Record Highs, Empty Savings Accounts, and the AI Spending Paradox

The stock market is celebrating record highs. The S&P 500 has reached its 25th all-time high of the year, market breadth has improved, financial stocks have delivered an unusual 10-week winning streak, and even defensive businesses are participating in the rally.

On the surface, the market looks strong.

Underneath it, the economy looks far less stable. Consumer spending is growing faster than income. The personal savings rate has fallen to 2.7%, near a 70-year low. At the same time, large technology companies are committing hundreds of billions of dollars to AI infrastructure while free cash flow deteriorates.

This is the central market paradox: equities are pricing in abundant growth, cheaper capital, and transformative AI productivity, while the consumer is increasingly constrained and inflation may limit the Federal Reserve's ability to provide support.

Quick summary: what is happening right now

• The S&P 500 is reaching repeated record highs, supported by broad participation beyond a narrow group of technology stocks.
• Retail speculation is increasing, with Robinhood margin interest reported at $215 million and event-contract activity exceeding crypto-related activity.
• Second-quarter GDP growth came in at 1.5%, below the 1.8% consensus expectation.
• Consumer spending rose 0.3% in June while personal income rose only 0.2%.
• The personal savings rate fell to 2.7%, leaving consumers with limited capacity to keep spending ahead of income.
• Amazon, Google, Meta, Microsoft, and other large technology firms are deploying enormous capital into AI chips, servers, networking, and data-center infrastructure.
• Meta's free cash flow reportedly fell 91%, while Amazon posted negative free cash flow during the spending surge.
• Markets are increasingly relying on the assumption that weak economic data will force the Fed to cut rates.
• That assumption is vulnerable if sticky services inflation, particularly Supercore PCE, continues accelerating.

The market rally is broad, and that matters

It is too easy to compare every technology-led rally to the 1999 dot-com bubble. That comparison misses an important difference: market breadth.

In the late 1990s, a small group of major technology companies carried the indexes higher while much of the market was already weakening. That was a narrow rally. Narrow leadership often creates a fragile market because a small number of stocks are responsible for holding up the index.

The current environment has shown much broader participation. Financials have completed a 10-week winning streak, an unusually long run not seen since 1989. Traditional and defensive companies have also moved higher. Casey's General Store, a Midwestern convenience-store operator associated with gasoline and pizza, has shown momentum comparable to far more glamorous semiconductor names.

Broad participation is normally a constructive sign. It suggests investors are not simply crowding into a few mega-cap technology companies.

However, breadth does not eliminate risk. It only tells us the rally is widespread. It does not tell us whether the capital supporting the rally is disciplined, sustainable, or dependent on cheap financing.

Retail speculation is rising again

The quality of market participation matters as much as the quantity.

Retail investors are returning to high-risk behavior. Robinhood margin interest reportedly reached $215 million, while revenue from event-contract betting reached $156 million. These contracts allow users to take binary positions on outcomes, including sports-related events such as World Cup matches.

That activity is important because it shows how some market participants are approaching risk. A brokerage account can become a tool for long-term investing and capital formation. It can also become a venue for leveraged speculation, short-term betting, and emotional decision-making.

When participants are chasing rapid gains, using margin, and treating financial markets like a sportsbook, risk tolerance is typically high. High risk tolerance can sustain a rally for longer than fundamentals justify. It can also accelerate downside once liquidity reverses.

GDP is growing, but the consumer is funding spending from savings

The second-quarter GDP figure came in at 1.5%, below the 1.8% expectation. That is not a recessionary number by itself, but it is a weak growth rate for an economy being valued by the equity market as if strong expansion is ahead.

Several components weighed on GDP:

• Government spending pulled back modestly.
• Business inventory growth declined.
• Imports reduced the headline GDP calculation.

The consumer data is the more important concern.

Consumer spending increased 0.3% in June. Personal income increased only 0.2%. That difference may appear small for one month, but the economic logic is straightforward: spending cannot consistently grow faster than income without another funding source.

That source is savings, debt, or both.

The personal savings rate has now fallen to 2.7%, near a 70-year low. Consumers are still transacting, which keeps GDP from falling more sharply. But they are doing so with less financial cushion.

Think of the consumer as the main cable on a suspension bridge. The bridge may look stable while traffic is moving. The real test is whether the cable can handle persistent tension. A 2.7% savings rate means households have less reserve capacity if job growth weakens, credit conditions tighten, or inflation rises again.

Spending eventually has to align with income. If income growth does not reaccelerate, consumer spending will have to slow.

AI infrastructure imports are distorting the GDP picture

There is a mechanical detail inside the GDP data that deserves attention. Imports subtract from GDP calculations. A portion of the imports weighing on second-quarter GDP was not traditional consumer-goods demand. It was corporate demand for AI-related equipment.

