Three days from Wednesday's pricing, the order book on Cerebras Systems has covered the deal more than twenty times at the raised $125 to $135 range. [1] That number is the lede everywhere, and it is the wrong lede. The lede is the mechanism the bookrunners used to get there.

Morgan Stanley, Citi, Barclays and UBS — the four banks running the Cerebras IPO — required indications of interest to be submitted as limit-only orders, with a price the buyer is willing to pay attached to every share allocation request. [2] An IPO order book ordinarily takes a mix of market and limit orders; the market orders are the sleeve banks use to gauge breadth. Cerebras's bookrunners cut the sleeve out. If you wanted to own Cerebras shares at the open Wednesday morning, the bank wanted to know your number first. The paper's Friday note had the book at $10 billion and 2.5x oversubscribed at the original $115-$125 range; the band has since moved up and the cover has compounded.

What Bloomberg's Crystal Tse and her colleagues understood, and what the broader semis press did not, is that this is a price-discovery weapon. A market-order book tells you that demand exists. A limit-order book tells you what demand exists at $135, what demand exists at $130, what demand exists at $145. The bookrunner can draw the supply-and-demand curve in real time. If the curve still steepens at the top of the range, the range goes up. If it flattens, the range stays.

The reason it matters now is the OpenAI counterparty triangle. Cerebras's S-1, filed in April, disclosed that OpenAI is its largest customer and a contracted purchaser of WSE-3 capacity through 2027. Greg Brockman's personal stake in Cerebras — disclosed Friday in the Musk-v-Altman trial as part of the five-stake roll-call (OpenAI, Stripe, Cerebras, CoreWeave, Helion) — sits on top of that customer relationship. [3] The buyside knows it. A standard market-order book would have buried what every fund's risk committee is actually asking, which is: what is the OpenAI-exposure premium, and what is the no-OpenAI floor.

The limit-order book makes them say it. That is a posture you take when you are confident the demand is real and you want the issuer, not the buyside, to capture the price discovery. It is the opposite of what bankers normally do. Normally, the bank wants the IPO pop because the pop is the bank's signal of underwriting credibility. A pop at the open transfers value from the issuer to the allocated buyer. The Cerebras structure transfers it back. Sunnyvale, not the IPO buyer, gets the upside.

The mechanic is also a stress test on the buyside. A long-only fund willing to put a $130 limit order in is making a public statement to the bookrunner that $130 is its private clearing price for Cerebras. The bookrunner, which also runs the long-only fund's prime brokerage book, has an internal record of that number forever. Funds that want to play games — bid the IPO and flip into the secondary at +20% — get caught by a limit that is below the price they would actually pay. The mechanism filters them out.

Bloomberg's May 8 follow-up confirmed the result: the book topped 20x at the raised range, up from a 10x cover at the original $115 to $125. [4] That is a true 20x — 20 times the share count being sold, demanded at the price the underwriters are now asking. The covered ratio at the bottom of the range is far higher; people with knowledge of the order book describe a ratio above 35x at $115. What that tells you is the price-elasticity is steep but not vertical. There is real demand at $135 and there is enormous demand below.

What this means for Wednesday is that the price range is likely to raise again before it prints. Two people involved in the pricing committee describe the working assumption as a final range of $135 to $145, with the actual print toward the top. That would put Cerebras's pricing-day market capitalization above $19 billion fully diluted — a number that, three weeks ago, would have been read as the ceiling, and that the limit-order book has now reset as the floor.

The other half of the story is what the limit-order mechanism implies for the next AI IPO. SpaceX's public S-1 window opens between May 18 and May 22, five trading days after Cerebras prices. People close to Goldman Sachs, which is leading the SpaceX process, describe internal discussions about whether to adopt the same limit-only structure. The argument for it is that SpaceX has the same characteristic Cerebras has — a top-heavy demand curve dominated by long-only funds that would flip the IPO if allowed to. The argument against it is that SpaceX is large enough that the breadth signal still matters; the Musk holdings premium is large enough that the bank may not want to tell the issuer what the buyside thinks the premium is worth.

Cerebras's ratio of book size to deal size is now the highest reported AI-IPO cover this cycle. CoreWeave's was high in March 2025 but at a smaller deal size. The closest comparison is Snowflake's 2020 IPO, which closed with a book Goldman described, then, as "exceptionally tight." Snowflake printed at $120, opened at $245, and closed Day 1 at $253. The pop was a $4 billion transfer from Snowflake to its Day 1 allocations. The Cerebras structure is, mechanically, the bank's confession that it does not want to repeat that trade.

The pricing committee meets Tuesday afternoon. The deal prices Wednesday after the close. The first share trades open Thursday morning under ticker CBRS. By Friday night the question of whether the limit-order book delivered is one print on a screen. By the following Monday, it is either a precedent SpaceX inherits or it is not.

-- THEO KAPLAN, San Francisco