Cerebras enters T-4 of its IPO with the simplest possible pitch and the hardest possible dependency. CNBC reported the AI chipmaker is targeting a $3.5 billion raise, and market coverage has treated the deal as a public test of AI-infrastructure demand. [1]

Friday's paper read the Day 5 IOI book and limit-only order requirement as bookrunner discipline rather than mere enthusiasm. Wednesday's paper put the $26.6 billion S-1/A and OpenAI counterparty triangle on the page. Saturday's question is whether the market is pricing the triangle or admiring the chip.

The public story is clean. AI inference demand is enormous. Nvidia's dominance leaves room for a specialized challenger. Cerebras has a dramatic product, a recognizable customer set, and the kind of revenue trajectory investors want to hear in a market that still pays for compute scarcity. [1] That is the version suited to a roadshow slide.

The prospectus story is less theatrical. An AI infrastructure company whose valuation depends heavily on a major customer is not only selling chips. It is selling confidence in the customer's future spend, payment discipline, strategic consistency, and absence of conflicts that could sour the relationship. In Cerebras's case, the paper has called that structure the OpenAI counterparty triangle because the buyer, the financing story, and the market multiple reinforce one another.

The limit-only order requirement made the triangle more visible. Bloomberg reported earlier in the week that bookrunners were seeking limit orders to gauge true demand at different valuations. [2] If demand were simply unlimited, the constraint would be unnecessary. Limit orders matter because price matters. Price matters because the market is not merely buying growth. It is assigning a discount or premium to concentration.

CNBC's account gives the headline number: $3.5 billion in proceeds. [1] The question behind it is what multiple public investors will tolerate for a company whose growth is bound to the spending path of the AI labs that created the compute panic. That makes Cerebras less like a generic semiconductor IPO and more like a structured bet on the next phase of OpenAI-led infrastructure.

X's useful instinct is suspicion of customer concentration dressed as inevitability. Its less useful habit is treating every connected party as proof of corruption before documents prove the case. Mainstream coverage has the opposite problem. It respects the document stack but often refuses to make the concentration the story unless a filing uses frightening enough prose.

The public market will not resolve the philosophical argument about AI power. It will do something more concrete. It will reveal the price at which investors are willing to convert one buyer's demand into many shareholders' risk.

That is why T-4 matters. The IPO is not only a vote on Cerebras. It is a vote on whether the market believes the AI boom has become infrastructure, or whether it is still a handful of counterparties borrowing certainty from one another.

-- THEO KAPLAN, San Francisco