Analysis
Cerebras priced at $185 in May, opened at $350, peaked at $386.34 on its first day, bottomed at $160.81 in late June, and now trades around $185 to $190 at roughly a $45 billion market cap. Four months, one complete circle.
That path is the most useful pricing document available to any AI compute company weighing a listing, and it is more instructive than the private marks those companies are currently carrying.
What the round trip actually tested
Cerebras sells AI accelerators and inference capacity -- a business whose revenue is concentrated in a small number of very large customers and whose cost base is capital equipment on a depreciation schedule. On day one the market priced the narrative. Over the following ninety days it repriced against reported earnings, and the answer it arrived at was: roughly the IPO price.
The private compute marks queued behind it were set on different logic:
- Crusoe -- $30B: no public filing, bankers retained.
- Fluidstack -- [$18B](/pulse/fluidstack-1-5-billion-jane-street-18-billion-valuation-2026), up from $7.5B in December 2025, on a round Jane Street led.
- Nscale -- $30B cap against a $103B contract backlog.
- SB Energy -- $439B backlog against $269M of revenue, per its Nasdaq registration.
Every one of those numbers is a forward claim on contracts, and the Cerebras experience is what happens when a public market gets to test a forward claim quarterly.
The distinction that will decide pricing
It is not compute versus applications. It is contracted versus committed. A backlog figure counts capacity a customer has agreed to buy; what a public market pays for is revenue a customer cannot walk away from without a penalty. Those are different documents, and the S-1 risk factors are where the difference shows up.
SpaceX offers the contrast, as Pulse laid out in its aftermarket scorecard: priced at $135 in June, closed day one near $161 -- a far smaller pop -- and has climbed since, with Oppenheimer raising its target to $280 on Sept. 2. A modest opening and a rising line is the shape issuers should want. The spike-and-round-trip is the shape that leaves employees underwater at the first lockup expiry, and it is the one the AI compute queue is currently priced for.
