year US Treasuries declined from 4.70% to 4.62%, and strong earnings together pushed risk assets higher.

This performance is significant because it shifts the entry point for the latest debate.From 2025 to early 2026, market arguments focused on whether AI capex was too high and whether macro rates would compress valuations; with this new high, the market has provided a phased answer: corporate earnings have not been crushed by high rates, but rather are accelerating. Yet prices lead the financials—what the index is now trading on is not “what happened in Q2” but rather “can current earnings momentum persist through 2027?”

To judge whether the new highs are healthy, at least three sets of accounts must be reviewed concurrently.

Therefore, a “new high” is neither proof of a bubble nor evidence of returns being realized.It only means the market is willing to price in higher future earnings today. What needs evaluating next is whether these future earnings come from repeatable operating income, or from one-off accounting gains, front-loaded capex, and multiple expansion.

2. The Profits Are Real, But the 47.4% Needs to Exclude Two Large Valuation Gains

FactSet data through July 31 showed that 61% of S&P 500 constituents had reported Q2 earnings, 86% of which beat expectations, above the five-year average of 78% and the ten-year average of 76%; 77% beat on revenues, with an overall revenue surprise of 2.9%.Q2 blended revenue grew by 14.1% year-over-year, with all 11 sectors recording growth. Out of 11 sectors, 10 had profit growth, and 8 achieved double-digit profit growth.

This means that current earnings are broad—outperformance ratios are high, revenue growth is widespread, and industry participation is high.Even without discussing AI, consumption, finance, industrials, and international operations are all contributing incremental growth. FactSet splits by revenue source show that companies with more than 50% overseas revenues saw Q2 profits up 74.7% and revenues up 20.5%; excluding Google, ExxonMobil, and Chevron, earnings and revenue still grew 27.5% and 16.0%. Outliers exist, but there’s genuine growth beyond the extremes.

The issue is that the eye-catching 47.4% earnings growth contains two items that shouldn’t be annualized as regular operating profit: Google recognized around $98 billion in investment valuation gains and Amazon recognized about $53.4 billion in pre-tax “other income” (mainly from valuation changes to its Anthropic stake).Amazon alone accounted for 76% of the S&P 500’s week-over-week net profit increase. These are real accounting profits but not cloud revenue, ad income, or cash from AI application sales.

The strength of earnings must be layered. Layer 1 is revenue and operating profit, which support sustainable multiples; Layer 2 is investment appreciation, which strengthens the balance sheet but fluctuates with market prices; Layer 3 is revenue from future AI projects, much of which hasn’t yet hit the P&L. Mixing all three overstates current AI returns and understates core operating growth.

In other words, Q2 is enough to refute “US stocks have no earnings support,” but not enough to prove “AI capex has already generated equivalent cash returns at the end customer level.”What the market really needs for the next few quarters is: even without large investment gains, adjusted earnings can still grow near double digits, or even above 20%, supported by both revenue and cash flow.

3. Market Breadth Is Improving, but SPY’s Outcome Is Still Driven by a Few Mega-Caps

There are two opposing narratives to this rally.One believes gains remain driven by a few big techs and are therefore fragile; the other sees equal-weight and small-cap outperformance as proof that concentration is no longer an issue. Both have some basis in fact.

According to State Street Global as of August 3, the top 10 SPY holdings represented 38.01% of the ETF.Nvidia, Apple, Microsoft, Amazon, two classes of Google shares, Broadcom, and Meta—a total of 8 securities from 7 firms—constitute 35.12%. Information technology holds a 36.65% weight, while communication services and consumer discretionary are 10.18% and 9.51% respectively. Therefore, the profits, capex, and valuation changes of a few mega-cap tech firms still directly determine SPY’s direction.

But the market breadth is indeed better than before.The S&P 500 equal-weight index gives every company about a 0.2% weight at quarterly rebalancing and is up 11.44% YTD through July 24; CME’s Q2 review shows the equal-weight index outperformed the market-weight at times, and the Russell 2000 did even better. As of August 4, the Russell 2000’s YTD gain of 22.4% is significantly higher than the S&P 500.

These two datasets aren’t contradictory.Breadth measures “how many stocks are rising”, while concentration measures “who determines the index outcome.”More sector and small-cap participation lowers the risk of “only a few firms are profitable,” but doesn’t change the math: when SPY moves 1% up or down, mega-cap techs contribute the most.

