AI Capital Spending Hits Record Highs and Wall Street Keeps Buying the Dip

Meta Platforms raised its 2025 capital expenditure guidance to $60-65 billion on July 30, up from a prior estimate of $53-55 billion. The stock dipped in after-hours trading, then recovered. Investors treated the pullback as a buying opportunity. The pattern repeated across Big Tech earnings season: massive AI spending announcements followed by brief sell-offs, followed by rapid recoveries. The "buy the f-ing dip" mentality now dominates the AI trade, and the numbers involved have grown large enough to reshape capital allocation across the entire technology sector.

The Scale of AI Capital Expenditure

The figures are staggering by any historical standard. Meta plans to spend more on AI infrastructure in 2025 than the entire GDP of Iceland. Microsoft committed $80 billion for AI data centers in its fiscal year 2025. Alphabet disclosed $75 billion in planned capital expenditure. Amazon Web Services allocated $100 billion. These four companies alone will pour more than $300 billion into AI infrastructure this year.

To put that in perspective, the entire US venture capital industry deployed roughly $170 billion in 2024. Four corporations are now outspending the combined output of thousands of venture firms, limited partners, and startup ecosystems. The capital is flowing into Nvidia GPUs, custom chips, data center construction, power generation, and cooling systems.

Meta CEO Mark Zuckerberg told investors on the earnings call that the company sees "a really big opportunity" and wants to invest "aggressively" during a period where competitors might hesitate. The language mirrors what Jeff Bezos said about Amazon Web Services in 2014, when AWS spending worried Wall Street analysts who could not yet see the revenue that cloud computing would eventually generate. AWS now produces more than $100 billion in annual revenue.

The optimistic case writes itself. AI models are improving rapidly. Enterprise adoption is accelerating. Advertising revenue is climbing because AI-driven ad targeting produces better results. Meta reported Q2 2025 revenue of $42.3 billion, up 22% year-over-year. The spending looks aggressive, but the revenue growth supports it, at least for now.

The Bull Case for Buying Every Dip

Wall Street's reaction to each spending increase follows a predictable script. Shares drop 3-5% after hours as traders digest the higher capex number. By the next morning, analysts publish notes arguing the spending is justified. Within 48 hours, the stock recovers and often sets new highs.

Meta stock gained more than 60% in 2025 before the latest earnings report. Nvidia is up more than 150% over the past twelve months. The Nasdaq Composite has outperformed nearly every other asset class globally. Investors who sold on AI capex fears in mid-2024 missed a historic rally.

The bull thesis rests on a few pillars. First, AI improves the core business. Meta's advertising algorithms, powered by its Llama family of open-source models, are producing measurably better click-through rates and conversion metrics. Revenue per user is rising across all geographies. Second, AI creates new revenue streams. Coding assistants, enterprise chatbots, and image generation tools represent greenfield markets. Third, infrastructure spending creates competitive moats. A company that spends $65 billion on GPU clusters and data centers is building something that smaller competitors cannot replicate.

Anthony Pompliano, the investor and newsletter writer behind The Pomp Letter, captured the prevailing sentiment: the AI trade is not over, and investors will continue to buy the dip. His argument is simple. The technology works. The revenue is growing. The companies spending the most are also the ones generating the most cash flow. Until that equation breaks, the market has no reason to punish the spenders.

The Bear Case Nobody Wants to Hear

Not everyone shares the enthusiasm. A growing minority of analysts and investors worry that AI capital spending has entered bubble territory.

David Cahn, a partner at Sequoia Capital, published a widely cited analysis in 2024 arguing that AI companies would need to generate $600 billion in annual revenue just to cover the infrastructure being built. His updated estimate for 2025 pushes that figure closer to $800 billion. Current AI-specific revenue across the entire industry is a fraction of that number.

The skeptics point to historical precedent. In the late 1990s, telecom companies spent hundreds of billions laying fiber optic cable. The infrastructure proved valuable decades later, but the companies that built it, WorldCom, Global Crossing, Lucent, went bankrupt. The investors who funded the buildout lost everything. The beneficiaries were the next generation of companies, Google, Netflix, Facebook, that used the cheap infrastructure without bearing the construction costs.

Could the same pattern repeat? The AI infrastructure being built today will not disappear. The GPUs, data centers, and power plants are real assets. But the question is whether today's spenders will capture the economic value, or whether some future set of companies will build on top of these assets at a fraction of the cost.

There is also the efficiency question. Each generation of AI models tends to require less computation to achieve the same performance level. Techniques like distillation, quantization, and mixture-of-experts architectures are already reducing inference costs by 80-90% compared to two years ago. If the cost of running AI drops faster than demand grows, much of the infrastructure being built today could end up underutilized.

Jim Chanos, the veteran short seller, has been vocal about what he calls "AI capex mania." He draws parallels to the railroad boom of the 1860s, where the railroad companies themselves earned poor returns even as the broader economy benefited enormously from the transportation network they built. His point is not that AI is fake. It is that building the infrastructure and profiting from the infrastructure are two different things.

