Happy (almost) Summer! After watching Kevin Warsh get sworn in at a White House ceremony two days ago, tracking three S-1 filings that could collectively hoover up more capital than every U.S. IPO since 2022 combined, and watching 26-year-old stock charts finally break to new highs — it felt like the right moment to ask the uncomfortable question out loud.
Are we at a party that ends gracefully, or one that ends with the furniture on fire?
The market is simultaneously flashing the neon signs of 1999 and the orchestral excess of 1927. Most commentators reach for the dot-com playbook. I think the original Roaring Twenties is the better map. Here's why...
Assembly Lines to AI Clusters
Ford's River Rouge complex was the largest industrial facility on earth in the 1920s — raw iron in one end, a Model T out the other. Steel, rubber, and oil became the picks-and-shovels of the age. GE and Westinghouse were electrifying factories and homes. The infrastructure buildout was the story, not just the product.
Today, Nvidia's GPU farms are the River Rouge of the AI era. Corning lays the fiber, Micron supplies the memory, the power grid strains under the load. AI data center capex is projected at nearly $1 trillion this year alone. Jensen Huang recently called it "the single largest infrastructure buildout in human history." That is exactly what people said about the electrical grid in 1925. They were right. The infrastructure cycle always pulls in the old industrial players — and then eventually outruns them.
Swampland to Starships
By 1925, Florida lots were changing hands multiple times a day. Buyers bid on land described only in pamphlets, using "binder contracts" — options on future options — to lever up without ever taking delivery. The boom peaked in 1926. The bust came before 1929. And nobody connected the two until it was too late.
SpaceX, OpenAI, and Anthropic are being valued at $900 billion to $2 trillion on businesses that are real, growing — and largely unprofitable at scale. The SpaceX S-1 hit May 20th: $1.75–2 trillion target valuation. OpenAI targets $1 trillion in September. Anthropic follows in October with over $1 trillion Combined: $3.75 trillion pricing in a five-month window — roughly the GDP of France.
The Florida binder contracts are today's IPO allocations. The pamphlets are pitch decks. The lots are orbital launch rights and foundation model APIs. The underlying assets are real. The arithmetic, at these prices, requires a great deal of imagination. Every era has an asset class where imagination outruns math — until it doesn't.
Ticker Tape to Twitter Feeds
The 1920s gave ordinary Americans real-time price data for the first time. Bucket shops let small investors speculate on margin without owning the underlying. By 1929, nine million Americans — in a population of 120 million — held stocks. Speculation had been fully democratized. That was both the triumph and the trap.
Today: zero-commission trading, zero day options, options on phones, Reddit-driven price action, and retail IPO allocations have extended the party to a new generation. The mechanics differ; the psychology is identical — easy access, leverage, and the primal fear of missing the decade's defining trade. Democratizing speculation is always celebrated right up until the margin calls arrive.
The Ghost Fleet & the Change of Guard
Woven through all three themes is an eerie subplot: the infrastructure names of the last technology boom are back at record highs. The companies that defined the dot-com era — Intel up nearly 200% in 2026, Cisco crossing century mark first time ever, Corning finally breaking its year-2000 peak, Micron at an $800 billion market cap — are all riding the AI infrastructure wave. Intel now trades at 108x forward earnings, richer than the most celebrated networking names were at their dot-com apex. Same ships, new cargo delivering the goods for the AI infrastructure
The Fed parallel is equally striking. In 1999, Alan Greenspan had spent three years warning of "irrational exuberance" — then watched the market double anyway. He finally raised rates six times between June 1999 and May 2000. The hikes were what ultimately punctured the bubble. Greenspan had the political independence to act. The market paid the price on his schedule.
Kevin Warsh just took the oath in the East Room two days ago, with the President watching. He inherits 3.8% inflation (three-year high, driven by the Iran war oil shock), real wages falling, and a political principal with a well-documented preference for rate cuts regardless of the data. The inversion from 1999 is the critical distinction: Greenspan had independence but waited too long. Warsh may lack the runway to use it at all. If he cuts while inflation sits at 3.8%, bond markets will reprice violently and every long-duration AI growth name will feel it. If he holds, the political pressure will be unlike anything the institution has faced in a generation. There are no clean exits from this inheritance.
1927 or 1999? Both. Simultaneously.
The three-layer reality:
The Real Thing. AI productivity gains are structural and compounding. The infrastructure buildout will run for years similar to internet infrastructure buildout because need for tokens would be exponential (When the dot-com bubble burst in 2000, global internet traffic averaged roughly 84 petabytes per month. Five years later, in 2005, monthly global traffic had grown significantly to about 2,426 petabytes (2.42 exabytes), representing roughly a 29-fold increase)
The Froth. Infrastructure names at 100x+ earnings, trillion-dollar IPOs priced on terminal-value imagination, retail speculation at historic participation rates. That's where 1999 lives.
The Structural Trap. A new Fed chair under political pressure, 3.8% inflation, a war that markets have priced as resolved when it isn't, and three capital-market whales surfacing simultaneously. That's where 1927 lives — and it's the layer most investors aren't pricing.
The 1927 lesson that the 1999 frame misses: the crash doesn't require fraud. It requires excess capital chasing a genuine thesis faster than that thesis can deliver returns. The 1929 bubble wasn't built on delusion. It was built on extrapolation.
What's the Play?
- Mega-IPOs: entry price over first-day pop. SpaceX is the most defensible — real cash flow, near-monopoly in commercial launch. Size for the five-year hold. The first-day pop is for insiders. The five-year hold is for investors.
- The Ghost Fleet: trade, don't marry. Infrastructure names at 100x earnings are real businesses priced for perfection. Know whether you're buying the thesis or the multiple.
- Warsh Watch is the underpriced risk. The Fed chair transition is the single most important variable for the second half of 2026. Watch his first two moves carefully — they will telegraph whether he has the independence to act or the political constraints that make 1999's ending look orderly by comparison.
- Cash is not trash. Three trillion-dollar IPOs competing for institutional capital creates dislocations in everything else. The patient investor who bought Amazon at $7 in late 2000 didn't need a time machine. Just dry powder and conviction.
Final Thought
We came into 2026 asking whether this was 1998 or 2000. The better map is a century older — assembly lines becoming AI clusters, swampland becoming starships, bucket shops becoming 24 hour trading on mobile apps. And a new Fed chair, like Greenspan before him, inheriting a market that has already decided what it wants to believe.
History doesn't repeat. But sometimes it rhymes so loudly you can hear it from the balcony.
The question for 2026–2027 isn't whether this ends. The question is whether you're positioned with enough real exposure to capture a genuine technological transformation — and enough intellectual honesty to know that the price you pay determines everything.
Enjoy the summer. It might be a memorable one.
/Shyam
Personal views and market observations, not investment advice. Do your own diligence.
Comments