SpaceX Rewrites IPO Records While Memory Chips Run Dry

The Largest IPO in Human History Is an AI Bet

SpaceX priced its public offering at $135 a share, raising as much as $86 billion at a valuation approaching $1.8 trillion. That tops the Saudi Aramco debut of 2019 in both proceeds and valuation, which is a sentence that should require sitting down. A rocket company built to colonize Mars is now the largest Wall Street debut ever, and the prospectus reads less like aerospace and more like an AI infrastructure pitch deck with thrusters attached.

The money is not going to Starship alone. A meaningful share is earmarked for Terafab, Musk's $55 billion AI chip plant, which just secured a tax break despite local legal threats and organized community opposition. The pattern is familiar. Promise jobs, extract concessions, route the savings into compute. Whatever you think of the man, the capital stack he is assembling, public equity, tax abatements, and government launch contracts, is the most efficient AI funding vehicle ever constructed.

The IPO matters beyond Musk. It establishes that public markets will absorb essentially any quantity of AI-adjacent paper at essentially any valuation. The previous ceiling was a Saudi oil monopoly. The new ceiling is a privately held space and AI conglomerate with a CEO who tweets through SEC filings. Price discovery is now a vibe.

Broadcom Reminds Everyone That Gravity Exists

While SpaceX was busy minting a new record, Broadcom lost roughly $250 billion in market capitalization after-hours on a revenue miss and a soft forecast. Fifteen percent gone on a single guidance number. The stock had been priced for an AI fairy tale, and the company delivered something closer to a quarterly report.

This is the second mode of the current market. Names go up because of AI, and they go down because of AI. The same narrative supplies both the levitation and the cliff. A 15 percent single-session drawdown on a chip giant should be a warning shot about how thin the support actually is under these valuations, but Monday's SpaceX listing suggests nobody is listening. The bid is infinite until it isn't.

The interesting question is whether Broadcom's stumble was a stock-specific disappointment or the first crack in the assumption that every AI vendor enjoys uncapped demand. Probably both. Custom silicon margins were always going to compress once hyperscalers had options. The market is finally noticing that AI infrastructure has competitors, not just customers.

The Real Story Is the Memory Shortage

Underneath the IPO theatrics and the Broadcom whiplash, the most consequential item this slot is the one nobody is talking about. Memory chip supply is now visibly insufficient for non-AI buyers. Automakers and medical device manufacturers are warning that they cannot get DRAM and NAND at acceptable prices or volumes, and business groups are openly lobbying the Trump administration to intervene.

This is what the AI buildout looks like at the physical layer. Hyperscalers are vacuuming up HBM and standard memory at prices the rest of the economy cannot match. The result is rationing by wallet. A pacemaker manufacturer and a hyperscaler walk into a fab, and the hyperscaler walks out with the wafers. The pacemaker company writes a letter to Commerce.

If this persists, expect two things. First, political pressure to allocate semiconductor capacity by sector, which will be ugly and inefficient and probably necessary. Second, a wave of memory capex announcements that will not produce actual chips for two to three years. In the meantime, cars get more expensive, medical devices get backlogged, and AI labs keep buying everything that ships.

What Today Actually Tells You

The through line is concentration. Capital is concentrating in a handful of AI-aligned issuers, with SpaceX now sitting on the largest war chest ever assembled at a public debut. Silicon is concentrating in a handful of end uses, with memory shortages spilling into industries that have nothing to do with language models. And political leverage is concentrating around whoever controls fabs, whoever controls launch, and whoever controls the largest model runs.

The doom number stays moderate because none of this is catastrophic on a single day. The trajectory, however, is unmistakable. A small number of companies are absorbing a large fraction of available capital and silicon, and the rest of the industrial economy is starting to file complaints.