Companies are importing:

• AI chips and semiconductor hardware
• Networking equipment
• Data-center servers
• Cooling systems and related infrastructure

This creates a strange split in the economy. Consumers are running down savings to maintain spending. Meanwhile, enterprise technology firms are aggressively spending on equipment to build the next generation of AI infrastructure.

One side of the economy is constrained by income. The other is investing heavily based on future expectations.

Big Tech's revenue growth is strong, but free cash flow is under pressure

On the headline level, large technology earnings have looked excellent. Many companies are reporting revenue growth in the 20% to 30% range. That is the number investors want to see.

But revenue is not the same as free cash flow.

Free cash flow measures what remains after a company pays operating expenses and capital expenditures. It is one of the cleanest measures of financial flexibility because it shows how much cash a company actually has available after funding the business.

The AI infrastructure race is putting meaningful pressure on that measure. Meta's free cash flow reportedly fell 91%. Amazon posted negative free cash flow. Across the major technology platforms, collective AI infrastructure spending is estimated at roughly $150 billion to $200 billion.

That level of capital expenditure requires a clear return on invested capital. At some point, investors will require evidence that these data centers, chips, servers, and networks can generate durable revenue and attractive margins.

The AI arms race may be driven by corporate anxiety

Professor Aswath Damodaran has framed the AI spending surge as a form of corporate anxiety. The logic is direct.

Mega-cap technology companies have traded at premium valuation multiples for years because investors view them as high-growth innovators. If the market starts treating them as mature, slower-growth businesses, valuation multiples compress. Even if revenue remains large, a lower multiple can erase trillions of dollars in market value.

That creates an incentive to preserve the growth narrative.

Building enormous AI infrastructure allows companies to position themselves as leaders in the next platform shift. But there is a major investment question: where is the fully monetized consumer product?

Many firms are spending heavily before the following issues are fully resolved:

• What AI products will consumers and enterprises pay for?
• What level of pricing power will those products have?
• How quickly can infrastructure spending convert into recurring revenue?
• Will returns exceed the cost of capital?

These are not minor questions. They determine whether AI CapEx becomes a long-term growth engine or an expensive overbuild.

Apple is taking a different approach

Apple stands out because it is not participating in the hardware arms race at the same scale as Amazon, Google, Meta, and Microsoft. Its capital expenditures are roughly $11 billion, far below the amounts being deployed by several competitors.

That approach reflects capital discipline.

Apple appears willing to let competitors absorb the early risks of building data centers, purchasing chips, funding research, and expanding computing capacity. Rather than spending aggressively before demand is proven, Apple can wait to see which AI applications consumers actually adopt.

This is a calculated strategy. It protects free cash flow and reduces the risk of spending billions on infrastructure before there is a clear monetization path.

It also highlights the central question facing the rest of Big Tech: are they building indispensable infrastructure, or are they simply racing because none of them can afford to appear behind?

The circular revenue illusion in AI

A deeper concern is the possibility of circular economics within the AI ecosystem.

The structure works like this:

1. A major technology company takes an equity stake in a private AI startup such as OpenAI or Anthropic.
2. The startup raises a new funding round at a higher private valuation.
3. The large technology investor records an unrealized gain based on the higher valuation of its stake.
4. The startup uses its new funding to purchase cloud services and computing capacity from the same large technology company.
5. The technology company records the startup's spending as cloud revenue.

Under mark-to-market accounting rules, a company can report gains from a higher valuation of an investment even though it has not sold the investment and received cash. Those gains are paper gains, not realized liquidity.

The issue becomes more significant when the startup then uses funding to buy cloud capacity from its strategic investor. The investor may benefit from both a higher reported investment value and higher reported cloud revenue.

Reportedly, as much as 85% of Amazon's net-income growth and 87% of Google's net-income growth in a recent quarter came from unrealized paper gains.

This does not mean every AI investment is artificial or unsustainable. It does mean investors need to separate three different things:

• Realized cash earnings
• Unrealized valuation gains
• Revenue that may be linked to the financing ecosystem itself

When liquidity is abundant, markets often tolerate these structures. When capital becomes scarce, markets demand actual free cash flow, durable end-user demand, and clean returns on investment.

The Ashenbrenner collapse shows the danger of correlated leverage

Leopold Ashenbrenner's trading blowup is a useful example of what happens when a credible macro thesis is paired with poor risk construction.

A former employee of OpenAI and FTX, Ashenbrenner published a widely discussed AI thesis titled Situational Awareness. The paper correctly identified the rising demand for AI infrastructure. On the strength of that thesis, he secured billions in funding and launched an investment fund.