This also explains why this rally is healthier than the pure “magnificent seven” rally, but does not mean valuation tolerance has improved proportionally:

• If the equal-weight index keeps outperforming and profits broaden, the index could stay high even with sector rotation within tech;
• If big tech profits meet expectations but small/mid-cap recovery continues, market returns will shift from concentration to broad participation;
• If both big techs see profit downgrades while equal-weight/small caps' strength is only short-term liquidity, SPY will still face noticeable pressure due to its structure.

Therefore, judging a new SPY high can’t be based solely on the number of gainers or only the top 10 weights.The truly useful clues are: the relative performance of equal-weight vs SPY, Russell 2000 vs S&P 500, and whether profit estimates outside the mega-caps are being revised up in tandem. Only when both “price breadth” and “earnings breadth” improve together is the rally more than just rotation.

4. Current Levels Do Not Merely Price 2026—They're Already Pricing 2027

FactSet reported forward 12-month S&P 500 P/E at 19.6x as of July 31, below 20.4x on June 30 and the five-year average of 19.9x, but still above the ten-year average of 19x.The lower P/E doesn’t mean the index got cheaper, because it came as profit forecasts were rapidly revised up; the index rose another 1.8% on August 4, moving valuations forward again.

More importantly, interest rates matter.A 19.6x P/E means a simple earnings yield of about 5.10%. Compared with the 10-year US Treasury yield of 4.62% on August 4, the spread is only 0.48 points. It’s not an official equity risk premium, and the two data points are not exactly synchronous, but it does show one thing: when risk-free rates stay near 4.6%, stock valuations rely on profit growth—not low rates boosting multiples naturally.

Goldman Sachs forecasts S&P 500 EPS at $340 for 2026 and $385 for 2027; JPMorgan puts it at $330 and $385, respectively.At 7,736.52, the current multiple is clearly factoring in earnings across years.

If 2027 EPS does reach $385, whether the index keeps rising will depend on what multiple the market is willing to pay then.18x corresponds to 6,930; 19x to 7,315; 20x to 7,700; 21x to 8,085. This is mathematical sensitivity, not a point forecast, but it does reveal the true constraint of current pricing:even if earnings deliver in full, if long-term rates rise or risk appetite drops, multiples could drop from 20x to 18-19x, and the index could fall below current levels.

Conversely, if 2027 earnings keep being revised up and the 10-year Treasury yield falls, while credit spreads stay steady, current valuations may not represent a top either.CME’s review of prior cycles shows that profit slumps, widening high yield credit spreads, and rising volatility are stronger top signals than any single valuation measure. Profits are still rising and credit spreads haven’t shown classic sustained deterioration, so one cannot equate “expensive” with “about to peak”.

5. Why AI Capex Can Boost Index Earnings Early, but Doesn’t Mean Returns Have Fully Materialized Yet

Understanding the relationship between SPY’s new high and AI investment returns means differentiating between revenue timing for various parts of the industry chain.

The suppliers of GPUs, networking and power equipment, and data center builders can recognize revenue promptly upon delivering products or services; cloud providers and large platforms capitalize server, data center, and related infra purchases, spending cash upfront with depreciation recognized as future costs; end-customers face a further lag as models are deployed, products commercialized, labor costs saved, or sales conversion boosted.The result is:AI supply chain revenue can show up in index earnings earlier, while buyer-side cash returns may not arrive for quarters or years.

Capex exceeding operating cash flow doesn’t mean an investment has failed.If backlog, cloud service demand, revenue per compute, and customer retention all rise, upfront spending can yield long-term assets and future cash flow. The problem is, if outlays grow persistently faster than cash generation, companies must rely on cash, debt, leasing, or other financing to plug the gap, and the market will demand faster realization of returns.

In this profit season, investment valuation gains for Google and Amazon add another layer of misalignment.These reflect AI ecosystem asset appreciation but do not equate to enterprise customers paying equivalent recurring income for AI services. Treating these gains as the same as GPU sales, cloud revenue, ad efficiency, or end-user SaaS makes “AI returns” appear to progress faster than underlying cash flow.

Thus, for AI investment returns to truly close the loop at the index level, at least four transitions must occur:

1. Infra investment turns into available compute, not perpetual construction work or underutilized assets;
2. Available compute becomes cloud, model API, or software revenue—unit prices can’t fall faster than usage grows;
3. Revenue turns into operating profit and cash flow—not persistent capex masking cash burn;
4. Operating cash flow covers incremental capex, interest, and shareholder returns, yielding sustainable free cash flow.

Only after these four steps can the market be said to be “realizing AI returns”; if just the first one or two, the index is really just pricing in future income streams ahead of time.