The Pomp Letter thesis is well-known: AI shows bubble characteristics historically, but top allocators remain committed.

History Rhymes on AI, But the Smart Money Holds Steady

Anthony Pompliano posed a sharp question to his readers this week: if AI looks like a bubble by every historical measure, why are the most disciplined investors in the world refusing to sell? The Pomp Letter, published May 2026, laid out the case that artificial intelligence capital spending mirrors the dot-com era in scale and speed, yet argued the underlying revenue trajectory sets this cycle apart. For anyone watching capital markets, the distinction matters. For anyone watching Bitcoin, it matters more.

The Numbers Behind the AI Spending Surge

The raw figures are staggering. Microsoft, Alphabet, Meta, and Amazon collectively committed over $200 billion in capital expenditure for AI infrastructure in fiscal 2025, a figure that has only accelerated through early 2026. Nvidia posted $130 billion in trailing twelve-month revenue as of Q1 2026, up from $61 billion just two years prior. The company's data center GPU shipments now account for roughly 85% of total revenue, a concentration that would alarm any fundamental analyst in a normal cycle.

Microsoft alone earmarked $80 billion for AI data center buildouts in fiscal 2025. Meta raised its 2025 capex guidance to $60-65 billion, up from an initial $37-40 billion range. These are not speculative startups burning venture capital. These are the largest companies on the planet, funded by operating cash flow, deploying at wartime speed.

Pompliano's observation is straightforward: this spending pattern looks almost identical to the telecom infrastructure boom of 1998-2000, when companies like WorldCom and Global Crossing poured hundreds of billions into fiber optic cable. That ended badly. The question is whether the analogy holds.

The Dot-Com Comparison and Its Limits

The parallels are real but imperfect. During the dot-com era, the Nasdaq Composite rose approximately 400% from 1995 to its March 2000 peak, driven by companies with negligible revenue and speculative business models. Pets.com went public with $619,000 in revenue and a $290 million valuation. Webvan burned through $1.2 billion before liquidating. The median price-to-sales ratio for Nasdaq stocks exceeded 30x at the peak.

Today's AI leaders look nothing like that. Nvidia trades at roughly 35x trailing earnings as of May 2026, expensive but grounded in actual profit. Microsoft's Intelligent Cloud segment generates over $100 billion in annualized revenue. Alphabet's AI-driven ad optimization has pushed Google Cloud past $50 billion in annual run rate. These companies are profitable, cash-generative, and growing.

The bears argue that this misses the point. David Einhorn of Greenlight Capital warned in late 2025 that market concentration itself is a bubble signal. The top seven stocks by market capitalization now represent over 35% of the S&P 500, a level of concentration not seen since the early 1970s "Nifty Fifty" era. Jim Chanos, the legendary short seller, pointed out that every bubble has real revenue at its core, the problem is overpayment for that revenue and overcapacity in the supply chain.

The Pomp Letter acknowledged this tension but came down on the side of patience. His argument: the investors with the best long-term track records, Berkshire Hathaway's continued Apple position, Renaissance Technologies' quiet accumulation of AI infrastructure plays, and sovereign wealth funds scaling their tech allocations, are not reducing exposure. If anything, they are adding.

Why the Smart Money Stays

The behavioral case for staying invested is more interesting than the financial one. Pompliano cited the historical pattern where the biggest investment returns come from enduring uncomfortable periods of apparent overvaluation. Amazon fell 94% from its dot-com peak but generated 200x returns for investors who held from 1997 through 2025. Cisco never recovered its 2000 high, but investors who bought at reasonable valuations in 2002-2003 still earned 8x.

The key variable is whether AI spending translates into sustained productivity gains. McKinsey's 2025 estimate suggested AI could add $4.4 trillion in annual value to the global economy by 2030. Goldman Sachs put the figure at $7 trillion in cumulative GDP impact. Even if these projections prove optimistic by half, the addressable market dwarfs current spending.

Bridgewater Associates, the world's largest hedge fund, published a research note in early 2026 arguing that AI capex is better understood as a new form of monetary velocity, not a bubble, but a structural shift in how corporations deploy capital. The firm drew parallels to electricity infrastructure spending in the 1920s, which appeared excessive at the time but proved foundational.

Stanley Druckenmiller, who famously rode the dot-com bubble up and exited near the top, told Bloomberg in January 2026 that he maintains AI-related positions. His reasoning: the difference between the internet in 1999 and AI in 2026 is that major corporations are generating immediate returns on their AI investments, not waiting for future adoption.