The Money Printer Connection

The AI trade does not exist in a vacuum. It exists inside a monetary system that has printed trillions of dollars over the past five years.

The Federal Reserve's balance sheet expanded from $4.2 trillion in early 2020 to a peak of $8.9 trillion in 2022. Even after quantitative tightening, it remains above $6.8 trillion. The M2 money supply in the United States has grown by roughly 40% since January 2020. In absolute terms, there are trillions more dollars chasing assets than there were five years ago.

This context matters. When money is abundant and interest rates, despite recent hikes, remain low by historical standards, capital flows toward the most compelling narrative. In 2020-2021, it was crypto and SPACs. In 2022-2023, it shifted to AI. The underlying driver is the same: an expanding money supply seeking returns in a system where holding cash means losing purchasing power to inflation.

Austrian economists would recognize this pattern immediately. Friedrich Hayek described the boom-bust cycle as a consequence of credit expansion that distorts the structure of production. Capital flows into projects that appear profitable at artificially low interest rates but prove unsustainable when monetary conditions normalize. The question is not whether AI is useful, it clearly is, but whether the scale of investment is calibrated to genuine demand or to the availability of cheap money.

Bitcoin offers an alternative framework. With a fixed supply of 21 million coins and a programmatic issuance schedule that cannot be altered by any central authority, Bitcoin forces discipline on capital allocation. There is no Bitcoin printer that can flood the market with new units to fund speculative buildouts. Every satoshi spent on infrastructure must be earned, not printed. This constraint, which critics call a limitation, is precisely what prevents the kind of malinvestment that characterizes late-cycle spending booms.

The irony is hard to miss. The same monetary expansion that funds the AI capex boom also erodes the purchasing power of the wages earned by the workers building the data centers. A dollar-denominated system incentivizes corporate spending, because holding cash is a losing strategy. A sound money system would force harder questions about whether $300 billion in annual AI infrastructure spending represents genuine value creation or credit-fueled overbuilding.

Concentration Risk and Market Structure

The AI trade has created an extraordinary concentration of market power. The "Magnificent Seven" tech stocks, Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, and Tesla, now represent more than 30% of the S&P 500's total market capitalization. Nvidia alone accounts for roughly 5% of the index.

This concentration creates fragility. A single disappointing earnings report from Nvidia or a shift in sentiment around AI spending could trigger a sell-off that ripples across retirement accounts, pension funds, and passive index funds worldwide. The vast majority of 401(k) holders are exposed to the AI trade whether they chose to be or not, simply by holding a standard S&P 500 index fund.

The market structure also raises questions about price discovery. When passive index funds automatically buy more of the stocks that have already gone up, a feedback loop emerges. Rising prices lead to higher index weights, which lead to more passive buying, which leads to higher prices. This dynamic can persist far longer than skeptics expect, but it also means that when the reversal comes, it can be violent.

Bitcoin's market structure is fundamentally different. No central committee decides how many bitcoins to issue. No passive index fund dominates price discovery. The market operates 24/7 across global exchanges with no circuit breakers and no Federal Reserve backstop. This openness introduces volatility, but it also produces more honest price signals. The price of bitcoin reflects what buyers are actually willing to pay, not what an algorithmic rebalancing formula dictates.

What to Watch

Three developments will determine whether the AI BTFD trade continues or breaks down.

First, watch Nvidia's next earnings report in late August. Nvidia is the picks-and-shovels play for the entire AI buildout. If data center revenue growth decelerates or if management signals softening demand from hyperscalers, the narrative could shift quickly. Consensus expects data center revenue above $35 billion for the quarter. Any miss would be significant.

Second, monitor AI revenue disclosure from the hyperscalers. Microsoft, Google, and Amazon have been vague about how much revenue their AI products actually generate. As spending scales into the hundreds of billions, investors will demand clearer attribution. If Q3 and Q4 earnings calls do not show AI revenue growing at a pace that justifies the capex, the market's patience could wear thin.

Third, track Federal Reserve policy and dollar liquidity conditions. The AI trade, like all risk-on trades, is sensitive to monetary tightening. If the Fed delays rate cuts or resumes quantitative tightening, the wall of capital supporting tech valuations could recede. Conversely, if rate cuts arrive in September as futures markets currently price, the AI trade likely has room to run.

Bitcoin sits in a unique position relative to all three variables. It benefits from dollar debasement but does not depend on corporate earnings or AI adoption timelines. It is uncorrelated to Nvidia's gross margins. And its supply schedule is indifferent to whatever the Federal Reserve decides. For investors worried that the AI trade is a late-cycle phenomenon funded by easy money, bitcoin offers something none of the Magnificent Seven can: monetary sovereignty with no counterparty risk.

---

Source: Pomp Letter