Dinner Plate Chips and Dead Coins
Tuesday, 19 May 2026 · Anonymous Financial Letter
The AI IPO machine just swallowed its first test subject whole. Bitcoin is bleeding quietly in the corner. Nobody is asking the obvious question.
Let me describe the trade for you, and you tell me if it sounds rational.
A company with $510 million in annual revenue — 62% of which came from a single university in Abu Dhabi last year — prices its IPO at $185 a share, above the expected range. The next morning it opens at $350. By afternoon it's touched $385. It closes at $311, giving a chipmaker with no profitable operating history prior to twelve months ago a market cap of roughly $66 billion, or $86 billion fully diluted if you're being honest about the warrants. Eight months before this, the same company was valued at $8.1 billion.
That's a tenfold increase. In eight months. During which it did not invent a new business model. It sold chips.
Welcome to the Cerebras (CBRS) IPO, the largest tech listing of 2026 and, if you squint at it correctly, a Rorschach test for everything currently wrong and possibly right about how capital is being allocated in this market.
The bull case is not stupid. Give it its due. The Wafer-Scale Engine 3 is a genuinely unusual piece of engineering — a single processor 58 times the size of a conventional GPU, purpose-built for AI inference, reportedly 15 times faster on token throughput with lower energy consumption. OpenAI signed a $20 billion compute deal. Amazon is deploying the chips on Bedrock. ARM and SoftBank tried to acquire the company weeks before the IPO, which either means they saw something transformative or they panicked and overpaid, and we'll find out which one when the lockup expires.
The bear case is also not stupid. Two-thirds of 2025 revenue came from UAE customers. The remaining customer concentration — this time in OpenAI, which holds warrants tied to Cerebras hitting a $40 billion valuation — is a structure that would get a first-year analyst sent back to their desk for revision. A $20 billion supply agreement with your largest customer, who also owns equity contingent on your success, is not a customer relationship. It's a merger that hasn't closed yet.
None of this stopped the book from being oversubscribed twenty times.
Twenty times. Think about what that means in practice. For every share available, nineteen buyers walked away empty. That isn't demand. That's a frenzy with paperwork.
The IPO Pipeline Is a Pressure Cooker
Cerebras was the opener. The headliner acts are still warming up backstage.
SpaceX is circling a potential $75 billion fundraise this summer. OpenAI's listing is a question of when, not if. Anthropic is in the wings. These are not ordinary companies entering a normal IPO market. They are brand-name anchors for the single biggest investment narrative since the internet, and they will land into a market already frothy with AI capex enthusiasm, above-consensus earnings, and a Nasdaq 100 that — let's say it plainly — has now surpassed every historical melt-up in the record books except the dot-com bubble's own terminal phase.
That comparison isn't alarmist. It's just arithmetic. The Nasdaq ran 400% from 1997 to March 2000 before unwinding 77% over the following two years. The current run isn't there yet. But the shape of the curve is recognizable to anyone who has looked at it long enough without needing to believe in it.
The S&P 500 is trading at a forward P/E of 20.9. The five-year average is 19.9. The ten-year average is 18.9. Nobody in the financial press is writing that sentence as a warning, which is itself a data point worth noting.
Meanwhile, Bitcoin
Down roughly 9–12% this year. Sitting below $80,000. Google searches for "buy bitcoin" at a twelve-month low, according to Kalshi. The CoinDesk 20 Index is off 19%. Retail interest — the lifeblood of every crypto bull cycle — has migrated. Not to gold. Not to bonds. To CBRS.
This is the rotation everyone saw coming and nobody fully priced. AI equity has absorbed the speculative energy that crypto relied on. Every dollar waiting for the OpenAI IPO is a dollar that didn't flow into Solana. Every institutional allocation reshuffling toward semiconductor infrastructure is a dollar that didn't push ETH to a new high. The Senate Banking Committee's Clarity Act advancing 15-9 out of committee gave Bitcoin a brief $82,000 bounce before it faded back. Regulatory clarity, it turns out, is not a substitute for narrative.
Crypto had narrative in 2020 and 2021. The narrative was: institutions are coming, nation-states are coming, the old financial system is being replaced. Some of that happened. Bitcoin is legal tender in El Salvador. Spot ETFs exist. Fidelity holds it. And yet the price is lower than it was at this point in 2024, and retail has moved on to something that feels more tangible — a chip, a data center, a company with a CEO who can tell you the processor is the size of a dinner plate.
Tangibility is underrated as a market force. People buy what they can picture.
The Question Nobody Is Asking
Here is what gets glossed over in the coverage: the AI capex cycle that is making Nvidia worth $2.7 trillion, that priced Cerebras at $86 billion fully diluted, that has hyperscalers committing $725 billion in infrastructure spend this year — all of it runs on the assumption that the returns exist on the other side.
Alphabet rose 34% in April after a genuine beat across cloud, advertising, and Waymo. Meta fell 9% after raising capex guidance to $125–145 billion for 2026, even while beating on earnings. The market rewarded one and punished the other. That's new. For two years, announcing more AI spending was an unconditional positive. Now investors are starting to price spending against evidence of returns. The divergence between Alphabet and Meta in Q1 wasn't about revenue — both beat. It was about whether the capital allocation story was believable.
That's the hairline crack in the cathedral. Not a collapse. A crack. But cathedrals don't announce when they're about to fall.
Headline CPI has accelerated to 3.8% year-over-year, with ground beef breaking $7 per pound and tomato prices surging. The Empire State Manufacturing Index jumped to 19.6 in May, well above estimates, but the prices subindexes surged alongside it. The most recent jobs report showed nonfarm payrolls above consensus at 115,000, while the U6 underemployment rate jumped 20 basis points to 8.2% and the job-finding rate fell to near its lowest level in over a decade. An economy that is creating jobs more slowly, showing rising underemployment, and accelerating consumer prices is not a soft landing. It's a stagflation sketch that hasn't been named yet because nobody wants to be the first to say it.
Kevin Warsh, four days into the job, inherits this. The FOMC minutes drop tomorrow alongside Nvidia's earnings. By Thursday morning, the market will have either reaffirmed the entire AI thesis or started a conversation about margin compression and return on invested capital. One of those conversations is more fun to have at a conference in Davos. The other one is more useful.
Cerebras closed Friday at $311. Bitcoin is at $76,000. The 30-year Treasury is at 5%. The Nasdaq has never been higher.
Every one of those numbers is internally logical. Taken together, they describe a market that is placing enormous simultaneous bets on growth, on disinflation, on AI productivity, and on the indefinite tolerance of bond markets for a government running 7% deficits in an expansion.
Some of those bets will pay off. Historically, fewer simultaneous maximal bets resolve cleanly than the people making them expect.
But the dinner plate chip is real. Give them that much.
· Not investment advice · For financially literate readers only