The Contrarian Case for Caution

Not everyone is convinced. Nouriel Roubini warned at the World Economic Forum in Davos in January 2026 that AI capex cycles follow a predictable arc: over-investment, overcapacity, margin compression, then consolidation. He pointed to the Gartner Hype Cycle as a framework, arguing that generative AI sits squarely in the "trough of disillusionment" for enterprise adoption despite continued spending at the infrastructure layer.

Elliott Management, the activist hedge fund run by Paul Singer, has taken a public position that Nvidia is overvalued and that GPU demand will normalize as hyperscalers build custom silicon. Amazon's Trainium chips, Google's TPU v6, and Microsoft's Maia accelerator all represent in-house alternatives that could erode Nvidia's pricing power within 18-24 months.

There is also the question of monetary policy. The Federal Reserve held rates at 4.25-4.50% through the first half of 2026, providing a more restrictive backdrop than the dot-com era's accommodative environment. Higher rates raise the discount rate on future earnings, which mathematically compresses the valuations of growth stocks. If AI revenue growth decelerates even modestly, the re-rating could be severe.

The most honest framing is probably this: AI is both a real technology shift and a speculative mania. These two things are not mutually exclusive. Railroads in the 1840s, electricity in the 1920s, and the internet in the 1990s were all genuine transformative technologies that also produced devastating financial bubbles. The technology survived and thrived. Many of the investors did not.

The Bitcoin Angle: Sound Money in an Age of Capital Misallocation

This is where the story gets interesting for anyone who thinks in terms of monetary soundness. The AI spending boom is funded almost entirely by fiat currency, much of it created through corporate debt issuance at interest rates subsidized by a decade of central bank manipulation. When Microsoft borrows $10 billion at 4.5% to build data centers, that debt is denominated in dollars whose purchasing power the Federal Reserve actively works to erode at 2% per year.

Austrian economists would recognize the pattern immediately. Friedrich Hayek described the boom-bust cycle as a consequence of artificially cheap credit directing capital into malinvestments, projects that appear profitable only because interest rates do not reflect genuine time preferences. Whether AI itself is a malinvestment is debatable. That the scale of AI investment is amplified by monetary distortion is not.

Bitcoin offers a different framework. A fixed supply of 21 million coins, a transparent issuance schedule halving roughly every four years, and no central authority capable of diluting holders to fund speculative buildouts. The most recent halving in April 2024 reduced the block subsidy to 3.125 BTC, and Bitcoin trades near $107,000 as of mid-May 2026. Its total market capitalization exceeds $2.1 trillion.

If the AI boom ends in a bust, the capital destroyed will be measured in trillions of dollars, much of it borne by passive index fund holders who had no say in the allocation. Bitcoin holders face no such dilution risk. The asset cannot be printed to fund the next generation of data centers. This asymmetry is structural, not speculative.

Pompliano himself has been publicly positioned in both AI and Bitcoin for years, a stance that reflects a practical truth: you can believe a technology is transformative while distrusting the monetary system that finances it. The smart money may be right to hold AI stocks. The smartest money holds Bitcoin alongside them.

What to Watch

Three signals will determine whether AI spending is a dot-com replay or a genuine infrastructure build.

First, enterprise AI revenue growth rates through the second half of 2026. Microsoft's Copilot adoption metrics, Salesforce's Agentforce revenue contribution, and Google Cloud's AI-specific bookings will reveal whether corporate customers are paying for AI at scale or still experimenting. If revenue growth stays above 30% year-over-year, the capex is justified. If it decelerates to 15-20%, expect a repricing.

Second, Nvidia's competitive moat. Custom silicon from hyperscalers will begin shipping in volume by late 2026. Watch Nvidia's gross margins closely. A decline from the current 75% range to below 65% would signal the beginning of commoditization, a classic late-cycle indicator in technology hardware.

Third, Federal Reserve rate decisions. If the Fed begins cutting rates in H2 2026, as futures markets currently price at 60% probability for a September cut, the liquidity tailwind could extend the AI trade regardless of fundamentals. If rates hold or rise, the gravity of discounted cash flow math will reassert itself. In either scenario, Bitcoin benefits: rate cuts weaken the dollar, rate holds expose fiat-funded malinvestment. The 21 million cap does not change either way.

---

Source: Pomp Letter