The fund then suffered one of the fastest multibillion-dollar losses in recent memory.

The problem was not necessarily the AI thesis. The problem was leverage and correlation.

The fund reportedly used a leveraged long-short trade:

• Long semiconductor hardware stocks
• Short enterprise software stocks

In a long-short trade, the investor expects the long side to outperform the short side. Borrowed capital amplifies returns if that spread moves in the expected direction.

It also amplifies losses if both sides move against the position.

That is what happened when the market rotated. Semiconductor stocks sold off, damaging the long positions. Enterprise software stocks rose, creating losses on the short positions. A leveraged portfolio can deteriorate quickly when both sides move against it.

Once equity falls below required thresholds, prime brokers issue margin calls. Forced liquidations follow. The investor must sell the declining long positions and buy back the rising short positions, often at the worst possible prices.

This is a textbook margin spiral.

The broader lesson is simple: a strong macro idea does not protect an investor from poor position sizing, leverage, liquidity risk, or correlation breakdown.

Markets are rotating from AI infrastructure to AI applications

The market is beginning to demand proof of monetization.

Capital is gradually moving away from companies that merely sell the building blocks of AI infrastructure and toward companies that can demonstrate measurable economic value from AI deployment.

Palantir is a clear example. Its stock rose 30% after an earnings report highlighted 150% growth in U.S. commercial revenue. The company is not simply selling chips, servers, or data-center capacity. It is integrating AI into enterprise workflows and positioning its products as tools that can improve operational efficiency today.

This distinction matters.

Infrastructure companies are selling the picks and shovels of the AI buildout. Application companies must prove that AI creates revenue, lowers costs, improves decisions, or increases productivity for paying customers.

The next stage of the AI trade will likely reward proof of value over abstract potential.

The "less bad" trade is spreading through the market

Speculation is not limited to technology. It is also appearing in distressed sectors through what can be called the "less bad" trade.

Floor & Decor provides a useful case study. The U.S. housing market remains frozen by mortgage rates around 7%. Transaction activity has slowed, and large remodeling projects are being deferred. Floor & Decor's stock had already declined roughly 70% from prior highs because the market expected prolonged weakness.

Then the company reported results that were not strong, but were less negative than feared.

The stock rallied sharply and moved above its 200-day moving average.

This is often driven by short covering. When hedge funds build large short positions around a terminal-decline thesis, a company does not need to produce excellent results to rally. It only needs to survive better than expected.

Short sellers then buy shares to close positions. That buying can push the price higher, forcing more short sellers to cover. The result is a relief rally.

Investors should understand the difference between a short-covering rally and a fundamental turnaround. A company can rally sharply without having solved the economic problem that caused its decline.

Bad economic data has become a bullish signal

One of the most dangerous market dynamics today is the "bad news is good news" mindset.

Markets are operating under the assumption that weak data will force the Federal Reserve to cut interest rates. Lower rates reduce borrowing costs and increase the present value of future cash flows. This especially benefits long-duration growth stocks, including technology companies.

The July labor data illustrates the point. The U.S. economy lost 23,000 jobs. Under a traditional economic framework, negative payroll growth signals weakening demand and a deteriorating labor market.

Yet equities rallied.

The market's logic was mechanical: fewer jobs may pressure the Fed to cut rates, and rate cuts may support equity valuations.

This is the modern version of the Fed put, the belief that the Federal Reserve will step in with easier policy whenever financial conditions deteriorate.

That belief encourages investors to take more risk because they expect policy support to limit downside.

Is job loss AI productivity or the start of a cyclical slowdown?

Wharton professor Jeremy Siegel argued that maintaining 1.5% GDP growth while losing 23,000 jobs could be evidence of AI-driven productivity. Under this interpretation, companies are using technology to preserve output while reducing labor costs.

From a corporate margin perspective, the argument is understandable. If a company can generate the same revenue with fewer employees, profit margins improve.

But there is another interpretation.

Weak GDP combined with falling payrolls may simply represent a standard cyclical slowdown. Companies may be reducing headcount because demand is weakening, not because AI has created a clean productivity miracle.

The difference is crucial.

• If job cuts are driven by genuine productivity, businesses may maintain output, margins, and investment.
• If job cuts are driven by falling demand, revenue growth may weaken next, followed by lower earnings and reduced capital spending.

Investors should not automatically treat layoffs as proof of technological progress. Context matters: revenue trends, margins, spending, hiring plans, and end-market demand all need to be evaluated together.

Sticky inflation could break the Fed-cut narrative

The bullish case relies heavily on one assumption: inflation is under control.

That assumption is not secure.

