Cerebras's OpenAI deal is the kind of validation an IPO banker dreams of, and the kind of concentration risk an IPO lawyer is paid to underline. Monday's paper said OpenAI concentration was the Cerebras risk factor to read first. Tuesday's filing read is sharper: validation and dependency are the same fact.

The S-1 says OpenAI selected Cerebras for a multi-year relationship valued at more than $20 billion, with 750 megawatts of high-speed AI compute and co-design of future models for future Cerebras hardware. [1] That paragraph is the sales pitch. A few pages later, the risk language tells investors that a substantial portion of revenue is expected to be driven by a limited number of customers, including OpenAI, G42, MBZUAI and AWS. [1]

Reuters' account of the IPO revival keeps the public frame familiar: a Nvidia rival comes back to market as AI listings warm, with OpenAI demand and prior national-security scrutiny around G42 as central context. [2] That is accurate. It is not the whole story.

The whole story is invoice architecture. OpenAI is not an ordinary logo. It is the market's most legible proof that a compute provider has something worth buying. But if the proof becomes too large a share of the revenue case, investors are buying customer power as well as chip speed.

The divergence is useful. X can hype the Nvidia challenger or warn that OpenAI owns the roadshow. Mainstream coverage can keep the IPO in the listing-recovery lane. The filing says to hold both thoughts. The customer that validates the company can also define it.

That does not make Cerebras weak. It makes the roadshow honest. The first question is not whether OpenAI matters. It is what Cerebras looks like if everyone else matters less.

The prospectus gives investors the right order of operations: admire the contract, then price the dependence.

-- THEO KAPLAN, San Francisco