Supercore PCE is a key measure to monitor. It focuses on underlying services inflation while stripping out volatile categories such as housing and energy. It is designed to capture persistent wage and service-price pressure, the type of inflation that is difficult for the Federal Reserve to ignore.

That measure is accelerating again.

Former Federal Reserve Governor Kevin Warsh has emphasized that the current Federal Open Market Committee remains committed to a 2% inflation target. If persistent inflation remains above target, the Fed has less flexibility to cut rates simply because equity markets want lower financing costs.

This creates a policy trap:

• GDP growth is weak at 1.5%.
• The consumer has limited savings capacity.
• Labor data is weakening.
• Services inflation remains sticky.
• Geopolitical tension could push oil prices higher.

If an energy shock related to Middle East tensions, including risks surrounding Iran, pushes crude oil prices sharply higher, headline inflation could reaccelerate. In that scenario, the Fed may not be able to deliver the rate cuts the market has already priced in.

It could even face pressure to keep policy restrictive for longer, or tighten again if inflation rises materially.

$2 trillion in AI debt depends on cheap capital

The AI infrastructure buildout has reportedly absorbed about $2 trillion in global off-balance-sheet debt. That spending is based on an expectation that future financing will remain available and reasonably cheap.

This is the major structural risk.

AI infrastructure can be funded aggressively when capital is cheap, equity valuations are high, and investors are focused on future potential. If interest rates stay elevated or move higher, the economics change quickly.

Higher borrowing costs expose weak business models. They force investors to distinguish between genuine cash generation and narrative-driven growth. They also make heavy capital expenditure less attractive because the required return on investment rises.

If AI infrastructure has not produced a clear monetizable application by the time liquidity tightens, the unwind could be severe.

The historical comparison is the late-1990s telecom buildout. Companies installed enormous quantities of fiber-optic infrastructure based on demand forecasts that were ultimately too early. The infrastructure itself had value, but the timing and financing structure damaged investors when expected demand did not arrive fast enough.

AI could follow a similar path if the technology is transformative but monetization takes longer than markets expect.

The ultimate AI paradox: who buys the output?

The deepest question is not whether AI can automate work. It clearly can automate certain tasks and may improve productivity across many industries.

The deeper question is what happens to consumer demand if automation eliminates a large share of knowledge-economy jobs.

The consumer drives roughly two-thirds of the U.S. economy. Consumers need income to spend. If companies use AI to eliminate millions of jobs, they may improve their own efficiency and margins in the short run.

But those displaced workers are also customers.

A fully automated economy can produce goods and services at lower cost. It cannot sustain demand if a growing share of households no longer has dependable income.

This is the AI paradox:

• Companies want to automate labor to improve margins.
• Workers need income to buy the products and services companies sell.
• Reducing labor income too aggressively can weaken the consumer base that supports corporate revenue.

Efficiency is valuable. But a system built entirely around cost reduction eventually runs into a demand problem.

What investors should watch from here

The market is not necessarily about to collapse. Broad participation, strong headline earnings, and genuine AI innovation can support elevated valuations for longer than skeptics expect.

But this is not a market to analyze only through index highs.

Focus on the underlying cables:

• Personal income growth versus consumer spending growth
• The personal savings rate and consumer-credit conditions
• Free cash flow at AI-heavy technology companies
• Capital expenditure growth relative to AI revenue and operating profit
• Unrealized gains versus realized earnings
• Supercore PCE and other sticky-inflation measures
• Federal Reserve policy expectations versus actual inflation data
• Oil prices and geopolitical risks
• Whether AI applications can prove measurable, recurring customer value

Record highs are not proof that risk has disappeared. In fact, record highs are often the right time to inspect the assumptions holding the market together.

Conclusion: the bridge is carrying heavy traffic

Today's market is supported by real strengths: broad participation, substantial technology revenue growth, and an investment cycle centered on a potentially transformative technology.

It is also carrying major structural weaknesses: a depleted consumer savings rate, aggressive AI capital expenditures, possible circular revenue structures, rising leverage, and an overconfident belief that the Federal Reserve will always provide a safety net.

The critical variable is not whether AI will matter. It will.

The critical variables are timing, monetization, financing costs, and consumer demand. If AI investment produces durable free cash flow before the consumer weakens and liquidity tightens, this cycle can continue. If capital costs rise before monetization arrives, markets will stop rewarding narratives and start demanding cash.

That is when the pressure on the underlying cables becomes visible.

SOURCES

U.S. Bureau of Economic Analysis: Gross Domestic Product

U.S. Bureau of Economic Analysis: Personal Income and Outlays

Federal Reserve: Federal Open Market Committee

Federal Reserve: Monetary Policy